Age Owner Branch data TLA Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * procarray.c
4 : : * POSTGRES process array code.
5 : : *
6 : : *
7 : : * This module maintains arrays of PGPROC substructures, as well as associated
8 : : * arrays in ProcGlobal, for all active backends. Although there are several
9 : : * uses for this, the principal one is as a means of determining the set of
10 : : * currently running transactions.
11 : : *
12 : : * Because of various subtle race conditions it is critical that a backend
13 : : * hold the correct locks while setting or clearing its xid (in
14 : : * ProcGlobal->xids[]/MyProc->xid). See notes in
15 : : * src/backend/access/transam/README.
16 : : *
17 : : * The process arrays now also include structures representing prepared
18 : : * transactions. The xid and subxids fields of these are valid, as are the
19 : : * myProcLocks lists. They can be distinguished from regular backend PGPROCs
20 : : * at need by checking for pid == 0.
21 : : *
22 : : * During hot standby, we also keep a list of XIDs representing transactions
23 : : * that are known to be running on the primary (or more precisely, were running
24 : : * as of the current point in the WAL stream). This list is kept in the
25 : : * KnownAssignedXids array, and is updated by watching the sequence of
26 : : * arriving XIDs. This is necessary because if we leave those XIDs out of
27 : : * snapshots taken for standby queries, then they will appear to be already
28 : : * complete, leading to MVCC failures. Note that in hot standby, the PGPROC
29 : : * array represents standby processes, which by definition are not running
30 : : * transactions that have XIDs.
31 : : *
32 : : * It is perhaps possible for a backend on the primary to terminate without
33 : : * writing an abort record for its transaction. While that shouldn't really
34 : : * happen, it would tie up KnownAssignedXids indefinitely, so we protect
35 : : * ourselves by pruning the array when a valid list of running XIDs arrives.
36 : : *
37 : : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
38 : : * Portions Copyright (c) 1994, Regents of the University of California
39 : : *
40 : : *
41 : : * IDENTIFICATION
42 : : * src/backend/storage/ipc/procarray.c
43 : : *
44 : : *-------------------------------------------------------------------------
45 : : */
46 : : #include "postgres.h"
47 : :
48 : : #include <signal.h>
49 : :
50 : : #include "access/subtrans.h"
51 : : #include "access/transam.h"
52 : : #include "access/twophase.h"
53 : : #include "access/xact.h"
54 : : #include "access/xlogutils.h"
55 : : #include "catalog/catalog.h"
56 : : #include "catalog/pg_authid.h"
57 : : #include "miscadmin.h"
58 : : #include "pgstat.h"
59 : : #include "postmaster/bgworker.h"
60 : : #include "port/pg_lfind.h"
61 : : #include "storage/proc.h"
62 : : #include "storage/procarray.h"
63 : : #include "storage/procsignal.h"
64 : : #include "storage/subsystems.h"
65 : : #include "utils/acl.h"
66 : : #include "utils/builtins.h"
67 : : #include "utils/injection_point.h"
68 : : #include "utils/lsyscache.h"
69 : : #include "utils/rel.h"
70 : : #include "utils/snapmgr.h"
71 : : #include "utils/wait_event.h"
72 : :
73 : : #define UINT32_ACCESS_ONCE(var) ((uint32)(*((volatile uint32 *)&(var))))
74 : :
75 : : /* Our shared memory area */
76 : : typedef struct ProcArrayStruct
77 : : {
78 : : int numProcs; /* number of valid procs entries */
79 : : int maxProcs; /* allocated size of procs array */
80 : :
81 : : /*
82 : : * Known assigned XIDs handling
83 : : */
84 : : int maxKnownAssignedXids; /* allocated size of array */
85 : : int numKnownAssignedXids; /* current # of valid entries */
86 : : int tailKnownAssignedXids; /* index of oldest valid element */
87 : : int headKnownAssignedXids; /* index of newest element, + 1 */
88 : :
89 : : /*
90 : : * Highest subxid that has been removed from KnownAssignedXids array to
91 : : * prevent overflow; or InvalidTransactionId if none. We track this for
92 : : * similar reasons to tracking overflowing cached subxids in PGPROC
93 : : * entries. Must hold exclusive ProcArrayLock to change this, and shared
94 : : * lock to read it.
95 : : */
96 : : TransactionId lastOverflowedXid;
97 : :
98 : : /* oldest xmin of any replication slot */
99 : : TransactionId replication_slot_xmin;
100 : : /* oldest catalog xmin of any replication slot */
101 : : TransactionId replication_slot_catalog_xmin;
102 : :
103 : : /* indexes into allProcs[], has PROCARRAY_MAXPROCS entries */
104 : : int pgprocnos[FLEXIBLE_ARRAY_MEMBER];
105 : : } ProcArrayStruct;
106 : :
107 : : static void ProcArrayShmemRequest(void *arg);
108 : : static void ProcArrayShmemInit(void *arg);
109 : : static void ProcArrayShmemAttach(void *arg);
110 : :
111 : : static ProcArrayStruct *procArray;
112 : :
113 : : const struct ShmemCallbacks ProcArrayShmemCallbacks = {
114 : : .request_fn = ProcArrayShmemRequest,
115 : : .init_fn = ProcArrayShmemInit,
116 : : .attach_fn = ProcArrayShmemAttach,
117 : : };
118 : :
119 : : /*
120 : : * State for the GlobalVisTest* family of functions. Those functions can
121 : : * e.g. be used to decide if a deleted row can be removed without violating
122 : : * MVCC semantics: If the deleted row's xmax is not considered to be running
123 : : * by anyone, the row can be removed.
124 : : *
125 : : * To avoid slowing down GetSnapshotData(), we don't calculate a precise
126 : : * cutoff XID while building a snapshot (looking at the frequently changing
127 : : * xmins scales badly). Instead we compute two boundaries while building the
128 : : * snapshot:
129 : : *
130 : : * 1) definitely_needed, indicating that rows deleted by XIDs >=
131 : : * definitely_needed are definitely still visible.
132 : : *
133 : : * 2) maybe_needed, indicating that rows deleted by XIDs < maybe_needed can
134 : : * definitely be removed
135 : : *
136 : : * When testing an XID that falls in between the two (i.e. XID >= maybe_needed
137 : : * && XID < definitely_needed), the boundaries can be recomputed (using
138 : : * ComputeXidHorizons()) to get a more accurate answer. This is cheaper than
139 : : * maintaining an accurate value all the time.
140 : : *
141 : : * As it is not cheap to compute accurate boundaries, we limit the number of
142 : : * times that happens in short succession. See GlobalVisTestShouldUpdate().
143 : : *
144 : : *
145 : : * There are three backend lifetime instances of this struct, optimized for
146 : : * different types of relations. As e.g. a normal user defined table in one
147 : : * database is inaccessible to backends connected to another database, a test
148 : : * specific to a relation can be more aggressive than a test for a shared
149 : : * relation. Currently we track four different states:
150 : : *
151 : : * 1) GlobalVisSharedRels, which only considers an XID's
152 : : * effects visible-to-everyone if neither snapshots in any database, nor a
153 : : * replication slot's xmin, nor a replication slot's catalog_xmin might
154 : : * still consider XID as running.
155 : : *
156 : : * 2) GlobalVisCatalogRels, which only considers an XID's
157 : : * effects visible-to-everyone if neither snapshots in the current
158 : : * database, nor a replication slot's xmin, nor a replication slot's
159 : : * catalog_xmin might still consider XID as running.
160 : : *
161 : : * I.e. the difference to GlobalVisSharedRels is that
162 : : * snapshot in other databases are ignored.
163 : : *
164 : : * 3) GlobalVisDataRels, which only considers an XID's
165 : : * effects visible-to-everyone if neither snapshots in the current
166 : : * database, nor a replication slot's xmin consider XID as running.
167 : : *
168 : : * I.e. the difference to GlobalVisCatalogRels is that
169 : : * replication slot's catalog_xmin is not taken into account.
170 : : *
171 : : * 4) GlobalVisTempRels, which only considers the current session, as temp
172 : : * tables are not visible to other sessions.
173 : : *
174 : : * GlobalVisTestFor(relation) returns the appropriate state
175 : : * for the relation.
176 : : *
177 : : * The boundaries are FullTransactionIds instead of TransactionIds to avoid
178 : : * wraparound dangers. There e.g. would otherwise exist no procarray state to
179 : : * prevent maybe_needed to become old enough after the GetSnapshotData()
180 : : * call.
181 : : *
182 : : * The typedef is in the header.
183 : : */
184 : : struct GlobalVisState
185 : : {
186 : : /* XIDs >= are considered running by some backend */
187 : : FullTransactionId definitely_needed;
188 : :
189 : : /* XIDs < are not considered to be running by any backend */
190 : : FullTransactionId maybe_needed;
191 : : };
192 : :
193 : : /*
194 : : * Result of ComputeXidHorizons().
195 : : */
196 : : typedef struct ComputeXidHorizonsResult
197 : : {
198 : : /*
199 : : * The value of TransamVariables->latestCompletedXid when
200 : : * ComputeXidHorizons() held ProcArrayLock.
201 : : */
202 : : FullTransactionId latest_completed;
203 : :
204 : : /*
205 : : * The same for procArray->replication_slot_xmin and
206 : : * procArray->replication_slot_catalog_xmin.
207 : : */
208 : : TransactionId slot_xmin;
209 : : TransactionId slot_catalog_xmin;
210 : :
211 : : /*
212 : : * Oldest xid that any backend might still consider running. This needs to
213 : : * include processes running VACUUM, in contrast to the normal visibility
214 : : * cutoffs, as vacuum needs to be able to perform pg_subtrans lookups when
215 : : * determining visibility, but doesn't care about rows above its xmin to
216 : : * be removed.
217 : : *
218 : : * This likely should only be needed to determine whether pg_subtrans can
219 : : * be truncated. It currently includes the effects of replication slots,
220 : : * for historical reasons. But that could likely be changed.
221 : : */
222 : : TransactionId oldest_considered_running;
223 : :
224 : : /*
225 : : * Oldest xid for which deleted tuples need to be retained in shared
226 : : * tables.
227 : : *
228 : : * This includes the effects of replication slots. If that's not desired,
229 : : * look at shared_oldest_nonremovable_raw;
230 : : */
231 : : TransactionId shared_oldest_nonremovable;
232 : :
233 : : /*
234 : : * Oldest xid that may be necessary to retain in shared tables. This is
235 : : * the same as shared_oldest_nonremovable, except that is not affected by
236 : : * replication slot's catalog_xmin.
237 : : *
238 : : * This is mainly useful to be able to send the catalog_xmin to upstream
239 : : * streaming replication servers via hot_standby_feedback, so they can
240 : : * apply the limit only when accessing catalog tables.
241 : : */
242 : : TransactionId shared_oldest_nonremovable_raw;
243 : :
244 : : /*
245 : : * Oldest xid for which deleted tuples need to be retained in non-shared
246 : : * catalog tables.
247 : : */
248 : : TransactionId catalog_oldest_nonremovable;
249 : :
250 : : /*
251 : : * Oldest xid for which deleted tuples need to be retained in normal user
252 : : * defined tables.
253 : : */
254 : : TransactionId data_oldest_nonremovable;
255 : :
256 : : /*
257 : : * Oldest xid for which deleted tuples need to be retained in this
258 : : * session's temporary tables.
259 : : */
260 : : TransactionId temp_oldest_nonremovable;
261 : : } ComputeXidHorizonsResult;
262 : :
263 : : /*
264 : : * Return value for GlobalVisHorizonKindForRel().
265 : : */
266 : : typedef enum GlobalVisHorizonKind
267 : : {
268 : : VISHORIZON_SHARED,
269 : : VISHORIZON_CATALOG,
270 : : VISHORIZON_DATA,
271 : : VISHORIZON_TEMP,
272 : : } GlobalVisHorizonKind;
273 : :
274 : : /*
275 : : * Reason codes for KnownAssignedXidsCompress().
276 : : */
277 : : typedef enum KAXCompressReason
278 : : {
279 : : KAX_NO_SPACE, /* need to free up space at array end */
280 : : KAX_PRUNE, /* we just pruned old entries */
281 : : KAX_TRANSACTION_END, /* we just committed/removed some XIDs */
282 : : KAX_STARTUP_PROCESS_IDLE, /* startup process is about to sleep */
283 : : } KAXCompressReason;
284 : :
285 : : static PGPROC *allProcs;
286 : :
287 : : /*
288 : : * Cache to reduce overhead of repeated calls to TransactionIdIsInProgress()
289 : : */
290 : : static TransactionId cachedXidIsNotInProgress = InvalidTransactionId;
291 : :
292 : : /*
293 : : * Bookkeeping for tracking emulated transactions in recovery
294 : : */
295 : :
296 : : static TransactionId *KnownAssignedXids;
297 : :
298 : : static bool *KnownAssignedXidsValid;
299 : :
300 : : static TransactionId latestObservedXid = InvalidTransactionId;
301 : :
302 : : /*
303 : : * If we're in STANDBY_SNAPSHOT_PENDING state, standbySnapshotPendingXmin is
304 : : * the highest xid that might still be running that we don't have in
305 : : * KnownAssignedXids.
306 : : */
307 : : static TransactionId standbySnapshotPendingXmin;
308 : :
309 : : /*
310 : : * State for visibility checks on different types of relations. See struct
311 : : * GlobalVisState for details. As shared, catalog, normal and temporary
312 : : * relations can have different horizons, one such state exists for each.
313 : : */
314 : : static GlobalVisState GlobalVisSharedRels;
315 : : static GlobalVisState GlobalVisCatalogRels;
316 : : static GlobalVisState GlobalVisDataRels;
317 : : static GlobalVisState GlobalVisTempRels;
318 : :
319 : : /*
320 : : * This backend's RecentXmin at the last time the accurate xmin horizon was
321 : : * recomputed, or InvalidTransactionId if it has not. Used to limit how many
322 : : * times accurate horizons are recomputed. See GlobalVisTestShouldUpdate().
323 : : */
324 : : static TransactionId ComputeXidHorizonsResultLastXmin;
325 : :
326 : : #ifdef XIDCACHE_DEBUG
327 : :
328 : : /* counters for XidCache measurement */
329 : : static long xc_by_recent_xmin = 0;
330 : : static long xc_by_known_xact = 0;
331 : : static long xc_by_my_xact = 0;
332 : : static long xc_by_latest_xid = 0;
333 : : static long xc_by_main_xid = 0;
334 : : static long xc_by_child_xid = 0;
335 : : static long xc_by_known_assigned = 0;
336 : : static long xc_no_overflow = 0;
337 : : static long xc_slow_answer = 0;
338 : :
339 : : #define xc_by_recent_xmin_inc() (xc_by_recent_xmin++)
340 : : #define xc_by_known_xact_inc() (xc_by_known_xact++)
341 : : #define xc_by_my_xact_inc() (xc_by_my_xact++)
342 : : #define xc_by_latest_xid_inc() (xc_by_latest_xid++)
343 : : #define xc_by_main_xid_inc() (xc_by_main_xid++)
344 : : #define xc_by_child_xid_inc() (xc_by_child_xid++)
345 : : #define xc_by_known_assigned_inc() (xc_by_known_assigned++)
346 : : #define xc_no_overflow_inc() (xc_no_overflow++)
347 : : #define xc_slow_answer_inc() (xc_slow_answer++)
348 : :
349 : : static void DisplayXidCache(void);
350 : : #else /* !XIDCACHE_DEBUG */
351 : :
352 : : #define xc_by_recent_xmin_inc() ((void) 0)
353 : : #define xc_by_known_xact_inc() ((void) 0)
354 : : #define xc_by_my_xact_inc() ((void) 0)
355 : : #define xc_by_latest_xid_inc() ((void) 0)
356 : : #define xc_by_main_xid_inc() ((void) 0)
357 : : #define xc_by_child_xid_inc() ((void) 0)
358 : : #define xc_by_known_assigned_inc() ((void) 0)
359 : : #define xc_no_overflow_inc() ((void) 0)
360 : : #define xc_slow_answer_inc() ((void) 0)
361 : : #endif /* XIDCACHE_DEBUG */
362 : :
363 : : /* Primitives for KnownAssignedXids array handling for standby */
364 : : static void KnownAssignedXidsCompress(KAXCompressReason reason, bool haveLock);
365 : : static void KnownAssignedXidsAdd(TransactionId from_xid, TransactionId to_xid,
366 : : bool exclusive_lock);
367 : : static bool KnownAssignedXidsSearch(TransactionId xid, bool remove);
368 : : static bool KnownAssignedXidExists(TransactionId xid);
369 : : static void KnownAssignedXidsRemove(TransactionId xid);
370 : : static void KnownAssignedXidsRemoveTree(TransactionId xid, int nsubxids,
371 : : TransactionId *subxids);
372 : : static void KnownAssignedXidsRemovePreceding(TransactionId removeXid);
373 : : static int KnownAssignedXidsGet(TransactionId *xarray, TransactionId xmax);
374 : : static int KnownAssignedXidsGetAndSetXmin(TransactionId *xarray,
375 : : TransactionId *xmin,
376 : : TransactionId xmax);
377 : : static TransactionId KnownAssignedXidsGetOldestXmin(void);
378 : : static void KnownAssignedXidsDisplay(int trace_level);
379 : : static void KnownAssignedXidsReset(void);
380 : : static inline void ProcArrayEndTransactionInternal(PGPROC *proc, TransactionId latestXid);
381 : : static void ProcArrayGroupClearXid(PGPROC *proc, TransactionId latestXid);
382 : : static void MaintainLatestCompletedXid(TransactionId latestXid);
383 : : static void MaintainLatestCompletedXidRecovery(TransactionId latestXid);
384 : :
385 : : static inline FullTransactionId FullXidRelativeTo(FullTransactionId rel,
386 : : TransactionId xid);
387 : : static void GlobalVisUpdateApply(ComputeXidHorizonsResult *horizons);
388 : :
389 : : /*
390 : : * Register the shared PGPROC array during postmaster startup.
391 : : */
392 : : static void
29 heikki.linnakangas@i 393 :GNC 1244 : ProcArrayShmemRequest(void *arg)
394 : : {
395 : : #define PROCARRAY_MAXPROCS (MaxBackends + max_prepared_xacts)
396 : :
397 : : /*
398 : : * During Hot Standby processing we have a data structure called
399 : : * KnownAssignedXids, created in shared memory. Local data structures are
400 : : * also created in various backends during GetSnapshotData(),
401 : : * TransactionIdIsInProgress() and GetRunningTransactionData(). All of the
402 : : * main structures created in those functions must be identically sized,
403 : : * since we may at times copy the whole of the data structures around. We
404 : : * refer to this size as TOTAL_MAX_CACHED_SUBXIDS.
405 : : *
406 : : * Ideally we'd only create this structure if we were actually doing hot
407 : : * standby in the current run, but we don't know that yet at the time
408 : : * shared memory is being set up.
409 : : */
410 : : #define TOTAL_MAX_CACHED_SUBXIDS \
411 : : ((PGPROC_MAX_CACHED_SUBXIDS + 1) * PROCARRAY_MAXPROCS)
412 : :
5850 tgl@sss.pgh.pa.us 413 [ + + ]:CBC 1244 : if (EnableHotStandby)
414 : : {
29 heikki.linnakangas@i 415 :GNC 1238 : ShmemRequestStruct(.name = "KnownAssignedXids",
416 : : .size = mul_size(sizeof(TransactionId), TOTAL_MAX_CACHED_SUBXIDS),
417 : : .ptr = (void **) &KnownAssignedXids,
418 : : );
419 : :
420 : 1238 : ShmemRequestStruct(.name = "KnownAssignedXidsValid",
421 : : .size = mul_size(sizeof(bool), TOTAL_MAX_CACHED_SUBXIDS),
422 : : .ptr = (void **) &KnownAssignedXidsValid,
423 : : );
424 : : }
425 : :
426 : : /* Register the ProcArray shared structure */
427 : 1244 : ShmemRequestStruct(.name = "Proc Array",
428 : : .size = add_size(offsetof(ProcArrayStruct, pgprocnos),
429 : : mul_size(sizeof(int), PROCARRAY_MAXPROCS)),
430 : : .ptr = (void **) &procArray,
431 : : );
7656 tgl@sss.pgh.pa.us 432 :GIC 1244 : }
433 : :
434 : : /*
435 : : * Initialize the shared PGPROC array during postmaster startup.
436 : : */
437 : : static void
29 heikki.linnakangas@i 438 :GNC 1241 : ProcArrayShmemInit(void *arg)
439 : : {
440 : 1241 : procArray->numProcs = 0;
441 : 1241 : procArray->maxProcs = PROCARRAY_MAXPROCS;
442 : 1241 : procArray->maxKnownAssignedXids = TOTAL_MAX_CACHED_SUBXIDS;
443 : 1241 : procArray->numKnownAssignedXids = 0;
444 : 1241 : procArray->tailKnownAssignedXids = 0;
445 : 1241 : procArray->headKnownAssignedXids = 0;
446 : 1241 : procArray->lastOverflowedXid = InvalidTransactionId;
447 : 1241 : procArray->replication_slot_xmin = InvalidTransactionId;
448 : 1241 : procArray->replication_slot_catalog_xmin = InvalidTransactionId;
449 : 1241 : TransamVariables->xactCompletionCount = 1;
450 : :
5275 rhaas@postgresql.org 451 :CBC 1241 : allProcs = ProcGlobal->allProcs;
29 heikki.linnakangas@i 452 :GNC 1241 : }
453 : :
454 : : static void
29 heikki.linnakangas@i 455 :UNC 0 : ProcArrayShmemAttach(void *arg)
456 : : {
457 : 0 : allProcs = ProcGlobal->allProcs;
7656 tgl@sss.pgh.pa.us 458 :LBC (1070) : }
459 : :
460 : : /*
461 : : * Add the specified PGPROC to the shared array.
462 : : */
463 : : void
7627 tgl@sss.pgh.pa.us 464 :CBC 18915 : ProcArrayAdd(PGPROC *proc)
465 : : {
803 heikki.linnakangas@i 466 : 18915 : int pgprocno = GetNumberFromPGProc(proc);
7656 tgl@sss.pgh.pa.us 467 : 18915 : ProcArrayStruct *arrayP = procArray;
468 : : int index;
469 : : int movecount;
470 : :
471 : : /* See ProcGlobal comment explaining why both locks are held */
472 : 18915 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
2090 andres@anarazel.de 473 : 18915 : LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
474 : :
7656 tgl@sss.pgh.pa.us 475 [ - + ]: 18915 : if (arrayP->numProcs >= arrayP->maxProcs)
476 : : {
477 : : /*
478 : : * Oops, no room. (This really shouldn't happen, since there is a
479 : : * fixed supply of PGPROC structs too, and so we should have failed
480 : : * earlier.)
481 : : */
7656 tgl@sss.pgh.pa.us 482 [ # # ]:UBC 0 : ereport(FATAL,
483 : : (errcode(ERRCODE_TOO_MANY_CONNECTIONS),
484 : : errmsg("sorry, too many clients already")));
485 : : }
486 : :
487 : : /*
488 : : * Keep the procs array sorted by (PGPROC *) so that we can utilize
489 : : * locality of references much better. This is useful while traversing the
490 : : * ProcArray because there is an increased likelihood of finding the next
491 : : * PGPROC structure in the cache.
492 : : *
493 : : * Since the occurrence of adding/removing a proc is much lower than the
494 : : * access to the ProcArray itself, the overhead should be marginal
495 : : */
5275 rhaas@postgresql.org 496 [ + + ]:CBC 47106 : for (index = 0; index < arrayP->numProcs; index++)
497 : : {
803 heikki.linnakangas@i 498 : 41746 : int this_procno = arrayP->pgprocnos[index];
499 : :
500 [ + - - + ]: 41746 : Assert(this_procno >= 0 && this_procno < (arrayP->maxProcs + NUM_AUXILIARY_PROCS));
501 [ - + ]: 41746 : Assert(allProcs[this_procno].pgxactoff == index);
502 : :
503 : : /* If we have found our right position in the array, break */
504 [ + + ]: 41746 : if (this_procno > pgprocno)
5275 rhaas@postgresql.org 505 : 13555 : break;
506 : : }
507 : :
1789 andres@anarazel.de 508 : 18915 : movecount = arrayP->numProcs - index;
509 : 18915 : memmove(&arrayP->pgprocnos[index + 1],
510 : 18915 : &arrayP->pgprocnos[index],
511 : : movecount * sizeof(*arrayP->pgprocnos));
512 : 18915 : memmove(&ProcGlobal->xids[index + 1],
513 : 18915 : &ProcGlobal->xids[index],
514 : : movecount * sizeof(*ProcGlobal->xids));
515 : 18915 : memmove(&ProcGlobal->subxidStates[index + 1],
516 : 18915 : &ProcGlobal->subxidStates[index],
517 : : movecount * sizeof(*ProcGlobal->subxidStates));
518 : 18915 : memmove(&ProcGlobal->statusFlags[index + 1],
519 : 18915 : &ProcGlobal->statusFlags[index],
520 : : movecount * sizeof(*ProcGlobal->statusFlags));
521 : :
803 heikki.linnakangas@i 522 : 18915 : arrayP->pgprocnos[index] = GetNumberFromPGProc(proc);
1789 andres@anarazel.de 523 : 18915 : proc->pgxactoff = index;
2090 524 : 18915 : ProcGlobal->xids[index] = proc->xid;
525 : 18915 : ProcGlobal->subxidStates[index] = proc->subxidStatus;
1996 alvherre@alvh.no-ip. 526 : 18915 : ProcGlobal->statusFlags[index] = proc->statusFlags;
527 : :
7656 tgl@sss.pgh.pa.us 528 : 18915 : arrayP->numProcs++;
529 : :
530 : : /* adjust pgxactoff for all following PGPROCs */
1789 andres@anarazel.de 531 : 18915 : index++;
2090 532 [ + + ]: 51817 : for (; index < arrayP->numProcs; index++)
533 : : {
1789 534 : 32902 : int procno = arrayP->pgprocnos[index];
535 : :
536 [ + - - + ]: 32902 : Assert(procno >= 0 && procno < (arrayP->maxProcs + NUM_AUXILIARY_PROCS));
537 [ - + ]: 32902 : Assert(allProcs[procno].pgxactoff == index - 1);
538 : :
539 : 32902 : allProcs[procno].pgxactoff = index;
540 : : }
541 : :
542 : : /*
543 : : * Release in reversed acquisition order, to reduce frequency of having to
544 : : * wait for XidGenLock while holding ProcArrayLock.
545 : : */
2090 546 : 18915 : LWLockRelease(XidGenLock);
7656 tgl@sss.pgh.pa.us 547 : 18915 : LWLockRelease(ProcArrayLock);
548 : 18915 : }
549 : :
550 : : /*
551 : : * Remove the specified PGPROC from the shared array.
552 : : *
553 : : * When latestXid is a valid XID, we are removing a live 2PC gxact from the
554 : : * array, and thus causing it to appear as "not running" anymore. In this
555 : : * case we must advance latestCompletedXid. (This is essentially the same
556 : : * as ProcArrayEndTransaction followed by removal of the PGPROC, but we take
557 : : * the ProcArrayLock only once, and don't damage the content of the PGPROC;
558 : : * twophase.c depends on the latter.)
559 : : */
560 : : void
6814 561 : 18889 : ProcArrayRemove(PGPROC *proc, TransactionId latestXid)
562 : : {
7656 563 : 18889 : ProcArrayStruct *arrayP = procArray;
564 : : int myoff;
565 : : int movecount;
566 : :
567 : : #ifdef XIDCACHE_DEBUG
568 : : /* dump stats at backend shutdown, but not prepared-xact end */
569 : : if (proc->pid != 0)
570 : : DisplayXidCache();
571 : : #endif
572 : :
573 : : /* See ProcGlobal comment explaining why both locks are held */
574 : 18889 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
2090 andres@anarazel.de 575 : 18889 : LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
576 : :
1789 577 : 18889 : myoff = proc->pgxactoff;
578 : :
579 [ + - - + ]: 18889 : Assert(myoff >= 0 && myoff < arrayP->numProcs);
580 [ - + ]: 18889 : Assert(ProcGlobal->allProcs[arrayP->pgprocnos[myoff]].pgxactoff == myoff);
581 : :
6814 tgl@sss.pgh.pa.us 582 [ + + ]: 18889 : if (TransactionIdIsValid(latestXid))
583 : : {
1789 andres@anarazel.de 584 [ - + ]: 304 : Assert(TransactionIdIsValid(ProcGlobal->xids[myoff]));
585 : :
586 : : /* Advance global latestCompletedXid while holding the lock */
2093 587 : 304 : MaintainLatestCompletedXid(latestXid);
588 : :
589 : : /* Same with xactCompletionCount */
879 heikki.linnakangas@i 590 : 304 : TransamVariables->xactCompletionCount++;
591 : :
1789 andres@anarazel.de 592 : 304 : ProcGlobal->xids[myoff] = InvalidTransactionId;
593 : 304 : ProcGlobal->subxidStates[myoff].overflowed = false;
594 : 304 : ProcGlobal->subxidStates[myoff].count = 0;
595 : : }
596 : : else
597 : : {
598 : : /* Shouldn't be trying to remove a live transaction here */
599 [ - + ]: 18585 : Assert(!TransactionIdIsValid(ProcGlobal->xids[myoff]));
600 : : }
601 : :
602 [ - + ]: 18889 : Assert(!TransactionIdIsValid(ProcGlobal->xids[myoff]));
603 [ - + ]: 18889 : Assert(ProcGlobal->subxidStates[myoff].count == 0);
604 [ - + ]: 18889 : Assert(ProcGlobal->subxidStates[myoff].overflowed == false);
605 : :
606 : 18889 : ProcGlobal->statusFlags[myoff] = 0;
607 : :
608 : : /* Keep the PGPROC array sorted. See notes above */
609 : 18889 : movecount = arrayP->numProcs - myoff - 1;
610 : 18889 : memmove(&arrayP->pgprocnos[myoff],
611 : 18889 : &arrayP->pgprocnos[myoff + 1],
612 : : movecount * sizeof(*arrayP->pgprocnos));
613 : 18889 : memmove(&ProcGlobal->xids[myoff],
614 : 18889 : &ProcGlobal->xids[myoff + 1],
615 : : movecount * sizeof(*ProcGlobal->xids));
616 : 18889 : memmove(&ProcGlobal->subxidStates[myoff],
617 : 18889 : &ProcGlobal->subxidStates[myoff + 1],
618 : : movecount * sizeof(*ProcGlobal->subxidStates));
619 : 18889 : memmove(&ProcGlobal->statusFlags[myoff],
620 : 18889 : &ProcGlobal->statusFlags[myoff + 1],
621 : : movecount * sizeof(*ProcGlobal->statusFlags));
622 : :
623 : 18889 : arrayP->pgprocnos[arrayP->numProcs - 1] = -1; /* for debugging */
624 : 18889 : arrayP->numProcs--;
625 : :
626 : : /*
627 : : * Adjust pgxactoff of following procs for removed PGPROC (note that
628 : : * numProcs already has been decremented).
629 : : */
630 [ + + ]: 55358 : for (int index = myoff; index < arrayP->numProcs; index++)
631 : : {
632 : 36469 : int procno = arrayP->pgprocnos[index];
633 : :
634 [ + - - + ]: 36469 : Assert(procno >= 0 && procno < (arrayP->maxProcs + NUM_AUXILIARY_PROCS));
635 [ - + ]: 36469 : Assert(allProcs[procno].pgxactoff - 1 == index);
636 : :
637 : 36469 : allProcs[procno].pgxactoff = index;
638 : : }
639 : :
640 : : /*
641 : : * Release in reversed acquisition order, to reduce frequency of having to
642 : : * wait for XidGenLock while holding ProcArrayLock.
643 : : */
2090 644 : 18889 : LWLockRelease(XidGenLock);
7656 tgl@sss.pgh.pa.us 645 : 18889 : LWLockRelease(ProcArrayLock);
646 : 18889 : }
647 : :
648 : :
649 : : /*
650 : : * ProcArrayEndTransaction -- mark a transaction as no longer running
651 : : *
652 : : * This is used interchangeably for commit and abort cases. The transaction
653 : : * commit/abort must already be reported to WAL and pg_xact.
654 : : *
655 : : * proc is currently always MyProc, but we pass it explicitly for flexibility.
656 : : * latestXid is the latest Xid among the transaction's main XID and
657 : : * subtransactions, or InvalidTransactionId if it has no XID. (We must ask
658 : : * the caller to pass latestXid, instead of computing it from the PGPROC's
659 : : * contents, because the subxid information in the PGPROC might be
660 : : * incomplete.)
661 : : */
662 : : void
6814 663 : 423007 : ProcArrayEndTransaction(PGPROC *proc, TransactionId latestXid)
664 : : {
665 [ + + ]: 423007 : if (TransactionIdIsValid(latestXid))
666 : : {
667 : : /*
668 : : * We must lock ProcArrayLock while clearing our advertised XID, so
669 : : * that we do not exit the set of "running" transactions while someone
670 : : * else is taking a snapshot. See discussion in
671 : : * src/backend/access/transam/README.
672 : : */
2090 andres@anarazel.de 673 [ - + ]: 165683 : Assert(TransactionIdIsValid(proc->xid));
674 : :
675 : : /*
676 : : * If we can immediately acquire ProcArrayLock, we clear our own XID
677 : : * and release the lock. If not, use group XID clearing to improve
678 : : * efficiency.
679 : : */
3925 rhaas@postgresql.org 680 [ + + ]: 165683 : if (LWLockConditionalAcquire(ProcArrayLock, LW_EXCLUSIVE))
681 : : {
2090 andres@anarazel.de 682 : 165460 : ProcArrayEndTransactionInternal(proc, latestXid);
3925 rhaas@postgresql.org 683 : 165460 : LWLockRelease(ProcArrayLock);
684 : : }
685 : : else
686 : 223 : ProcArrayGroupClearXid(proc, latestXid);
687 : : }
688 : : else
689 : : {
690 : : /*
691 : : * If we have no XID, we don't need to lock, since we won't affect
692 : : * anyone else's calculation of a snapshot. We might change their
693 : : * estimate of global xmin, but that's OK.
694 : : */
2090 andres@anarazel.de 695 [ - + ]: 257324 : Assert(!TransactionIdIsValid(proc->xid));
696 [ - + ]: 257324 : Assert(proc->subxidStatus.count == 0);
697 [ - + ]: 257324 : Assert(!proc->subxidStatus.overflowed);
698 : :
793 heikki.linnakangas@i 699 : 257324 : proc->vxid.lxid = InvalidLocalTransactionId;
2091 andres@anarazel.de 700 : 257324 : proc->xmin = InvalidTransactionId;
701 : :
702 : : /* be sure this is cleared in abort */
1488 rhaas@postgresql.org 703 : 257324 : proc->delayChkptFlags = 0;
704 : :
705 : : /* must be cleared with xid/xmin: */
706 : : /* avoid unnecessarily dirtying shared cachelines */
1996 alvherre@alvh.no-ip. 707 [ + + ]: 257324 : if (proc->statusFlags & PROC_VACUUM_STATE_MASK)
708 : : {
2120 andres@anarazel.de 709 [ - + ]: 16856 : Assert(!LWLockHeldByMe(ProcArrayLock));
1986 alvherre@alvh.no-ip. 710 : 16856 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
1996 711 [ - + ]: 16856 : Assert(proc->statusFlags == ProcGlobal->statusFlags[proc->pgxactoff]);
712 : 16856 : proc->statusFlags &= ~PROC_VACUUM_STATE_MASK;
713 : 16856 : ProcGlobal->statusFlags[proc->pgxactoff] = proc->statusFlags;
2120 andres@anarazel.de 714 : 16856 : LWLockRelease(ProcArrayLock);
715 : : }
716 : : }
6814 tgl@sss.pgh.pa.us 717 : 423007 : }
718 : :
719 : : /*
720 : : * Mark a write transaction as no longer running.
721 : : *
722 : : * We don't do any locking here; caller must handle that.
723 : : */
724 : : static inline void
2090 andres@anarazel.de 725 : 165683 : ProcArrayEndTransactionInternal(PGPROC *proc, TransactionId latestXid)
726 : : {
1692 fujii@postgresql.org 727 : 165683 : int pgxactoff = proc->pgxactoff;
728 : :
729 : : /*
730 : : * Note: we need exclusive lock here because we're going to change other
731 : : * processes' PGPROC entries.
732 : : */
1994 alvherre@alvh.no-ip. 733 [ - + ]: 165683 : Assert(LWLockHeldByMeInMode(ProcArrayLock, LW_EXCLUSIVE));
2090 andres@anarazel.de 734 [ - + ]: 165683 : Assert(TransactionIdIsValid(ProcGlobal->xids[pgxactoff]));
735 [ - + ]: 165683 : Assert(ProcGlobal->xids[pgxactoff] == proc->xid);
736 : :
737 : 165683 : ProcGlobal->xids[pgxactoff] = InvalidTransactionId;
738 : 165683 : proc->xid = InvalidTransactionId;
793 heikki.linnakangas@i 739 : 165683 : proc->vxid.lxid = InvalidLocalTransactionId;
2091 andres@anarazel.de 740 : 165683 : proc->xmin = InvalidTransactionId;
741 : :
742 : : /* be sure this is cleared in abort */
1488 rhaas@postgresql.org 743 : 165683 : proc->delayChkptFlags = 0;
744 : :
745 : : /* must be cleared with xid/xmin: */
746 : : /* avoid unnecessarily dirtying shared cachelines */
1996 alvherre@alvh.no-ip. 747 [ + + ]: 165683 : if (proc->statusFlags & PROC_VACUUM_STATE_MASK)
748 : : {
749 : 866 : proc->statusFlags &= ~PROC_VACUUM_STATE_MASK;
750 : 866 : ProcGlobal->statusFlags[proc->pgxactoff] = proc->statusFlags;
751 : : }
752 : :
753 : : /* Clear the subtransaction-XID cache too while holding the lock */
2090 andres@anarazel.de 754 [ + - - + ]: 165683 : Assert(ProcGlobal->subxidStates[pgxactoff].count == proc->subxidStatus.count &&
755 : : ProcGlobal->subxidStates[pgxactoff].overflowed == proc->subxidStatus.overflowed);
756 [ + + - + ]: 165683 : if (proc->subxidStatus.count > 0 || proc->subxidStatus.overflowed)
757 : : {
758 : 518 : ProcGlobal->subxidStates[pgxactoff].count = 0;
759 : 518 : ProcGlobal->subxidStates[pgxactoff].overflowed = false;
760 : 518 : proc->subxidStatus.count = 0;
761 : 518 : proc->subxidStatus.overflowed = false;
762 : : }
763 : :
764 : : /* Also advance global latestCompletedXid while holding the lock */
2093 765 : 165683 : MaintainLatestCompletedXid(latestXid);
766 : :
767 : : /* Same with xactCompletionCount */
879 heikki.linnakangas@i 768 : 165683 : TransamVariables->xactCompletionCount++;
3925 rhaas@postgresql.org 769 : 165683 : }
770 : :
771 : : /*
772 : : * ProcArrayGroupClearXid -- group XID clearing
773 : : *
774 : : * When we cannot immediately acquire ProcArrayLock in exclusive mode at
775 : : * commit time, add ourselves to a list of processes that need their XIDs
776 : : * cleared. The first process to add itself to the list will acquire
777 : : * ProcArrayLock in exclusive mode and perform ProcArrayEndTransactionInternal
778 : : * on behalf of all group members. This avoids a great deal of contention
779 : : * around ProcArrayLock when many processes are trying to commit at once,
780 : : * since the lock need not be repeatedly handed off from one committing
781 : : * process to the next.
782 : : */
783 : : static void
784 : 223 : ProcArrayGroupClearXid(PGPROC *proc, TransactionId latestXid)
785 : : {
803 heikki.linnakangas@i 786 : 223 : int pgprocno = GetNumberFromPGProc(proc);
2734 andres@anarazel.de 787 : 223 : PROC_HDR *procglobal = ProcGlobal;
788 : : uint32 nextidx;
789 : : uint32 wakeidx;
790 : :
791 : : /* We should definitely have an XID to clear. */
2090 792 [ - + ]: 223 : Assert(TransactionIdIsValid(proc->xid));
793 : :
794 : : /* Add ourselves to the list of processes needing a group XID clear. */
3736 rhaas@postgresql.org 795 : 223 : proc->procArrayGroupMember = true;
796 : 223 : proc->procArrayGroupMemberXid = latestXid;
2391 noah@leadboat.com 797 : 223 : nextidx = pg_atomic_read_u32(&procglobal->procArrayGroupFirst);
798 : : while (true)
799 : : {
3736 rhaas@postgresql.org 800 : 223 : pg_atomic_write_u32(&proc->procArrayGroupNext, nextidx);
801 : :
802 [ + - ]: 223 : if (pg_atomic_compare_exchange_u32(&procglobal->procArrayGroupFirst,
803 : : &nextidx,
804 : : (uint32) pgprocno))
3925 805 : 223 : break;
806 : : }
807 : :
808 : : /*
809 : : * If the list was not empty, the leader will clear our XID. It is
810 : : * impossible to have followers without a leader because the first process
811 : : * that has added itself to the list will always have nextidx as
812 : : * INVALID_PROC_NUMBER.
813 : : */
793 heikki.linnakangas@i 814 [ + + ]: 223 : if (nextidx != INVALID_PROC_NUMBER)
815 : : {
3407 rhaas@postgresql.org 816 : 52 : int extraWaits = 0;
817 : :
818 : : /* Sleep until the leader clears our XID. */
3315 819 : 52 : pgstat_report_wait_start(WAIT_EVENT_PROCARRAY_GROUP_UPDATE);
820 : : for (;;)
821 : : {
822 : : /* acts as a read barrier */
3431 tgl@sss.pgh.pa.us 823 : 52 : PGSemaphoreLock(proc->sem);
3736 rhaas@postgresql.org 824 [ + - ]: 52 : if (!proc->procArrayGroupMember)
3897 825 : 52 : break;
3897 rhaas@postgresql.org 826 :UBC 0 : extraWaits++;
827 : : }
3315 rhaas@postgresql.org 828 :CBC 52 : pgstat_report_wait_end();
829 : :
793 heikki.linnakangas@i 830 [ - + ]: 52 : Assert(pg_atomic_read_u32(&proc->procArrayGroupNext) == INVALID_PROC_NUMBER);
831 : :
832 : : /* Fix semaphore count for any absorbed wakeups */
3925 rhaas@postgresql.org 833 [ - + ]: 52 : while (extraWaits-- > 0)
3431 tgl@sss.pgh.pa.us 834 :UBC 0 : PGSemaphoreUnlock(proc->sem);
3925 rhaas@postgresql.org 835 :CBC 52 : return;
836 : : }
837 : :
838 : : /* We are the leader. Acquire the lock on behalf of everyone. */
839 : 171 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
840 : :
841 : : /*
842 : : * Now that we've got the lock, clear the list of processes waiting for
843 : : * group XID clearing, saving a pointer to the head of the list. Trying
844 : : * to pop elements one at a time could lead to an ABA problem.
845 : : */
2782 akorotkov@postgresql 846 : 171 : nextidx = pg_atomic_exchange_u32(&procglobal->procArrayGroupFirst,
847 : : INVALID_PROC_NUMBER);
848 : :
849 : : /* Remember head of list so we can perform wakeups after dropping lock. */
3925 rhaas@postgresql.org 850 : 171 : wakeidx = nextidx;
851 : :
852 : : /* Walk the list and clear all XIDs. */
793 heikki.linnakangas@i 853 [ + + ]: 394 : while (nextidx != INVALID_PROC_NUMBER)
854 : : {
1692 fujii@postgresql.org 855 : 223 : PGPROC *nextproc = &allProcs[nextidx];
856 : :
857 : 223 : ProcArrayEndTransactionInternal(nextproc, nextproc->procArrayGroupMemberXid);
858 : :
859 : : /* Move to next proc in list. */
860 : 223 : nextidx = pg_atomic_read_u32(&nextproc->procArrayGroupNext);
861 : : }
862 : :
863 : : /* We're done with the lock now. */
3925 rhaas@postgresql.org 864 : 171 : LWLockRelease(ProcArrayLock);
865 : :
866 : : /*
867 : : * Now that we've released the lock, go back and wake everybody up. We
868 : : * don't do this under the lock so as to keep lock hold times to a
869 : : * minimum. The system calls we need to perform to wake other processes
870 : : * up are probably much slower than the simple memory writes we did while
871 : : * holding the lock.
872 : : */
793 heikki.linnakangas@i 873 [ + + ]: 394 : while (wakeidx != INVALID_PROC_NUMBER)
874 : : {
1692 fujii@postgresql.org 875 : 223 : PGPROC *nextproc = &allProcs[wakeidx];
876 : :
877 : 223 : wakeidx = pg_atomic_read_u32(&nextproc->procArrayGroupNext);
793 heikki.linnakangas@i 878 : 223 : pg_atomic_write_u32(&nextproc->procArrayGroupNext, INVALID_PROC_NUMBER);
879 : :
880 : : /* ensure all previous writes are visible before follower continues. */
3897 rhaas@postgresql.org 881 : 223 : pg_write_barrier();
882 : :
1692 fujii@postgresql.org 883 : 223 : nextproc->procArrayGroupMember = false;
884 : :
885 [ + + ]: 223 : if (nextproc != MyProc)
886 : 52 : PGSemaphoreUnlock(nextproc->sem);
887 : : }
888 : : }
889 : :
890 : : /*
891 : : * ProcArrayClearTransaction -- clear the transaction fields
892 : : *
893 : : * This is used after successfully preparing a 2-phase transaction. We are
894 : : * not actually reporting the transaction's XID as no longer running --- it
895 : : * will still appear as running because the 2PC's gxact is in the ProcArray
896 : : * too. We just have to clear out our own PGPROC.
897 : : */
898 : : void
6814 tgl@sss.pgh.pa.us 899 : 297 : ProcArrayClearTransaction(PGPROC *proc)
900 : : {
901 : : int pgxactoff;
902 : :
903 : : /*
904 : : * Currently we need to lock ProcArrayLock exclusively here, as we
905 : : * increment xactCompletionCount below. We also need it at least in shared
906 : : * mode for pgproc->pgxactoff to stay the same below.
907 : : *
908 : : * We could however, as this action does not actually change anyone's view
909 : : * of the set of running XIDs (our entry is duplicate with the gxact that
910 : : * has already been inserted into the ProcArray), lower the lock level to
911 : : * shared if we were to make xactCompletionCount an atomic variable. But
912 : : * that doesn't seem worth it currently, as a 2PC commit is heavyweight
913 : : * enough for this not to be the bottleneck. If it ever becomes a
914 : : * bottleneck it may also be worth considering to combine this with the
915 : : * subsequent ProcArrayRemove()
916 : : */
2085 andres@anarazel.de 917 : 297 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
918 : :
2090 919 : 297 : pgxactoff = proc->pgxactoff;
920 : :
921 : 297 : ProcGlobal->xids[pgxactoff] = InvalidTransactionId;
922 : 297 : proc->xid = InvalidTransactionId;
923 : :
793 heikki.linnakangas@i 924 : 297 : proc->vxid.lxid = InvalidLocalTransactionId;
2091 andres@anarazel.de 925 : 297 : proc->xmin = InvalidTransactionId;
926 : :
1996 alvherre@alvh.no-ip. 927 [ - + ]: 297 : Assert(!(proc->statusFlags & PROC_VACUUM_STATE_MASK));
1488 rhaas@postgresql.org 928 [ - + ]: 297 : Assert(!proc->delayChkptFlags);
929 : :
930 : : /*
931 : : * Need to increment completion count even though transaction hasn't
932 : : * really committed yet. The reason for that is that GetSnapshotData()
933 : : * omits the xid of the current transaction, thus without the increment we
934 : : * otherwise could end up reusing the snapshot later. Which would be bad,
935 : : * because it might not count the prepared transaction as running.
936 : : */
879 heikki.linnakangas@i 937 : 297 : TransamVariables->xactCompletionCount++;
938 : :
939 : : /* Clear the subtransaction-XID cache too */
2090 andres@anarazel.de 940 [ + - - + ]: 297 : Assert(ProcGlobal->subxidStates[pgxactoff].count == proc->subxidStatus.count &&
941 : : ProcGlobal->subxidStates[pgxactoff].overflowed == proc->subxidStatus.overflowed);
942 [ + + - + ]: 297 : if (proc->subxidStatus.count > 0 || proc->subxidStatus.overflowed)
943 : : {
944 : 99 : ProcGlobal->subxidStates[pgxactoff].count = 0;
945 : 99 : ProcGlobal->subxidStates[pgxactoff].overflowed = false;
946 : 99 : proc->subxidStatus.count = 0;
947 : 99 : proc->subxidStatus.overflowed = false;
948 : : }
949 : :
950 : 297 : LWLockRelease(ProcArrayLock);
6814 tgl@sss.pgh.pa.us 951 : 297 : }
952 : :
953 : : /*
954 : : * Update TransamVariables->latestCompletedXid to point to latestXid if
955 : : * currently older.
956 : : */
957 : : static void
2093 andres@anarazel.de 958 : 166842 : MaintainLatestCompletedXid(TransactionId latestXid)
959 : : {
879 heikki.linnakangas@i 960 : 166842 : FullTransactionId cur_latest = TransamVariables->latestCompletedXid;
961 : :
2093 andres@anarazel.de 962 [ - + ]: 166842 : Assert(FullTransactionIdIsValid(cur_latest));
963 [ - + ]: 166842 : Assert(!RecoveryInProgress());
964 [ - + ]: 166842 : Assert(LWLockHeldByMe(ProcArrayLock));
965 : :
966 [ + + ]: 166842 : if (TransactionIdPrecedes(XidFromFullTransactionId(cur_latest), latestXid))
967 : : {
879 heikki.linnakangas@i 968 : 145960 : TransamVariables->latestCompletedXid =
2093 andres@anarazel.de 969 : 145960 : FullXidRelativeTo(cur_latest, latestXid);
970 : : }
971 : :
972 [ + + - + ]: 166842 : Assert(IsBootstrapProcessingMode() ||
973 : : FullTransactionIdIsNormal(TransamVariables->latestCompletedXid));
974 : 166842 : }
975 : :
976 : : /*
977 : : * Same as MaintainLatestCompletedXid, except for use during WAL replay.
978 : : */
979 : : static void
980 : 24594 : MaintainLatestCompletedXidRecovery(TransactionId latestXid)
981 : : {
879 heikki.linnakangas@i 982 : 24594 : FullTransactionId cur_latest = TransamVariables->latestCompletedXid;
983 : : FullTransactionId rel;
984 : :
2093 andres@anarazel.de 985 [ - + - - ]: 24594 : Assert(AmStartupProcess() || !IsUnderPostmaster);
986 [ - + ]: 24594 : Assert(LWLockHeldByMe(ProcArrayLock));
987 : :
988 : : /*
989 : : * Need a FullTransactionId to compare latestXid with. Can't rely on
990 : : * latestCompletedXid to be initialized in recovery. But in recovery it's
991 : : * safe to access nextXid without a lock for the startup process.
992 : : */
879 heikki.linnakangas@i 993 : 24594 : rel = TransamVariables->nextXid;
994 [ - + ]: 24594 : Assert(FullTransactionIdIsValid(TransamVariables->nextXid));
995 : :
2093 andres@anarazel.de 996 [ + + + + ]: 49067 : if (!FullTransactionIdIsValid(cur_latest) ||
997 : 24473 : TransactionIdPrecedes(XidFromFullTransactionId(cur_latest), latestXid))
998 : : {
879 heikki.linnakangas@i 999 : 17859 : TransamVariables->latestCompletedXid =
2093 andres@anarazel.de 1000 : 17859 : FullXidRelativeTo(rel, latestXid);
1001 : : }
1002 : :
879 heikki.linnakangas@i 1003 [ - + ]: 24594 : Assert(FullTransactionIdIsNormal(TransamVariables->latestCompletedXid));
2093 andres@anarazel.de 1004 : 24594 : }
1005 : :
1006 : : /*
1007 : : * ProcArrayInitRecovery -- initialize recovery xid mgmt environment
1008 : : *
1009 : : * Remember up to where the startup process initialized the CLOG and subtrans
1010 : : * so we can ensure it's initialized gaplessly up to the point where necessary
1011 : : * while in recovery.
1012 : : */
1013 : : void
4699 simon@2ndQuadrant.co 1014 : 121 : ProcArrayInitRecovery(TransactionId initializedUptoXID)
1015 : : {
1016 [ - + ]: 121 : Assert(standbyState == STANDBY_INITIALIZED);
1017 [ - + ]: 121 : Assert(TransactionIdIsNormal(initializedUptoXID));
1018 : :
1019 : : /*
1020 : : * we set latestObservedXid to the xid SUBTRANS has been initialized up
1021 : : * to, so we can extend it from that point onwards in
1022 : : * RecordKnownAssignedTransactionIds, and when we get consistent in
1023 : : * ProcArrayApplyRecoveryInfo().
1024 : : */
1025 : 121 : latestObservedXid = initializedUptoXID;
1026 [ - + ]: 121 : TransactionIdRetreat(latestObservedXid);
1027 : 121 : }
1028 : :
1029 : : /*
1030 : : * ProcArrayApplyRecoveryInfo -- apply recovery info about xids
1031 : : *
1032 : : * Takes us through 3 states: Initialized, Pending and Ready.
1033 : : * Normal case is to go all the way to Ready straight away, though there
1034 : : * are atypical cases where we need to take it in steps.
1035 : : *
1036 : : * Use the data about running transactions on the primary to create the initial
1037 : : * state of KnownAssignedXids. We also use these records to regularly prune
1038 : : * KnownAssignedXids because we know it is possible that some transactions
1039 : : * with FATAL errors fail to write abort records, which could cause eventual
1040 : : * overflow.
1041 : : *
1042 : : * See comments for LogStandbySnapshot().
1043 : : */
1044 : : void
5981 1045 : 843 : ProcArrayApplyRecoveryInfo(RunningTransactions running)
1046 : : {
1047 : : TransactionId *xids;
1048 : : TransactionId advanceNextXid;
1049 : : int nxids;
1050 : : int i;
1051 : :
1052 [ - + ]: 843 : Assert(standbyState >= STANDBY_INITIALIZED);
5835 1053 [ - + ]: 843 : Assert(TransactionIdIsValid(running->nextXid));
1054 [ - + ]: 843 : Assert(TransactionIdIsValid(running->oldestRunningXid));
1055 [ - + ]: 843 : Assert(TransactionIdIsNormal(running->latestCompletedXid));
1056 : :
1057 : : /*
1058 : : * Remove stale transactions, if any.
1059 : : */
5981 1060 : 843 : ExpireOldKnownAssignedTransactionIds(running->oldestRunningXid);
1061 : :
1062 : : /*
1063 : : * Adjust TransamVariables->nextXid before StandbyReleaseOldLocks(),
1064 : : * because we will need it up to date for accessing two-phase transactions
1065 : : * in StandbyReleaseOldLocks().
1066 : : */
837 akorotkov@postgresql 1067 : 843 : advanceNextXid = running->nextXid;
1068 [ - + ]: 843 : TransactionIdRetreat(advanceNextXid);
1069 : 843 : AdvanceNextFullTransactionIdPastXid(advanceNextXid);
1070 [ - + ]: 843 : Assert(FullTransactionIdIsValid(TransamVariables->nextXid));
1071 : :
1072 : : /*
1073 : : * Remove stale locks, if any.
1074 : : */
2880 simon@2ndQuadrant.co 1075 : 843 : StandbyReleaseOldLocks(running->oldestRunningXid);
1076 : :
1077 : : /*
1078 : : * If our snapshot is already valid, nothing else to do...
1079 : : */
5981 1080 [ + + ]: 843 : if (standbyState == STANDBY_SNAPSHOT_READY)
1081 : 722 : return;
1082 : :
1083 : : /*
1084 : : * If our initial RunningTransactionsData had an overflowed snapshot then
1085 : : * we knew we were missing some subxids from our snapshot. If we continue
1086 : : * to see overflowed snapshots then we might never be able to start up, so
1087 : : * we make another test to see if our snapshot is now valid. We know that
1088 : : * the missing subxids are equal to or earlier than nextXid. After we
1089 : : * initialise we continue to apply changes during recovery, so once the
1090 : : * oldestRunningXid is later than the nextXid from the initial snapshot we
1091 : : * know that we no longer have missing information and can mark the
1092 : : * snapshot as valid.
1093 : : */
1094 [ - + ]: 121 : if (standbyState == STANDBY_SNAPSHOT_PENDING)
1095 : : {
1096 : : /*
1097 : : * If the snapshot isn't overflowed or if its empty we can reset our
1098 : : * pending state and use this snapshot instead.
1099 : : */
677 heikki.linnakangas@i 1100 [ # # # # ]:UBC 0 : if (running->subxid_status != SUBXIDS_MISSING || running->xcnt == 0)
1101 : : {
1102 : : /*
1103 : : * If we have already collected known assigned xids, we need to
1104 : : * throw them away before we apply the recovery snapshot.
1105 : : */
5079 simon@2ndQuadrant.co 1106 : 0 : KnownAssignedXidsReset();
5298 1107 : 0 : standbyState = STANDBY_INITIALIZED;
1108 : : }
1109 : : else
1110 : : {
1111 [ # # ]: 0 : if (TransactionIdPrecedes(standbySnapshotPendingXmin,
1112 : : running->oldestRunningXid))
1113 : : {
1114 : 0 : standbyState = STANDBY_SNAPSHOT_READY;
876 michael@paquier.xyz 1115 [ # # ]: 0 : elog(DEBUG1,
1116 : : "recovery snapshots are now enabled");
1117 : : }
1118 : : else
1119 [ # # ]: 0 : elog(DEBUG1,
1120 : : "recovery snapshot waiting for non-overflowed snapshot or "
1121 : : "until oldest active xid on standby is at least %u (now %u)",
1122 : : standbySnapshotPendingXmin,
1123 : : running->oldestRunningXid);
5298 simon@2ndQuadrant.co 1124 : 0 : return;
1125 : : }
1126 : : }
1127 : :
5836 simon@2ndQuadrant.co 1128 [ - + ]:CBC 121 : Assert(standbyState == STANDBY_INITIALIZED);
1129 : :
1130 : : /*
1131 : : * NB: this can be reached at least twice, so make sure new code can deal
1132 : : * with that.
1133 : : */
1134 : :
1135 : : /*
1136 : : * Nobody else is running yet, but take locks anyhow
1137 : : */
1138 : 121 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
1139 : :
1140 : : /*
1141 : : * KnownAssignedXids is sorted so we cannot just add the xids, we have to
1142 : : * sort them first.
1143 : : *
1144 : : * Some of the new xids are top-level xids and some are subtransactions.
1145 : : * We don't call SubTransSetParent because it doesn't matter yet. If we
1146 : : * aren't overflowed then all xids will fit in snapshot and so we don't
1147 : : * need subtrans. If we later overflow, an xid assignment record will add
1148 : : * xids to subtrans. If RunningTransactionsData is overflowed then we
1149 : : * don't have enough information to correctly update subtrans anyway.
1150 : : */
1151 : :
1152 : : /*
1153 : : * Allocate a temporary array to avoid modifying the array passed as
1154 : : * argument.
1155 : : */
146 michael@paquier.xyz 1156 :GNC 121 : xids = palloc_array(TransactionId, running->xcnt + running->subxcnt);
1157 : :
1158 : : /*
1159 : : * Add to the temp array any xids which have not already completed.
1160 : : */
5628 heikki.linnakangas@i 1161 :CBC 121 : nxids = 0;
4902 simon@2ndQuadrant.co 1162 [ + + ]: 126 : for (i = 0; i < running->xcnt + running->subxcnt; i++)
1163 : : {
5836 1164 : 5 : TransactionId xid = running->xids[i];
1165 : :
1166 : : /*
1167 : : * The running-xacts snapshot can contain xids that were still visible
1168 : : * in the procarray when the snapshot was taken, but were already
1169 : : * WAL-logged as completed. They're not running anymore, so ignore
1170 : : * them.
1171 : : */
5981 1172 [ + - - + ]: 5 : if (TransactionIdDidCommit(xid) || TransactionIdDidAbort(xid))
5981 simon@2ndQuadrant.co 1173 :UBC 0 : continue;
1174 : :
5836 simon@2ndQuadrant.co 1175 :CBC 5 : xids[nxids++] = xid;
1176 : : }
1177 : :
1178 [ + + ]: 121 : if (nxids > 0)
1179 : : {
5079 1180 [ - + ]: 5 : if (procArray->numKnownAssignedXids != 0)
1181 : : {
5079 simon@2ndQuadrant.co 1182 :UBC 0 : LWLockRelease(ProcArrayLock);
1183 [ # # ]: 0 : elog(ERROR, "KnownAssignedXids is not empty");
1184 : : }
1185 : :
1186 : : /*
1187 : : * Sort the array so that we can add them safely into
1188 : : * KnownAssignedXids.
1189 : : *
1190 : : * We have to sort them logically, because in KnownAssignedXidsAdd we
1191 : : * call TransactionIdFollowsOrEquals and so on. But we know these XIDs
1192 : : * come from RUNNING_XACTS, which means there are only normal XIDs
1193 : : * from the same epoch, so this is safe.
1194 : : */
1559 tomas.vondra@postgre 1195 :CBC 5 : qsort(xids, nxids, sizeof(TransactionId), xidLogicalComparator);
1196 : :
1197 : : /*
1198 : : * Add the sorted snapshot into KnownAssignedXids. The running-xacts
1199 : : * snapshot may include duplicated xids because of prepared
1200 : : * transactions, so ignore them.
1201 : : */
5836 simon@2ndQuadrant.co 1202 [ + + ]: 10 : for (i = 0; i < nxids; i++)
1203 : : {
2760 michael@paquier.xyz 1204 [ - + - - ]: 5 : if (i > 0 && TransactionIdEquals(xids[i - 1], xids[i]))
1205 : : {
2760 michael@paquier.xyz 1206 [ # # ]:UBC 0 : elog(DEBUG1,
1207 : : "found duplicated transaction %u for KnownAssignedXids insertion",
1208 : : xids[i]);
1209 : 0 : continue;
1210 : : }
5628 heikki.linnakangas@i 1211 :CBC 5 : KnownAssignedXidsAdd(xids[i], xids[i], true);
1212 : : }
1213 : :
876 michael@paquier.xyz 1214 : 5 : KnownAssignedXidsDisplay(DEBUG3);
1215 : : }
1216 : :
5836 simon@2ndQuadrant.co 1217 : 121 : pfree(xids);
1218 : :
1219 : : /*
1220 : : * latestObservedXid is at least set to the point where SUBTRANS was
1221 : : * started up to (cf. ProcArrayInitRecovery()) or to the biggest xid
1222 : : * RecordKnownAssignedTransactionIds() was called for. Initialize
1223 : : * subtrans from thereon, up to nextXid - 1.
1224 : : *
1225 : : * We need to duplicate parts of RecordKnownAssignedTransactionId() here,
1226 : : * because we've just added xids to the known assigned xids machinery that
1227 : : * haven't gone through RecordKnownAssignedTransactionId().
1228 : : */
4699 1229 [ - + ]: 121 : Assert(TransactionIdIsNormal(latestObservedXid));
4547 heikki.linnakangas@i 1230 [ - + ]: 121 : TransactionIdAdvance(latestObservedXid);
4699 simon@2ndQuadrant.co 1231 [ - + ]: 242 : while (TransactionIdPrecedes(latestObservedXid, running->nextXid))
1232 : : {
4699 simon@2ndQuadrant.co 1233 :UBC 0 : ExtendSUBTRANS(latestObservedXid);
1234 [ # # ]: 0 : TransactionIdAdvance(latestObservedXid);
1235 : : }
4382 bruce@momjian.us 1236 [ - + ]:CBC 121 : TransactionIdRetreat(latestObservedXid); /* = running->nextXid - 1 */
1237 : :
1238 : : /* ----------
1239 : : * Now we've got the running xids we need to set the global values that
1240 : : * are used to track snapshots as they evolve further.
1241 : : *
1242 : : * - latestCompletedXid which will be the xmax for snapshots
1243 : : * - lastOverflowedXid which shows whether snapshots overflow
1244 : : * - nextXid
1245 : : *
1246 : : * If the snapshot overflowed, then we still initialise with what we know,
1247 : : * but the recovery snapshot isn't fully valid yet because we know there
1248 : : * are some subxids missing. We don't know the specific subxids that are
1249 : : * missing, so conservatively assume the last one is latestObservedXid.
1250 : : * ----------
1251 : : */
677 heikki.linnakangas@i 1252 [ - + ]: 121 : if (running->subxid_status == SUBXIDS_MISSING)
1253 : : {
5836 simon@2ndQuadrant.co 1254 :UBC 0 : standbyState = STANDBY_SNAPSHOT_PENDING;
1255 : :
1256 : 0 : standbySnapshotPendingXmin = latestObservedXid;
5628 heikki.linnakangas@i 1257 : 0 : procArray->lastOverflowedXid = latestObservedXid;
1258 : : }
1259 : : else
1260 : : {
5836 simon@2ndQuadrant.co 1261 :CBC 121 : standbyState = STANDBY_SNAPSHOT_READY;
1262 : :
1263 : 121 : standbySnapshotPendingXmin = InvalidTransactionId;
1264 : :
1265 : : /*
1266 : : * If the 'xids' array didn't include all subtransactions, we have to
1267 : : * mark any snapshots taken as overflowed.
1268 : : */
677 heikki.linnakangas@i 1269 [ + + ]: 121 : if (running->subxid_status == SUBXIDS_IN_SUBTRANS)
1270 : 26 : procArray->lastOverflowedXid = latestObservedXid;
1271 : : else
1272 : : {
1273 [ - + ]: 95 : Assert(running->subxid_status == SUBXIDS_IN_ARRAY);
1274 : 95 : procArray->lastOverflowedXid = InvalidTransactionId;
1275 : : }
1276 : : }
1277 : :
1278 : : /*
1279 : : * If a transaction wrote a commit record in the gap between taking and
1280 : : * logging the snapshot then latestCompletedXid may already be higher than
1281 : : * the value from the snapshot, so check before we use the incoming value.
1282 : : * It also might not yet be set at all.
1283 : : */
2093 andres@anarazel.de 1284 : 121 : MaintainLatestCompletedXidRecovery(running->latestCompletedXid);
1285 : :
1286 : : /*
1287 : : * NB: No need to increment TransamVariables->xactCompletionCount here,
1288 : : * nobody can see it yet.
1289 : : */
1290 : :
5202 tgl@sss.pgh.pa.us 1291 : 121 : LWLockRelease(ProcArrayLock);
1292 : :
876 michael@paquier.xyz 1293 : 121 : KnownAssignedXidsDisplay(DEBUG3);
5981 simon@2ndQuadrant.co 1294 [ + - ]: 121 : if (standbyState == STANDBY_SNAPSHOT_READY)
876 michael@paquier.xyz 1295 [ + + ]: 121 : elog(DEBUG1, "recovery snapshots are now enabled");
1296 : : else
876 michael@paquier.xyz 1297 [ # # ]:UBC 0 : elog(DEBUG1,
1298 : : "recovery snapshot waiting for non-overflowed snapshot or "
1299 : : "until oldest active xid on standby is at least %u (now %u)",
1300 : : standbySnapshotPendingXmin,
1301 : : running->oldestRunningXid);
1302 : : }
1303 : :
1304 : : /*
1305 : : * ProcArrayApplyXidAssignment
1306 : : * Process an XLOG_XACT_ASSIGNMENT WAL record
1307 : : */
1308 : : void
5981 simon@2ndQuadrant.co 1309 :CBC 21 : ProcArrayApplyXidAssignment(TransactionId topxid,
1310 : : int nsubxids, TransactionId *subxids)
1311 : : {
1312 : : TransactionId max_xid;
1313 : : int i;
1314 : :
5836 1315 [ - + ]: 21 : Assert(standbyState >= STANDBY_INITIALIZED);
1316 : :
5981 1317 : 21 : max_xid = TransactionIdLatest(topxid, nsubxids, subxids);
1318 : :
1319 : : /*
1320 : : * Mark all the subtransactions as observed.
1321 : : *
1322 : : * NOTE: This will fail if the subxid contains too many previously
1323 : : * unobserved xids to fit into known-assigned-xids. That shouldn't happen
1324 : : * as the code stands, because xid-assignment records should never contain
1325 : : * more than PGPROC_MAX_CACHED_SUBXIDS entries.
1326 : : */
1327 : 21 : RecordKnownAssignedTransactionIds(max_xid);
1328 : :
1329 : : /*
1330 : : * Notice that we update pg_subtrans with the top-level xid, rather than
1331 : : * the parent xid. This is a difference between normal processing and
1332 : : * recovery, yet is still correct in all cases. The reason is that
1333 : : * subtransaction commit is not marked in clog until commit processing, so
1334 : : * all aborted subtransactions have already been clearly marked in clog.
1335 : : * As a result we are able to refer directly to the top-level
1336 : : * transaction's state rather than skipping through all the intermediate
1337 : : * states in the subtransaction tree. This should be the first time we
1338 : : * have attempted to SubTransSetParent().
1339 : : */
1340 [ + + ]: 1365 : for (i = 0; i < nsubxids; i++)
3295 1341 : 1344 : SubTransSetParent(subxids[i], topxid);
1342 : :
1343 : : /* KnownAssignedXids isn't maintained yet, so we're done for now */
4547 heikki.linnakangas@i 1344 [ - + ]: 21 : if (standbyState == STANDBY_INITIALIZED)
4547 heikki.linnakangas@i 1345 :UBC 0 : return;
1346 : :
1347 : : /*
1348 : : * Uses same locking as transaction commit
1349 : : */
5981 simon@2ndQuadrant.co 1350 :CBC 21 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
1351 : :
1352 : : /*
1353 : : * Remove subxids from known-assigned-xacts.
1354 : : */
5851 tgl@sss.pgh.pa.us 1355 : 21 : KnownAssignedXidsRemoveTree(InvalidTransactionId, nsubxids, subxids);
1356 : :
1357 : : /*
1358 : : * Advance lastOverflowedXid to be at least the last of these subxids.
1359 : : */
5981 simon@2ndQuadrant.co 1360 [ + - ]: 21 : if (TransactionIdPrecedes(procArray->lastOverflowedXid, max_xid))
1361 : 21 : procArray->lastOverflowedXid = max_xid;
1362 : :
1363 : 21 : LWLockRelease(ProcArrayLock);
1364 : : }
1365 : :
1366 : : /*
1367 : : * TransactionIdIsInProgress -- is given transaction running in some backend
1368 : : *
1369 : : * Aside from some shortcuts such as checking RecentXmin and our own Xid,
1370 : : * there are four possibilities for finding a running transaction:
1371 : : *
1372 : : * 1. The given Xid is a main transaction Id. We will find this out cheaply
1373 : : * by looking at ProcGlobal->xids.
1374 : : *
1375 : : * 2. The given Xid is one of the cached subxact Xids in the PGPROC array.
1376 : : * We can find this out cheaply too.
1377 : : *
1378 : : * 3. In Hot Standby mode, we must search the KnownAssignedXids list to see
1379 : : * if the Xid is running on the primary.
1380 : : *
1381 : : * 4. Search the SubTrans tree to find the Xid's topmost parent, and then see
1382 : : * if that is running according to ProcGlobal->xids[] or KnownAssignedXids.
1383 : : * This is the slowest way, but sadly it has to be done always if the others
1384 : : * failed, unless we see that the cached subxact sets are complete (none have
1385 : : * overflowed).
1386 : : *
1387 : : * ProcArrayLock has to be held while we do 1, 2, 3. If we save the top Xids
1388 : : * while doing 1 and 3, we can release the ProcArrayLock while we do 4.
1389 : : * This buys back some concurrency (and we can't retrieve the main Xids from
1390 : : * ProcGlobal->xids[] again anyway; see GetNewTransactionId).
1391 : : */
1392 : : bool
7656 tgl@sss.pgh.pa.us 1393 : 16612608 : TransactionIdIsInProgress(TransactionId xid)
1394 : : {
1395 : : static TransactionId *xids = NULL;
1396 : : static TransactionId *other_xids;
1397 : : XidCacheStatus *other_subxidstates;
6801 1398 : 16612608 : int nxids = 0;
7656 1399 : 16612608 : ProcArrayStruct *arrayP = procArray;
1400 : : TransactionId topxid;
1401 : : TransactionId latestCompletedXid;
1402 : : int mypgxactoff;
1403 : : int numProcs;
1404 : : int j;
1405 : :
1406 : : /*
1407 : : * Don't bother checking a transaction older than RecentXmin; it could not
1408 : : * possibly still be running. (Note: in particular, this guarantees that
1409 : : * we reject InvalidTransactionId, FrozenTransactionId, etc as not
1410 : : * running.)
1411 : : */
1412 [ + + ]: 16612608 : if (TransactionIdPrecedes(xid, RecentXmin))
1413 : : {
1414 : : xc_by_recent_xmin_inc();
1415 : 13441361 : return false;
1416 : : }
1417 : :
1418 : : /*
1419 : : * We may have just checked the status of this transaction, so if it is
1420 : : * already known to be completed, we can fall out without any access to
1421 : : * shared memory.
1422 : : */
1408 heikki.linnakangas@i 1423 [ + + ]: 3171247 : if (TransactionIdEquals(cachedXidIsNotInProgress, xid))
1424 : : {
1425 : : xc_by_known_xact_inc();
6629 tgl@sss.pgh.pa.us 1426 : 1419239 : return false;
1427 : : }
1428 : :
1429 : : /*
1430 : : * Also, we can handle our own transaction (and subtransactions) without
1431 : : * any access to shared memory.
1432 : : */
6801 1433 [ + + ]: 1752008 : if (TransactionIdIsCurrentTransactionId(xid))
1434 : : {
1435 : : xc_by_my_xact_inc();
1436 : 219471 : return true;
1437 : : }
1438 : :
1439 : : /*
1440 : : * If first time through, get workspace to remember main XIDs in. We
1441 : : * malloc it permanently to avoid repeated palloc/pfree overhead.
1442 : : */
1443 [ + + ]: 1532537 : if (xids == NULL)
1444 : : {
1445 : : /*
1446 : : * In hot standby mode, reserve enough space to hold all xids in the
1447 : : * known-assigned list. If we later finish recovery, we no longer need
1448 : : * the bigger array, but we don't bother to shrink it.
1449 : : */
5851 1450 [ - + ]: 1002 : int maxxids = RecoveryInProgress() ? TOTAL_MAX_CACHED_SUBXIDS : arrayP->maxProcs;
1451 : :
5981 simon@2ndQuadrant.co 1452 : 1002 : xids = (TransactionId *) malloc(maxxids * sizeof(TransactionId));
6801 tgl@sss.pgh.pa.us 1453 [ - + ]: 1002 : if (xids == NULL)
6801 tgl@sss.pgh.pa.us 1454 [ # # ]:UBC 0 : ereport(ERROR,
1455 : : (errcode(ERRCODE_OUT_OF_MEMORY),
1456 : : errmsg("out of memory")));
1457 : : }
1458 : :
2090 andres@anarazel.de 1459 :CBC 1532537 : other_xids = ProcGlobal->xids;
1460 : 1532537 : other_subxidstates = ProcGlobal->subxidStates;
1461 : :
7656 tgl@sss.pgh.pa.us 1462 : 1532537 : LWLockAcquire(ProcArrayLock, LW_SHARED);
1463 : :
1464 : : /*
1465 : : * Now that we have the lock, we can check latestCompletedXid; if the
1466 : : * target Xid is after that, it's surely still running.
1467 : : */
2093 andres@anarazel.de 1468 : 1532537 : latestCompletedXid =
879 heikki.linnakangas@i 1469 : 1532537 : XidFromFullTransactionId(TransamVariables->latestCompletedXid);
2093 andres@anarazel.de 1470 [ + + ]: 1532537 : if (TransactionIdPrecedes(latestCompletedXid, xid))
1471 : : {
6799 tgl@sss.pgh.pa.us 1472 : 1509845 : LWLockRelease(ProcArrayLock);
1473 : : xc_by_latest_xid_inc();
1474 : 1509845 : return true;
1475 : : }
1476 : :
1477 : : /* No shortcuts, gotta grovel through the array */
2090 andres@anarazel.de 1478 : 22692 : mypgxactoff = MyProc->pgxactoff;
1479 : 22692 : numProcs = arrayP->numProcs;
1692 fujii@postgresql.org 1480 [ + + ]: 240892 : for (int pgxactoff = 0; pgxactoff < numProcs; pgxactoff++)
1481 : : {
1482 : : int pgprocno;
1483 : : PGPROC *proc;
1484 : : TransactionId pxid;
1485 : : int pxids;
1486 : :
1487 : : /* Ignore ourselves --- dealt with it above */
2090 andres@anarazel.de 1488 [ + + ]: 223213 : if (pgxactoff == mypgxactoff)
6801 tgl@sss.pgh.pa.us 1489 : 20347 : continue;
1490 : :
1491 : : /* Fetch xid just once - see GetNewTransactionId */
2090 andres@anarazel.de 1492 : 202866 : pxid = UINT32_ACCESS_ONCE(other_xids[pgxactoff]);
1493 : :
7656 tgl@sss.pgh.pa.us 1494 [ + + ]: 202866 : if (!TransactionIdIsValid(pxid))
1495 : 126640 : continue;
1496 : :
1497 : : /*
1498 : : * Step 1: check the main Xid
1499 : : */
1500 [ + + ]: 76226 : if (TransactionIdEquals(pxid, xid))
1501 : : {
6801 1502 : 4884 : LWLockRelease(ProcArrayLock);
1503 : : xc_by_main_xid_inc();
1504 : 4884 : return true;
1505 : : }
1506 : :
1507 : : /*
1508 : : * We can ignore main Xids that are younger than the target Xid, since
1509 : : * the target could not possibly be their child.
1510 : : */
7656 1511 [ + + ]: 71342 : if (TransactionIdPrecedes(xid, pxid))
1512 : 38227 : continue;
1513 : :
1514 : : /*
1515 : : * Step 2: check the cached child-Xids arrays
1516 : : */
2090 andres@anarazel.de 1517 : 33115 : pxids = other_subxidstates[pgxactoff].count;
2734 1518 : 33115 : pg_read_barrier(); /* pairs with barrier in GetNewTransactionId() */
2090 1519 : 33115 : pgprocno = arrayP->pgprocnos[pgxactoff];
1520 : 33115 : proc = &allProcs[pgprocno];
2734 1521 [ + + ]: 47682 : for (j = pxids - 1; j >= 0; j--)
1522 : : {
1523 : : /* Fetch xid just once - see GetNewTransactionId */
1524 : 14696 : TransactionId cxid = UINT32_ACCESS_ONCE(proc->subxids.xids[j]);
1525 : :
7656 tgl@sss.pgh.pa.us 1526 [ + + ]: 14696 : if (TransactionIdEquals(cxid, xid))
1527 : : {
6801 1528 : 129 : LWLockRelease(ProcArrayLock);
1529 : : xc_by_child_xid_inc();
1530 : 129 : return true;
1531 : : }
1532 : : }
1533 : :
1534 : : /*
1535 : : * Save the main Xid for step 4. We only need to remember main Xids
1536 : : * that have uncached children. (Note: there is no race condition
1537 : : * here because the overflowed flag cannot be cleared, only set, while
1538 : : * we hold ProcArrayLock. So we can't miss an Xid that we need to
1539 : : * worry about.)
1540 : : */
2090 andres@anarazel.de 1541 [ + + ]: 32986 : if (other_subxidstates[pgxactoff].overflowed)
7656 tgl@sss.pgh.pa.us 1542 : 157 : xids[nxids++] = pxid;
1543 : : }
1544 : :
1545 : : /*
1546 : : * Step 3: in hot standby mode, check the known-assigned-xids list. XIDs
1547 : : * in the list must be treated as running.
1548 : : */
5981 simon@2ndQuadrant.co 1549 [ - + ]: 17679 : if (RecoveryInProgress())
1550 : : {
1551 : : /* none of the PGPROC entries should have XIDs in hot standby mode */
5981 simon@2ndQuadrant.co 1552 [ # # ]:LBC (1) : Assert(nxids == 0);
1553 : :
5851 tgl@sss.pgh.pa.us 1554 [ # # ]: (1) : if (KnownAssignedXidExists(xid))
1555 : : {
5981 simon@2ndQuadrant.co 1556 :UBC 0 : LWLockRelease(ProcArrayLock);
1557 : : xc_by_known_assigned_inc();
1558 : 0 : return true;
1559 : : }
1560 : :
1561 : : /*
1562 : : * If the KnownAssignedXids overflowed, we have to check pg_subtrans
1563 : : * too. Fetch all xids from KnownAssignedXids that are lower than
1564 : : * xid, since if xid is a subtransaction its parent will always have a
1565 : : * lower value. Note we will collect both main and subXIDs here, but
1566 : : * there's no help for it.
1567 : : */
5981 simon@2ndQuadrant.co 1568 [ # # ]:LBC (1) : if (TransactionIdPrecedesOrEquals(xid, procArray->lastOverflowedXid))
5981 simon@2ndQuadrant.co 1569 :UBC 0 : nxids = KnownAssignedXidsGet(xids, xid);
1570 : : }
1571 : :
7656 tgl@sss.pgh.pa.us 1572 :CBC 17679 : LWLockRelease(ProcArrayLock);
1573 : :
1574 : : /*
1575 : : * If none of the relevant caches overflowed, we know the Xid is not
1576 : : * running without even looking at pg_subtrans.
1577 : : */
1578 [ + + ]: 17679 : if (nxids == 0)
1579 : : {
1580 : : xc_no_overflow_inc();
1408 heikki.linnakangas@i 1581 : 17522 : cachedXidIsNotInProgress = xid;
6801 tgl@sss.pgh.pa.us 1582 : 17522 : return false;
1583 : : }
1584 : :
1585 : : /*
1586 : : * Step 4: have to check pg_subtrans.
1587 : : *
1588 : : * At this point, we know it's either a subtransaction of one of the Xids
1589 : : * in xids[], or it's not running. If it's an already-failed
1590 : : * subtransaction, we want to say "not running" even though its parent may
1591 : : * still be running. So first, check pg_xact to see if it's been aborted.
1592 : : */
1593 : : xc_slow_answer_inc();
1594 : :
7656 1595 [ - + ]: 157 : if (TransactionIdDidAbort(xid))
1596 : : {
1408 heikki.linnakangas@i 1597 :UBC 0 : cachedXidIsNotInProgress = xid;
6801 tgl@sss.pgh.pa.us 1598 : 0 : return false;
1599 : : }
1600 : :
1601 : : /*
1602 : : * It isn't aborted, so check whether the transaction tree it belongs to
1603 : : * is still running (or, more precisely, whether it was running when we
1604 : : * held ProcArrayLock).
1605 : : */
7656 tgl@sss.pgh.pa.us 1606 :CBC 157 : topxid = SubTransGetTopmostTransaction(xid);
1607 [ - + ]: 157 : Assert(TransactionIdIsValid(topxid));
1321 michael@paquier.xyz 1608 [ + - + - ]: 314 : if (!TransactionIdEquals(topxid, xid) &&
1609 : 157 : pg_lfind32(topxid, xids, nxids))
1610 : 157 : return true;
1611 : :
1408 heikki.linnakangas@i 1612 :UBC 0 : cachedXidIsNotInProgress = xid;
6801 tgl@sss.pgh.pa.us 1613 : 0 : return false;
1614 : : }
1615 : :
1616 : :
1617 : : /*
1618 : : * Determine XID horizons.
1619 : : *
1620 : : * This is used by wrapper functions like GetOldestNonRemovableTransactionId()
1621 : : * (for VACUUM), GetReplicationHorizons() (for hot_standby_feedback), etc as
1622 : : * well as "internally" by GlobalVisUpdate() (see comment above struct
1623 : : * GlobalVisState).
1624 : : *
1625 : : * See the definition of ComputeXidHorizonsResult for the various computed
1626 : : * horizons.
1627 : : *
1628 : : * For VACUUM separate horizons (used to decide which deleted tuples must
1629 : : * be preserved), for shared and non-shared tables are computed. For shared
1630 : : * relations backends in all databases must be considered, but for non-shared
1631 : : * relations that's not required, since only backends in my own database could
1632 : : * ever see the tuples in them. Also, we can ignore concurrently running lazy
1633 : : * VACUUMs because (a) they must be working on other tables, and (b) they
1634 : : * don't need to do snapshot-based lookups.
1635 : : *
1636 : : * This also computes a horizon used to truncate pg_subtrans. For that
1637 : : * backends in all databases have to be considered, and concurrently running
1638 : : * lazy VACUUMs cannot be ignored, as they still may perform pg_subtrans
1639 : : * accesses.
1640 : : *
1641 : : * Note: we include all currently running xids in the set of considered xids.
1642 : : * This ensures that if a just-started xact has not yet set its snapshot,
1643 : : * when it does set the snapshot it cannot set xmin less than what we compute.
1644 : : * See notes in src/backend/access/transam/README.
1645 : : *
1646 : : * Note: despite the above, it's possible for the calculated values to move
1647 : : * backwards on repeated calls. The calculated values are conservative, so
1648 : : * that anything older is definitely not considered as running by anyone
1649 : : * anymore, but the exact values calculated depend on a number of things. For
1650 : : * example, if there are no transactions running in the current database, the
1651 : : * horizon for normal tables will be latestCompletedXid. If a transaction
1652 : : * begins after that, its xmin will include in-progress transactions in other
1653 : : * databases that started earlier, so another call will return a lower value.
1654 : : * Nonetheless it is safe to vacuum a table in the current database with the
1655 : : * first result. There are also replication-related effects: a walsender
1656 : : * process can set its xmin based on transactions that are no longer running
1657 : : * on the primary but are still being replayed on the standby, thus possibly
1658 : : * making the values go backwards. In this case there is a possibility that
1659 : : * we lose data that the standby would like to have, but unless the standby
1660 : : * uses a replication slot to make its xmin persistent there is little we can
1661 : : * do about that --- data is only protected if the walsender runs continuously
1662 : : * while queries are executed on the standby. (The Hot Standby code deals
1663 : : * with such cases by failing standby queries that needed to access
1664 : : * already-removed data, so there's no integrity bug.)
1665 : : *
1666 : : * Note: the approximate horizons (see definition of GlobalVisState) are
1667 : : * updated by the computations done here. That's currently required for
1668 : : * correctness and a small optimization. Without doing so it's possible that
1669 : : * heap vacuum's call to heap_page_prune_and_freeze() uses a more conservative
1670 : : * horizon than later when deciding which tuples can be removed - which the
1671 : : * code doesn't expect (breaking HOT).
1672 : : */
1673 : : static void
2092 andres@anarazel.de 1674 :CBC 127882 : ComputeXidHorizons(ComputeXidHorizonsResult *h)
1675 : : {
7656 tgl@sss.pgh.pa.us 1676 : 127882 : ProcArrayStruct *arrayP = procArray;
1677 : : TransactionId kaxmin;
2092 andres@anarazel.de 1678 : 127882 : bool in_recovery = RecoveryInProgress();
2090 1679 : 127882 : TransactionId *other_xids = ProcGlobal->xids;
1680 : :
1681 : : /* inferred after ProcArrayLock is released */
1435 alvherre@alvh.no-ip. 1682 : 127882 : h->catalog_oldest_nonremovable = InvalidTransactionId;
1683 : :
6814 tgl@sss.pgh.pa.us 1684 : 127882 : LWLockAcquire(ProcArrayLock, LW_SHARED);
1685 : :
879 heikki.linnakangas@i 1686 : 127882 : h->latest_completed = TransamVariables->latestCompletedXid;
1687 : :
1688 : : /*
1689 : : * We initialize the MIN() calculation with latestCompletedXid + 1. This
1690 : : * is a lower bound for the XIDs that might appear in the ProcArray later,
1691 : : * and so protects us against overestimating the result due to future
1692 : : * additions.
1693 : : */
1694 : : {
1695 : : TransactionId initial;
1696 : :
2092 andres@anarazel.de 1697 : 127882 : initial = XidFromFullTransactionId(h->latest_completed);
1698 [ - + ]: 127882 : Assert(TransactionIdIsValid(initial));
1699 [ - + ]: 127882 : TransactionIdAdvance(initial);
1700 : :
1701 : 127882 : h->oldest_considered_running = initial;
1702 : 127882 : h->shared_oldest_nonremovable = initial;
1703 : 127882 : h->data_oldest_nonremovable = initial;
1704 : :
1705 : : /*
1706 : : * Only modifications made by this backend affect the horizon for
1707 : : * temporary relations. Instead of a check in each iteration of the
1708 : : * loop over all PGPROCs it is cheaper to just initialize to the
1709 : : * current top-level xid any.
1710 : : *
1711 : : * Without an assigned xid we could use a horizon as aggressive as
1712 : : * GetNewTransactionId(), but we can get away with the much cheaper
1713 : : * latestCompletedXid + 1: If this backend has no xid there, by
1714 : : * definition, can't be any newer changes in the temp table than
1715 : : * latestCompletedXid.
1716 : : */
2015 1717 [ + + ]: 127882 : if (TransactionIdIsValid(MyProc->xid))
1718 : 48520 : h->temp_oldest_nonremovable = MyProc->xid;
1719 : : else
1720 : 79362 : h->temp_oldest_nonremovable = initial;
1721 : : }
1722 : :
1723 : : /*
1724 : : * Fetch slot horizons while ProcArrayLock is held - the
1725 : : * LWLockAcquire/LWLockRelease are a barrier, ensuring this happens inside
1726 : : * the lock.
1727 : : */
2092 1728 : 127882 : h->slot_xmin = procArray->replication_slot_xmin;
1729 : 127882 : h->slot_catalog_xmin = procArray->replication_slot_catalog_xmin;
1730 : :
1731 [ + + ]: 763358 : for (int index = 0; index < arrayP->numProcs; index++)
1732 : : {
5077 bruce@momjian.us 1733 : 635476 : int pgprocno = arrayP->pgprocnos[index];
2734 andres@anarazel.de 1734 : 635476 : PGPROC *proc = &allProcs[pgprocno];
1996 alvherre@alvh.no-ip. 1735 : 635476 : int8 statusFlags = ProcGlobal->statusFlags[index];
1736 : : TransactionId xid;
1737 : : TransactionId xmin;
1738 : :
1739 : : /* Fetch xid just once - see GetNewTransactionId */
2088 andres@anarazel.de 1740 : 635476 : xid = UINT32_ACCESS_ONCE(other_xids[index]);
2091 1741 : 635476 : xmin = UINT32_ACCESS_ONCE(proc->xmin);
1742 : :
1743 : : /*
1744 : : * Consider both the transaction's Xmin, and its Xid.
1745 : : *
1746 : : * We must check both because a transaction might have an Xmin but not
1747 : : * (yet) an Xid; conversely, if it has an Xid, that could determine
1748 : : * some not-yet-set Xmin.
1749 : : */
2092 1750 : 635476 : xmin = TransactionIdOlder(xmin, xid);
1751 : :
1752 : : /* if neither is set, this proc doesn't influence the horizon */
2053 peter@eisentraut.org 1753 [ + + ]: 635476 : if (!TransactionIdIsValid(xmin))
7219 alvherre@alvh.no-ip. 1754 : 339382 : continue;
1755 : :
1756 : : /*
1757 : : * Don't ignore any procs when determining which transactions might be
1758 : : * considered running. While slots should ensure logical decoding
1759 : : * backends are protected even without this check, it can't hurt to
1760 : : * include them here as well..
1761 : : */
2092 andres@anarazel.de 1762 : 296094 : h->oldest_considered_running =
1763 : 296094 : TransactionIdOlder(h->oldest_considered_running, xmin);
1764 : :
1765 : : /*
1766 : : * Skip over backends either vacuuming (which is ok with rows being
1767 : : * removed, as long as pg_subtrans is not truncated) or doing logical
1768 : : * decoding (which manages xmin separately, check below).
1769 : : */
1996 alvherre@alvh.no-ip. 1770 [ + + ]: 296094 : if (statusFlags & (PROC_IN_VACUUM | PROC_IN_LOGICAL_DECODING))
2092 andres@anarazel.de 1771 : 18730 : continue;
1772 : :
1773 : : /* shared tables need to take backends in all databases into account */
1774 : 277364 : h->shared_oldest_nonremovable =
1775 : 277364 : TransactionIdOlder(h->shared_oldest_nonremovable, xmin);
1776 : :
1777 : : /*
1778 : : * Normally sessions in other databases are ignored for anything but
1779 : : * the shared horizon.
1780 : : *
1781 : : * However, include them when MyDatabaseId is not (yet) set. A
1782 : : * backend in the process of starting up must not compute a "too
1783 : : * aggressive" horizon, otherwise we could end up using it to prune
1784 : : * still-needed data away. If the current backend never connects to a
1785 : : * database this is harmless, because data_oldest_nonremovable will
1786 : : * never be utilized.
1787 : : *
1788 : : * Also, sessions marked with PROC_AFFECTS_ALL_HORIZONS should always
1789 : : * be included. (This flag is used for hot standby feedback, which
1790 : : * can't be tied to a specific database.)
1791 : : *
1792 : : * Also, while in recovery we cannot compute an accurate per-database
1793 : : * horizon, as all xids are managed via the KnownAssignedXids
1794 : : * machinery.
1795 : : */
1481 tgl@sss.pgh.pa.us 1796 [ + + ]: 277364 : if (proc->databaseId == MyDatabaseId ||
1797 [ + + ]: 12636 : MyDatabaseId == InvalidOid ||
1798 [ + + - + ]: 608 : (statusFlags & PROC_AFFECTS_ALL_HORIZONS) ||
1799 : : in_recovery)
1800 : : {
1435 alvherre@alvh.no-ip. 1801 : 276761 : h->data_oldest_nonremovable =
1802 : 276761 : TransactionIdOlder(h->data_oldest_nonremovable, xmin);
1803 : : }
1804 : : }
1805 : :
1806 : : /*
1807 : : * If in recovery fetch oldest xid in KnownAssignedXids, will be applied
1808 : : * after lock is released.
1809 : : */
2092 andres@anarazel.de 1810 [ + + ]: 127882 : if (in_recovery)
1811 : 382 : kaxmin = KnownAssignedXidsGetOldestXmin();
1812 : :
1813 : : /*
1814 : : * No other information from shared state is needed, release the lock
1815 : : * immediately. The rest of the computations can be done without a lock.
1816 : : */
1817 : 127882 : LWLockRelease(ProcArrayLock);
1818 : :
1819 [ + + ]: 127882 : if (in_recovery)
1820 : : {
1821 : 382 : h->oldest_considered_running =
1822 : 382 : TransactionIdOlder(h->oldest_considered_running, kaxmin);
1823 : 382 : h->shared_oldest_nonremovable =
1824 : 382 : TransactionIdOlder(h->shared_oldest_nonremovable, kaxmin);
1825 : 382 : h->data_oldest_nonremovable =
1826 : 382 : TransactionIdOlder(h->data_oldest_nonremovable, kaxmin);
1827 : : /* temp relations cannot be accessed in recovery */
1828 : : }
1829 : :
1107 1830 [ - + ]: 127882 : Assert(TransactionIdPrecedesOrEquals(h->oldest_considered_running,
1831 : : h->shared_oldest_nonremovable));
1832 [ - + ]: 127882 : Assert(TransactionIdPrecedesOrEquals(h->shared_oldest_nonremovable,
1833 : : h->data_oldest_nonremovable));
1834 : :
1835 : : /*
1836 : : * Check whether there are replication slots requiring an older xmin.
1837 : : */
2092 1838 : 127882 : h->shared_oldest_nonremovable =
1839 : 127882 : TransactionIdOlder(h->shared_oldest_nonremovable, h->slot_xmin);
1840 : 127882 : h->data_oldest_nonremovable =
1841 : 127882 : TransactionIdOlder(h->data_oldest_nonremovable, h->slot_xmin);
1842 : :
1843 : : /*
1844 : : * The only difference between catalog / data horizons is that the slot's
1845 : : * catalog xmin is applied to the catalog one (so catalogs can be accessed
1846 : : * for logical decoding). Initialize with data horizon, and then back up
1847 : : * further if necessary. Have to back up the shared horizon as well, since
1848 : : * that also can contain catalogs.
1849 : : */
1850 : 127882 : h->shared_oldest_nonremovable_raw = h->shared_oldest_nonremovable;
1851 : 127882 : h->shared_oldest_nonremovable =
1852 : 127882 : TransactionIdOlder(h->shared_oldest_nonremovable,
1853 : : h->slot_catalog_xmin);
1435 alvherre@alvh.no-ip. 1854 : 127882 : h->catalog_oldest_nonremovable = h->data_oldest_nonremovable;
2092 andres@anarazel.de 1855 : 127882 : h->catalog_oldest_nonremovable =
1856 : 127882 : TransactionIdOlder(h->catalog_oldest_nonremovable,
1857 : : h->slot_catalog_xmin);
1858 : :
1859 : : /*
1860 : : * It's possible that slots backed up the horizons further than
1861 : : * oldest_considered_running. Fix.
1862 : : */
1863 : 127882 : h->oldest_considered_running =
1864 : 127882 : TransactionIdOlder(h->oldest_considered_running,
1865 : : h->shared_oldest_nonremovable);
1866 : 127882 : h->oldest_considered_running =
1867 : 127882 : TransactionIdOlder(h->oldest_considered_running,
1868 : : h->catalog_oldest_nonremovable);
1869 : 127882 : h->oldest_considered_running =
1870 : 127882 : TransactionIdOlder(h->oldest_considered_running,
1871 : : h->data_oldest_nonremovable);
1872 : :
1873 : : /*
1874 : : * shared horizons have to be at least as old as the oldest visible in
1875 : : * current db
1876 : : */
1877 [ - + ]: 127882 : Assert(TransactionIdPrecedesOrEquals(h->shared_oldest_nonremovable,
1878 : : h->data_oldest_nonremovable));
1879 [ - + ]: 127882 : Assert(TransactionIdPrecedesOrEquals(h->shared_oldest_nonremovable,
1880 : : h->catalog_oldest_nonremovable));
1881 : :
1882 : : /*
1883 : : * Horizons need to ensure that pg_subtrans access is still possible for
1884 : : * the relevant backends.
1885 : : */
1886 [ - + ]: 127882 : Assert(TransactionIdPrecedesOrEquals(h->oldest_considered_running,
1887 : : h->shared_oldest_nonremovable));
1888 [ - + ]: 127882 : Assert(TransactionIdPrecedesOrEquals(h->oldest_considered_running,
1889 : : h->catalog_oldest_nonremovable));
1890 [ - + ]: 127882 : Assert(TransactionIdPrecedesOrEquals(h->oldest_considered_running,
1891 : : h->data_oldest_nonremovable));
2015 1892 [ - + ]: 127882 : Assert(TransactionIdPrecedesOrEquals(h->oldest_considered_running,
1893 : : h->temp_oldest_nonremovable));
2092 1894 [ + + - + ]: 127882 : Assert(!TransactionIdIsValid(h->slot_xmin) ||
1895 : : TransactionIdPrecedesOrEquals(h->oldest_considered_running,
1896 : : h->slot_xmin));
1897 [ + + - + ]: 127882 : Assert(!TransactionIdIsValid(h->slot_catalog_xmin) ||
1898 : : TransactionIdPrecedesOrEquals(h->oldest_considered_running,
1899 : : h->slot_catalog_xmin));
1900 : :
1901 : : /* update approximate horizons with the computed horizons */
1902 : 127882 : GlobalVisUpdateApply(h);
1903 : 127882 : }
1904 : :
1905 : : /*
1906 : : * Determine what kind of visibility horizon needs to be used for a
1907 : : * relation. If rel is NULL, the most conservative horizon is used.
1908 : : */
1909 : : static inline GlobalVisHorizonKind
1746 1910 : 20044435 : GlobalVisHorizonKindForRel(Relation rel)
1911 : : {
1912 : : /*
1913 : : * Other relkinds currently don't contain xids, nor always the necessary
1914 : : * logical decoding markers.
1915 : : */
1916 [ + + + + : 20044435 : Assert(!rel ||
+ + - + ]
1917 : : rel->rd_rel->relkind == RELKIND_RELATION ||
1918 : : rel->rd_rel->relkind == RELKIND_MATVIEW ||
1919 : : rel->rd_rel->relkind == RELKIND_TOASTVALUE);
1920 : :
1921 [ + + + + : 20044435 : if (rel == NULL || rel->rd_rel->relisshared || RecoveryInProgress())
+ + ]
1922 : 133563 : return VISHORIZON_SHARED;
1923 [ + + ]: 19910872 : else if (IsCatalogRelation(rel) ||
1924 [ + + + + : 15535114 : RelationIsAccessibleInLogicalDecoding(rel))
+ + - + -
- - - + -
+ + - + -
- + + ]
1925 : 4375763 : return VISHORIZON_CATALOG;
1926 [ + + + + ]: 15535109 : else if (!RELATION_IS_LOCAL(rel))
1927 : 15413019 : return VISHORIZON_DATA;
1928 : : else
1929 : 122090 : return VISHORIZON_TEMP;
1930 : : }
1931 : :
1932 : : /*
1933 : : * Return the oldest XID for which deleted tuples must be preserved in the
1934 : : * passed table.
1935 : : *
1936 : : * If rel is not NULL the horizon may be considerably more recent than
1937 : : * otherwise (i.e. fewer tuples will be removable). In the NULL case a horizon
1938 : : * that is correct (but not optimal) for all relations will be returned.
1939 : : *
1940 : : * This is used by VACUUM to decide which deleted tuples must be preserved in
1941 : : * the passed in table.
1942 : : */
1943 : : TransactionId
2092 1944 : 43927 : GetOldestNonRemovableTransactionId(Relation rel)
1945 : : {
1946 : : ComputeXidHorizonsResult horizons;
1947 : :
1948 : 43927 : ComputeXidHorizons(&horizons);
1949 : :
1746 1950 [ + + + + : 43927 : switch (GlobalVisHorizonKindForRel(rel))
- ]
1951 : : {
1952 : 3838 : case VISHORIZON_SHARED:
1953 : 3838 : return horizons.shared_oldest_nonremovable;
1954 : 12623 : case VISHORIZON_CATALOG:
1955 : 12623 : return horizons.catalog_oldest_nonremovable;
1956 : 11761 : case VISHORIZON_DATA:
1957 : 11761 : return horizons.data_oldest_nonremovable;
1958 : 15705 : case VISHORIZON_TEMP:
1959 : 15705 : return horizons.temp_oldest_nonremovable;
1960 : : }
1961 : :
1962 : : /* just to prevent compiler warnings */
1746 andres@anarazel.de 1963 :UBC 0 : return InvalidTransactionId;
1964 : : }
1965 : :
1966 : : /*
1967 : : * Return the oldest transaction id any currently running backend might still
1968 : : * consider running. This should not be used for visibility / pruning
1969 : : * determinations (see GetOldestNonRemovableTransactionId()), but for
1970 : : * decisions like up to where pg_subtrans can be truncated.
1971 : : */
1972 : : TransactionId
2092 andres@anarazel.de 1973 :CBC 1914 : GetOldestTransactionIdConsideredRunning(void)
1974 : : {
1975 : : ComputeXidHorizonsResult horizons;
1976 : :
1977 : 1914 : ComputeXidHorizons(&horizons);
1978 : :
1979 : 1914 : return horizons.oldest_considered_running;
1980 : : }
1981 : :
1982 : : /*
1983 : : * Return the visibility horizons for a hot standby feedback message.
1984 : : */
1985 : : void
1986 : 51 : GetReplicationHorizons(TransactionId *xmin, TransactionId *catalog_xmin)
1987 : : {
1988 : : ComputeXidHorizonsResult horizons;
1989 : :
1990 : 51 : ComputeXidHorizons(&horizons);
1991 : :
1992 : : /*
1993 : : * Don't want to use shared_oldest_nonremovable here, as that contains the
1994 : : * effect of replication slot's catalog_xmin. We want to send a separate
1995 : : * feedback for the catalog horizon, so the primary can remove data table
1996 : : * contents more aggressively.
1997 : : */
1998 : 51 : *xmin = horizons.shared_oldest_nonremovable_raw;
1999 : 51 : *catalog_xmin = horizons.slot_catalog_xmin;
7656 tgl@sss.pgh.pa.us 2000 : 51 : }
2001 : :
2002 : : /*
2003 : : * GetMaxSnapshotXidCount -- get max size for snapshot XID array
2004 : : *
2005 : : * We have to export this for use by snapmgr.c.
2006 : : */
2007 : : int
5309 2008 : 37657 : GetMaxSnapshotXidCount(void)
2009 : : {
2010 : 37657 : return procArray->maxProcs;
2011 : : }
2012 : :
2013 : : /*
2014 : : * GetMaxSnapshotSubxidCount -- get max size for snapshot sub-XID array
2015 : : *
2016 : : * We have to export this for use by snapmgr.c.
2017 : : */
2018 : : int
2019 : 37451 : GetMaxSnapshotSubxidCount(void)
2020 : : {
2021 : 37451 : return TOTAL_MAX_CACHED_SUBXIDS;
2022 : : }
2023 : :
2024 : : /*
2025 : : * Helper function for GetSnapshotData() that checks if the bulk of the
2026 : : * visibility information in the snapshot is still valid. If so, it updates
2027 : : * the fields that need to change and returns true. Otherwise it returns
2028 : : * false.
2029 : : *
2030 : : * This very likely can be evolved to not need ProcArrayLock held (at very
2031 : : * least in the case we already hold a snapshot), but that's for another day.
2032 : : */
2033 : : static bool
2087 andres@anarazel.de 2034 : 2349712 : GetSnapshotDataReuse(Snapshot snapshot)
2035 : : {
2036 : : uint64 curXactCompletionCount;
2037 : :
2038 [ - + ]: 2349712 : Assert(LWLockHeldByMe(ProcArrayLock));
2039 : :
2040 [ + + ]: 2349712 : if (unlikely(snapshot->snapXactCompletionCount == 0))
2041 : 35196 : return false;
2042 : :
879 heikki.linnakangas@i 2043 : 2314516 : curXactCompletionCount = TransamVariables->xactCompletionCount;
2087 andres@anarazel.de 2044 [ + + ]: 2314516 : if (curXactCompletionCount != snapshot->snapXactCompletionCount)
2045 : 474797 : return false;
2046 : :
2047 : : /*
2048 : : * If the current xactCompletionCount is still the same as it was at the
2049 : : * time the snapshot was built, we can be sure that rebuilding the
2050 : : * contents of the snapshot the hard way would result in the same snapshot
2051 : : * contents:
2052 : : *
2053 : : * As explained in transam/README, the set of xids considered running by
2054 : : * GetSnapshotData() cannot change while ProcArrayLock is held. Snapshot
2055 : : * contents only depend on transactions with xids and xactCompletionCount
2056 : : * is incremented whenever a transaction with an xid finishes (while
2057 : : * holding ProcArrayLock exclusively). Thus the xactCompletionCount check
2058 : : * ensures we would detect if the snapshot would have changed.
2059 : : *
2060 : : * As the snapshot contents are the same as it was before, it is safe to
2061 : : * re-enter the snapshot's xmin into the PGPROC array. None of the rows
2062 : : * visible under the snapshot could already have been removed (that'd
2063 : : * require the set of running transactions to change) and it fulfills the
2064 : : * requirement that concurrent GetSnapshotData() calls yield the same
2065 : : * xmin.
2066 : : */
2067 [ + + ]: 1839719 : if (!TransactionIdIsValid(MyProc->xmin))
2068 : 588046 : MyProc->xmin = TransactionXmin = snapshot->xmin;
2069 : :
2070 : 1839719 : RecentXmin = snapshot->xmin;
2071 [ - + ]: 1839719 : Assert(TransactionIdPrecedesOrEquals(TransactionXmin, RecentXmin));
2072 : :
2073 : 1839719 : snapshot->curcid = GetCurrentCommandId(false);
2074 : 1839719 : snapshot->active_count = 0;
2075 : 1839719 : snapshot->regd_count = 0;
2076 : 1839719 : snapshot->copied = false;
2077 : :
2078 : 1839719 : return true;
2079 : : }
2080 : :
2081 : : /*
2082 : : * GetSnapshotData -- returns information about running transactions.
2083 : : *
2084 : : * The returned snapshot includes xmin (lowest still-running xact ID),
2085 : : * xmax (highest completed xact ID + 1), and a list of running xact IDs
2086 : : * in the range xmin <= xid < xmax. It is used as follows:
2087 : : * All xact IDs < xmin are considered finished.
2088 : : * All xact IDs >= xmax are considered still running.
2089 : : * For an xact ID xmin <= xid < xmax, consult list to see whether
2090 : : * it is considered running or not.
2091 : : * This ensures that the set of transactions seen as "running" by the
2092 : : * current xact will not change after it takes the snapshot.
2093 : : *
2094 : : * All running top-level XIDs are included in the snapshot, except for lazy
2095 : : * VACUUM processes. We also try to include running subtransaction XIDs,
2096 : : * but since PGPROC has only a limited cache area for subxact XIDs, full
2097 : : * information may not be available. If we find any overflowed subxid arrays,
2098 : : * we have to mark the snapshot's subxid data as overflowed, and extra work
2099 : : * *may* need to be done to determine what's running (see XidInMVCCSnapshot()).
2100 : : *
2101 : : * We also update the following backend-global variables:
2102 : : * TransactionXmin: the oldest xmin of any snapshot in use in the
2103 : : * current transaction (this is the same as MyProc->xmin).
2104 : : * RecentXmin: the xmin computed for the most recent snapshot. XIDs
2105 : : * older than this are known not running any more.
2106 : : *
2107 : : * And try to advance the bounds of GlobalVis{Shared,Catalog,Data,Temp}Rels
2108 : : * for the benefit of the GlobalVisTest* family of functions.
2109 : : *
2110 : : * Note: this function should probably not be called with an argument that's
2111 : : * not statically allocated (see xip allocation below).
2112 : : */
2113 : : Snapshot
6567 alvherre@alvh.no-ip. 2114 : 2349712 : GetSnapshotData(Snapshot snapshot)
2115 : : {
7656 tgl@sss.pgh.pa.us 2116 : 2349712 : ProcArrayStruct *arrayP = procArray;
2090 andres@anarazel.de 2117 : 2349712 : TransactionId *other_xids = ProcGlobal->xids;
2118 : : TransactionId xmin;
2119 : : TransactionId xmax;
1692 fujii@postgresql.org 2120 : 2349712 : int count = 0;
7184 tgl@sss.pgh.pa.us 2121 : 2349712 : int subcount = 0;
5981 simon@2ndQuadrant.co 2122 : 2349712 : bool suboverflowed = false;
2123 : : FullTransactionId latest_completed;
2124 : : TransactionId oldestxid;
2125 : : int mypgxactoff;
2126 : : TransactionId myxid;
2127 : : uint64 curXactCompletionCount;
2128 : :
2734 andres@anarazel.de 2129 : 2349712 : TransactionId replication_slot_xmin = InvalidTransactionId;
2130 : 2349712 : TransactionId replication_slot_catalog_xmin = InvalidTransactionId;
2131 : :
7656 tgl@sss.pgh.pa.us 2132 [ - + ]: 2349712 : Assert(snapshot != NULL);
2133 : :
2134 : : /*
2135 : : * Allocating space for maxProcs xids is usually overkill; numProcs would
2136 : : * be sufficient. But it seems better to do the malloc while not holding
2137 : : * the lock, so we can't look at numProcs. Likewise, we allocate much
2138 : : * more subxip storage than is probably needed.
2139 : : *
2140 : : * This does open a possibility for avoiding repeated malloc/free: since
2141 : : * maxProcs does not change at runtime, we can simply reuse the previous
2142 : : * xip arrays if any. (This relies on the fact that all callers pass
2143 : : * static SnapshotData structs.)
2144 : : */
2145 [ + + ]: 2349712 : if (snapshot->xip == NULL)
2146 : : {
2147 : : /*
2148 : : * First call for this snapshot. Snapshot is same size whether or not
2149 : : * we are in recovery, see later comments.
2150 : : */
2151 : 35185 : snapshot->xip = (TransactionId *)
5309 2152 : 35185 : malloc(GetMaxSnapshotXidCount() * sizeof(TransactionId));
7656 2153 [ - + ]: 35185 : if (snapshot->xip == NULL)
7656 tgl@sss.pgh.pa.us 2154 [ # # ]:UBC 0 : ereport(ERROR,
2155 : : (errcode(ERRCODE_OUT_OF_MEMORY),
2156 : : errmsg("out of memory")));
7184 tgl@sss.pgh.pa.us 2157 [ - + ]:CBC 35185 : Assert(snapshot->subxip == NULL);
2158 : 35185 : snapshot->subxip = (TransactionId *)
5309 2159 : 35185 : malloc(GetMaxSnapshotSubxidCount() * sizeof(TransactionId));
7184 2160 [ - + ]: 35185 : if (snapshot->subxip == NULL)
7184 tgl@sss.pgh.pa.us 2161 [ # # ]:UBC 0 : ereport(ERROR,
2162 : : (errcode(ERRCODE_OUT_OF_MEMORY),
2163 : : errmsg("out of memory")));
2164 : : }
2165 : :
2166 : : /*
2167 : : * It is sufficient to get shared lock on ProcArrayLock, even if we are
2168 : : * going to set MyProc->xmin.
2169 : : */
7184 tgl@sss.pgh.pa.us 2170 :CBC 2349712 : LWLockAcquire(ProcArrayLock, LW_SHARED);
2171 : :
2087 andres@anarazel.de 2172 [ + + ]: 2349712 : if (GetSnapshotDataReuse(snapshot))
2173 : : {
2174 : 1839719 : LWLockRelease(ProcArrayLock);
2175 : 1839719 : return snapshot;
2176 : : }
2177 : :
879 heikki.linnakangas@i 2178 : 509993 : latest_completed = TransamVariables->latestCompletedXid;
2090 andres@anarazel.de 2179 : 509993 : mypgxactoff = MyProc->pgxactoff;
2180 : 509993 : myxid = other_xids[mypgxactoff];
2181 [ - + ]: 509993 : Assert(myxid == MyProc->xid);
2182 : :
879 heikki.linnakangas@i 2183 : 509993 : oldestxid = TransamVariables->oldestXid;
2184 : 509993 : curXactCompletionCount = TransamVariables->xactCompletionCount;
2185 : :
2186 : : /* xmax is always latestCompletedXid + 1 */
2093 andres@anarazel.de 2187 : 509993 : xmax = XidFromFullTransactionId(latest_completed);
6814 tgl@sss.pgh.pa.us 2188 [ - + ]: 509993 : TransactionIdAdvance(xmax);
2093 andres@anarazel.de 2189 [ - + ]: 509993 : Assert(TransactionIdIsNormal(xmax));
2190 : :
2191 : : /* initialize xmin calculation with xmax */
2092 2192 : 509993 : xmin = xmax;
2193 : :
2194 : : /* take own xid into account, saves a check inside the loop */
2090 2195 [ + + + - : 509993 : if (TransactionIdIsNormal(myxid) && NormalTransactionIdPrecedes(myxid, xmin))
- + + + ]
2196 : 45916 : xmin = myxid;
2197 : :
5860 simon@2ndQuadrant.co 2198 : 509993 : snapshot->takenDuringRecovery = RecoveryInProgress();
2199 : :
5861 2200 [ + + ]: 509993 : if (!snapshot->takenDuringRecovery)
2201 : : {
1692 fujii@postgresql.org 2202 : 508444 : int numProcs = arrayP->numProcs;
2090 andres@anarazel.de 2203 : 508444 : TransactionId *xip = snapshot->xip;
5077 bruce@momjian.us 2204 : 508444 : int *pgprocnos = arrayP->pgprocnos;
2090 andres@anarazel.de 2205 : 508444 : XidCacheStatus *subxidStates = ProcGlobal->subxidStates;
1996 alvherre@alvh.no-ip. 2206 : 508444 : uint8 *allStatusFlags = ProcGlobal->statusFlags;
2207 : :
2208 : : /*
2209 : : * First collect set of pgxactoff/xids that need to be included in the
2210 : : * snapshot.
2211 : : */
1692 fujii@postgresql.org 2212 [ + + ]: 4745374 : for (int pgxactoff = 0; pgxactoff < numProcs; pgxactoff++)
2213 : : {
2214 : : /* Fetch xid just once - see GetNewTransactionId */
2090 andres@anarazel.de 2215 : 4236930 : TransactionId xid = UINT32_ACCESS_ONCE(other_xids[pgxactoff]);
2216 : : uint8 statusFlags;
2217 : :
2218 [ - + ]: 4236930 : Assert(allProcs[arrayP->pgprocnos[pgxactoff]].pgxactoff == pgxactoff);
2219 : :
2220 : : /*
2221 : : * If the transaction has no XID assigned, we can skip it; it
2222 : : * won't have sub-XIDs either.
2223 : : */
2224 [ + + ]: 4236930 : if (likely(xid == InvalidTransactionId))
6817 tgl@sss.pgh.pa.us 2225 : 3075574 : continue;
2226 : :
2227 : : /*
2228 : : * We don't include our own XIDs (if any) in the snapshot. It
2229 : : * needs to be included in the xmin computation, but we did so
2230 : : * outside the loop.
2231 : : */
2090 andres@anarazel.de 2232 [ + + ]: 1161356 : if (pgxactoff == mypgxactoff)
2233 : 80120 : continue;
2234 : :
2235 : : /*
2236 : : * The only way we are able to get here with a non-normal xid is
2237 : : * during bootstrap - with this backend using
2238 : : * BootstrapTransactionId. But the above test should filter that
2239 : : * out.
2240 : : */
2241 [ - + ]: 1081236 : Assert(TransactionIdIsNormal(xid));
2242 : :
2243 : : /*
2244 : : * If the XID is >= xmax, we can skip it; such transactions will
2245 : : * be treated as running anyway (and any sub-XIDs will also be >=
2246 : : * xmax).
2247 : : */
2248 [ + - - + : 1081236 : if (!NormalTransactionIdPrecedes(xid, xmax))
+ + ]
5077 bruce@momjian.us 2249 : 261768 : continue;
2250 : :
2251 : : /*
2252 : : * Skip over backends doing logical decoding which manages xmin
2253 : : * separately (check below) and ones running LAZY VACUUM.
2254 : : */
1996 alvherre@alvh.no-ip. 2255 : 819468 : statusFlags = allStatusFlags[pgxactoff];
2256 [ + + ]: 819468 : if (statusFlags & (PROC_IN_LOGICAL_DECODING | PROC_IN_VACUUM))
2090 andres@anarazel.de 2257 : 22 : continue;
2258 : :
5254 rhaas@postgresql.org 2259 [ + - - + : 819446 : if (NormalTransactionIdPrecedes(xid, xmin))
+ + ]
2260 : 416168 : xmin = xid;
2261 : :
2262 : : /* Add XID to snapshot. */
2090 andres@anarazel.de 2263 : 819446 : xip[count++] = xid;
2264 : :
2265 : : /*
2266 : : * Save subtransaction XIDs if possible (if we've already
2267 : : * overflowed, there's no point). Note that the subxact XIDs must
2268 : : * be later than their parent, so no need to check them against
2269 : : * xmin. We could filter against xmax, but it seems better not to
2270 : : * do that much work while holding the ProcArrayLock.
2271 : : *
2272 : : * The other backend can add more subxids concurrently, but cannot
2273 : : * remove any. Hence it's important to fetch nxids just once.
2274 : : * Should be safe to use memcpy, though. (We needn't worry about
2275 : : * missing any xids added concurrently, because they must postdate
2276 : : * xmax.)
2277 : : *
2278 : : * Again, our own XIDs are not included in the snapshot.
2279 : : */
5254 rhaas@postgresql.org 2280 [ + + ]: 819446 : if (!suboverflowed)
2281 : : {
2282 : :
2090 andres@anarazel.de 2283 [ + + ]: 819442 : if (subxidStates[pgxactoff].overflowed)
5861 simon@2ndQuadrant.co 2284 : 20 : suboverflowed = true;
2285 : : else
2286 : : {
2090 andres@anarazel.de 2287 : 819422 : int nsubxids = subxidStates[pgxactoff].count;
2288 : :
2289 [ + + ]: 819422 : if (nsubxids > 0)
2290 : : {
2291 : 8151 : int pgprocno = pgprocnos[pgxactoff];
2734 2292 : 8151 : PGPROC *proc = &allProcs[pgprocno];
2293 : :
2294 : 8151 : pg_read_barrier(); /* pairs with GetNewTransactionId */
2295 : :
5861 simon@2ndQuadrant.co 2296 : 8151 : memcpy(snapshot->subxip + subcount,
1183 peter@eisentraut.org 2297 : 8151 : proc->subxids.xids,
2298 : : nsubxids * sizeof(TransactionId));
2090 andres@anarazel.de 2299 : 8151 : subcount += nsubxids;
2300 : : }
2301 : : }
2302 : : }
2303 : : }
2304 : : }
2305 : : else
2306 : : {
2307 : : /*
2308 : : * We're in hot standby, so get XIDs from KnownAssignedXids.
2309 : : *
2310 : : * We store all xids directly into subxip[]. Here's why:
2311 : : *
2312 : : * In recovery we don't know which xids are top-level and which are
2313 : : * subxacts, a design choice that greatly simplifies xid processing.
2314 : : *
2315 : : * It seems like we would want to try to put xids into xip[] only, but
2316 : : * that is fairly small. We would either need to make that bigger or
2317 : : * to increase the rate at which we WAL-log xid assignment; neither is
2318 : : * an appealing choice.
2319 : : *
2320 : : * We could try to store xids into xip[] first and then into subxip[]
2321 : : * if there are too many xids. That only works if the snapshot doesn't
2322 : : * overflow because we do not search subxip[] in that case. A simpler
2323 : : * way is to just store all xids in the subxip array because this is
2324 : : * by far the bigger array. We just leave the xip array empty.
2325 : : *
2326 : : * Either way we need to change the way XidInMVCCSnapshot() works
2327 : : * depending upon when the snapshot was taken, or change normal
2328 : : * snapshot processing so it matches.
2329 : : *
2330 : : * Note: It is possible for recovery to end before we finish taking
2331 : : * the snapshot, and for newly assigned transaction ids to be added to
2332 : : * the ProcArray. xmax cannot change while we hold ProcArrayLock, so
2333 : : * those newly added transaction ids would be filtered away, so we
2334 : : * need not be concerned about them.
2335 : : */
5851 tgl@sss.pgh.pa.us 2336 : 1549 : subcount = KnownAssignedXidsGetAndSetXmin(snapshot->subxip, &xmin,
2337 : : xmax);
2338 : :
2339 [ + + ]: 1549 : if (TransactionIdPrecedesOrEquals(xmin, procArray->lastOverflowedXid))
5981 simon@2ndQuadrant.co 2340 : 8 : suboverflowed = true;
2341 : : }
2342 : :
2343 : :
2344 : : /*
2345 : : * Fetch into local variable while ProcArrayLock is held - the
2346 : : * LWLockRelease below is a barrier, ensuring this happens inside the
2347 : : * lock.
2348 : : */
4477 rhaas@postgresql.org 2349 : 509993 : replication_slot_xmin = procArray->replication_slot_xmin;
4446 2350 : 509993 : replication_slot_catalog_xmin = procArray->replication_slot_catalog_xmin;
2351 : :
2091 andres@anarazel.de 2352 [ + + ]: 509993 : if (!TransactionIdIsValid(MyProc->xmin))
2353 : 267268 : MyProc->xmin = TransactionXmin = xmin;
2354 : :
7656 tgl@sss.pgh.pa.us 2355 : 509993 : LWLockRelease(ProcArrayLock);
2356 : :
2357 : : /* maintain state for GlobalVis* */
2358 : : {
2359 : : TransactionId def_vis_xid;
2360 : : TransactionId def_vis_xid_data;
2361 : : FullTransactionId def_vis_fxid;
2362 : : FullTransactionId def_vis_fxid_data;
2363 : : FullTransactionId oldestfxid;
2364 : :
2365 : : /*
2366 : : * Converting oldestXid is only safe when xid horizon cannot advance,
2367 : : * i.e. holding locks. While we don't hold the lock anymore, all the
2368 : : * necessary data has been gathered with lock held.
2369 : : */
2092 andres@anarazel.de 2370 : 509993 : oldestfxid = FullXidRelativeTo(latest_completed, oldestxid);
2371 : :
2372 : : /* Check whether there's a replication slot requiring an older xmin. */
2373 : : def_vis_xid_data =
1107 2374 : 509993 : TransactionIdOlder(xmin, replication_slot_xmin);
2375 : :
2376 : : /*
2377 : : * Rows in non-shared, non-catalog tables possibly could be vacuumed
2378 : : * if older than this xid.
2379 : : */
2092 2380 : 509993 : def_vis_xid = def_vis_xid_data;
2381 : :
2382 : : /*
2383 : : * Check whether there's a replication slot requiring an older catalog
2384 : : * xmin.
2385 : : */
2386 : : def_vis_xid =
2387 : 509993 : TransactionIdOlder(replication_slot_catalog_xmin, def_vis_xid);
2388 : :
2389 : 509993 : def_vis_fxid = FullXidRelativeTo(latest_completed, def_vis_xid);
2390 : 509993 : def_vis_fxid_data = FullXidRelativeTo(latest_completed, def_vis_xid_data);
2391 : :
2392 : : /*
2393 : : * Check if we can increase upper bound. As a previous
2394 : : * GlobalVisUpdate() might have computed more aggressive values, don't
2395 : : * overwrite them if so.
2396 : : */
2397 : : GlobalVisSharedRels.definitely_needed =
2398 : 509993 : FullTransactionIdNewer(def_vis_fxid,
2399 : : GlobalVisSharedRels.definitely_needed);
2400 : : GlobalVisCatalogRels.definitely_needed =
2401 : 509993 : FullTransactionIdNewer(def_vis_fxid,
2402 : : GlobalVisCatalogRels.definitely_needed);
2403 : : GlobalVisDataRels.definitely_needed =
2404 : 509993 : FullTransactionIdNewer(def_vis_fxid_data,
2405 : : GlobalVisDataRels.definitely_needed);
2406 : : /* See temp_oldest_nonremovable computation in ComputeXidHorizons() */
2015 2407 [ + + ]: 509993 : if (TransactionIdIsNormal(myxid))
2408 : : GlobalVisTempRels.definitely_needed =
2409 : 80006 : FullXidRelativeTo(latest_completed, myxid);
2410 : : else
2411 : : {
2412 : 429987 : GlobalVisTempRels.definitely_needed = latest_completed;
2413 : 429987 : FullTransactionIdAdvance(&GlobalVisTempRels.definitely_needed);
2414 : : }
2415 : :
2416 : : /*
2417 : : * Check if we know that we can initialize or increase the lower
2418 : : * bound. Currently the only cheap way to do so is to use
2419 : : * TransamVariables->oldestXid as input.
2420 : : *
2421 : : * We should definitely be able to do better. We could e.g. put a
2422 : : * global lower bound value into TransamVariables.
2423 : : */
2424 : : GlobalVisSharedRels.maybe_needed =
2092 2425 : 509993 : FullTransactionIdNewer(GlobalVisSharedRels.maybe_needed,
2426 : : oldestfxid);
2427 : : GlobalVisCatalogRels.maybe_needed =
2428 : 509993 : FullTransactionIdNewer(GlobalVisCatalogRels.maybe_needed,
2429 : : oldestfxid);
2430 : : GlobalVisDataRels.maybe_needed =
2431 : 509993 : FullTransactionIdNewer(GlobalVisDataRels.maybe_needed,
2432 : : oldestfxid);
2433 : : /* accurate value known */
2015 2434 : 509993 : GlobalVisTempRels.maybe_needed = GlobalVisTempRels.definitely_needed;
2435 : : }
2436 : :
7656 tgl@sss.pgh.pa.us 2437 : 509993 : RecentXmin = xmin;
2090 andres@anarazel.de 2438 [ - + ]: 509993 : Assert(TransactionIdPrecedesOrEquals(TransactionXmin, RecentXmin));
2439 : :
7656 tgl@sss.pgh.pa.us 2440 : 509993 : snapshot->xmin = xmin;
2441 : 509993 : snapshot->xmax = xmax;
2442 : 509993 : snapshot->xcnt = count;
7184 2443 : 509993 : snapshot->subxcnt = subcount;
5981 simon@2ndQuadrant.co 2444 : 509993 : snapshot->suboverflowed = suboverflowed;
2087 andres@anarazel.de 2445 : 509993 : snapshot->snapXactCompletionCount = curXactCompletionCount;
2446 : :
6731 tgl@sss.pgh.pa.us 2447 : 509993 : snapshot->curcid = GetCurrentCommandId(false);
2448 : :
2449 : : /*
2450 : : * This is a new snapshot, so set both refcounts are zero, and mark it as
2451 : : * not copied in persistent memory.
2452 : : */
6567 alvherre@alvh.no-ip. 2453 : 509993 : snapshot->active_count = 0;
2454 : 509993 : snapshot->regd_count = 0;
2455 : 509993 : snapshot->copied = false;
2456 : :
7656 tgl@sss.pgh.pa.us 2457 : 509993 : return snapshot;
2458 : : }
2459 : :
2460 : : /*
2461 : : * ProcArrayInstallImportedXmin -- install imported xmin into MyProc->xmin
2462 : : *
2463 : : * This is called when installing a snapshot imported from another
2464 : : * transaction. To ensure that OldestXmin doesn't go backwards, we must
2465 : : * check that the source transaction is still running, and we'd better do
2466 : : * that atomically with installing the new xmin.
2467 : : *
2468 : : * Returns true if successful, false if source xact is no longer running.
2469 : : */
2470 : : bool
3247 andres@anarazel.de 2471 : 18 : ProcArrayInstallImportedXmin(TransactionId xmin,
2472 : : VirtualTransactionId *sourcevxid)
2473 : : {
5309 tgl@sss.pgh.pa.us 2474 : 18 : bool result = false;
2475 : 18 : ProcArrayStruct *arrayP = procArray;
2476 : : int index;
2477 : :
2478 [ - + ]: 18 : Assert(TransactionIdIsNormal(xmin));
3247 andres@anarazel.de 2479 [ - + ]: 18 : if (!sourcevxid)
5309 tgl@sss.pgh.pa.us 2480 :UBC 0 : return false;
2481 : :
2482 : : /* Get lock so source xact can't end while we're doing this */
5309 tgl@sss.pgh.pa.us 2483 :CBC 18 : LWLockAcquire(ProcArrayLock, LW_SHARED);
2484 : :
2485 : : /*
2486 : : * Find the PGPROC entry of the source transaction. (This could use
2487 : : * GetPGProcByNumber(), unless it's a prepared xact. But this isn't
2488 : : * performance critical.)
2489 : : */
2490 [ + - ]: 18 : for (index = 0; index < arrayP->numProcs; index++)
2491 : : {
5077 bruce@momjian.us 2492 : 18 : int pgprocno = arrayP->pgprocnos[index];
2734 andres@anarazel.de 2493 : 18 : PGPROC *proc = &allProcs[pgprocno];
1996 alvherre@alvh.no-ip. 2494 : 18 : int statusFlags = ProcGlobal->statusFlags[index];
2495 : : TransactionId xid;
2496 : :
2497 : : /* Ignore procs running LAZY VACUUM */
2498 [ - + ]: 18 : if (statusFlags & PROC_IN_VACUUM)
5309 tgl@sss.pgh.pa.us 2499 :UBC 0 : continue;
2500 : :
2501 : : /* We are only interested in the specific virtual transaction. */
793 heikki.linnakangas@i 2502 [ - + ]:CBC 18 : if (proc->vxid.procNumber != sourcevxid->procNumber)
3247 andres@anarazel.de 2503 :LBC (2) : continue;
793 heikki.linnakangas@i 2504 [ - + ]:CBC 18 : if (proc->vxid.lxid != sourcevxid->localTransactionId)
5309 tgl@sss.pgh.pa.us 2505 :UBC 0 : continue;
2506 : :
2507 : : /*
2508 : : * We check the transaction's database ID for paranoia's sake: if it's
2509 : : * in another DB then its xmin does not cover us. Caller should have
2510 : : * detected this already, so we just treat any funny cases as
2511 : : * "transaction not found".
2512 : : */
5309 tgl@sss.pgh.pa.us 2513 [ - + ]:CBC 18 : if (proc->databaseId != MyDatabaseId)
5309 tgl@sss.pgh.pa.us 2514 :UBC 0 : continue;
2515 : :
2516 : : /*
2517 : : * Likewise, let's just make real sure its xmin does cover us.
2518 : : */
2091 andres@anarazel.de 2519 :CBC 18 : xid = UINT32_ACCESS_ONCE(proc->xmin);
5309 tgl@sss.pgh.pa.us 2520 [ + - ]: 18 : if (!TransactionIdIsNormal(xid) ||
2521 [ - + ]: 18 : !TransactionIdPrecedesOrEquals(xid, xmin))
5309 tgl@sss.pgh.pa.us 2522 :UBC 0 : continue;
2523 : :
2524 : : /*
2525 : : * We're good. Install the new xmin. As in GetSnapshotData, set
2526 : : * TransactionXmin too. (Note that because snapmgr.c called
2527 : : * GetSnapshotData first, we'll be overwriting a valid xmin here, so
2528 : : * we don't check that.)
2529 : : */
2091 andres@anarazel.de 2530 :CBC 18 : MyProc->xmin = TransactionXmin = xmin;
2531 : :
5309 tgl@sss.pgh.pa.us 2532 : 18 : result = true;
2533 : 18 : break;
2534 : : }
2535 : :
2536 : 18 : LWLockRelease(ProcArrayLock);
2537 : :
2538 : 18 : return result;
2539 : : }
2540 : :
2541 : : /*
2542 : : * ProcArrayInstallRestoredXmin -- install restored xmin into MyProc->xmin
2543 : : *
2544 : : * This is like ProcArrayInstallImportedXmin, but we have a pointer to the
2545 : : * PGPROC of the transaction from which we imported the snapshot, rather than
2546 : : * an XID.
2547 : : *
2548 : : * Note that this function also copies statusFlags from the source `proc` in
2549 : : * order to avoid the case where MyProc's xmin needs to be skipped for
2550 : : * computing xid horizon.
2551 : : *
2552 : : * Returns true if successful, false if source xact is no longer running.
2553 : : */
2554 : : bool
4023 rhaas@postgresql.org 2555 : 2220 : ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
2556 : : {
2557 : 2220 : bool result = false;
2558 : : TransactionId xid;
2559 : :
2560 [ - + ]: 2220 : Assert(TransactionIdIsNormal(xmin));
2561 [ - + ]: 2220 : Assert(proc != NULL);
2562 : :
2563 : : /*
2564 : : * Get an exclusive lock so that we can copy statusFlags from source proc.
2565 : : */
1628 akapila@postgresql.o 2566 : 2220 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
2567 : :
2568 : : /*
2569 : : * Be certain that the referenced PGPROC has an advertised xmin which is
2570 : : * no later than the one we're installing, so that the system-wide xmin
2571 : : * can't go backwards. Also, make sure it's running in the same database,
2572 : : * so that the per-database xmin cannot go backwards.
2573 : : */
2091 andres@anarazel.de 2574 : 2220 : xid = UINT32_ACCESS_ONCE(proc->xmin);
4023 rhaas@postgresql.org 2575 [ + - + - ]: 2220 : if (proc->databaseId == MyDatabaseId &&
2576 [ + - ]: 2220 : TransactionIdIsNormal(xid) &&
2577 : 2220 : TransactionIdPrecedesOrEquals(xid, xmin))
2578 : : {
2579 : : /*
2580 : : * Install xmin and propagate the statusFlags that affect how the
2581 : : * value is interpreted by vacuum.
2582 : : */
2091 andres@anarazel.de 2583 : 2220 : MyProc->xmin = TransactionXmin = xmin;
1447 alvherre@alvh.no-ip. 2584 : 2220 : MyProc->statusFlags = (MyProc->statusFlags & ~PROC_XMIN_FLAGS) |
2585 : 2220 : (proc->statusFlags & PROC_XMIN_FLAGS);
2586 : 2220 : ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags;
2587 : :
4023 rhaas@postgresql.org 2588 : 2220 : result = true;
2589 : : }
2590 : :
2591 : 2220 : LWLockRelease(ProcArrayLock);
2592 : :
2593 : 2220 : return result;
2594 : : }
2595 : :
2596 : : /*
2597 : : * GetRunningTransactionData -- returns information about running transactions.
2598 : : *
2599 : : * Similar to GetSnapshotData but returns more information. We include
2600 : : * all PGPROCs with an assigned TransactionId, even VACUUM processes and
2601 : : * prepared transactions.
2602 : : *
2603 : : * We acquire XidGenLock and ProcArrayLock, but the caller is responsible for
2604 : : * releasing them. Acquiring XidGenLock ensures that no new XIDs enter the proc
2605 : : * array until the caller has WAL-logged this snapshot, and releases the
2606 : : * lock. Acquiring ProcArrayLock ensures that no transactions commit until the
2607 : : * lock is released.
2608 : : *
2609 : : * The returned data structure is statically allocated; caller should not
2610 : : * modify it, and must not assume it is valid past the next call.
2611 : : *
2612 : : * This is never executed during recovery so there is no need to look at
2613 : : * KnownAssignedXids.
2614 : : *
2615 : : * Dummy PGPROCs from prepared transaction are included, meaning that this
2616 : : * may return entries with duplicated TransactionId values coming from
2617 : : * transaction finishing to prepare. Nothing is done about duplicated
2618 : : * entries here to not hold on ProcArrayLock more than necessary.
2619 : : *
2620 : : * We don't worry about updating other counters, we want to keep this as
2621 : : * simple as possible and leave GetSnapshotData() as the primary code for
2622 : : * that bookkeeping.
2623 : : *
2624 : : * Note that if any transaction has overflowed its cached subtransactions
2625 : : * then there is no real need include any subtransactions.
2626 : : *
2627 : : * If 'dbid' is valid, only gather transactions running in that database.
2628 : : */
2629 : : RunningTransactions
28 alvherre@kurilemu.de 2630 :GNC 1544 : GetRunningTransactionData(Oid dbid)
2631 : : {
2632 : : /* result workspace */
2633 : : static RunningTransactionsData CurrentRunningXactsData;
2634 : :
5981 simon@2ndQuadrant.co 2635 :CBC 1544 : ProcArrayStruct *arrayP = procArray;
2090 andres@anarazel.de 2636 : 1544 : TransactionId *other_xids = ProcGlobal->xids;
5851 tgl@sss.pgh.pa.us 2637 : 1544 : RunningTransactions CurrentRunningXacts = &CurrentRunningXactsData;
2638 : : TransactionId latestCompletedXid;
2639 : : TransactionId oldestRunningXid;
2640 : : TransactionId oldestDatabaseRunningXid;
2641 : : TransactionId *xids;
2642 : : int index;
2643 : : int count;
2644 : : int subcount;
2645 : : bool suboverflowed;
2646 : :
5981 simon@2ndQuadrant.co 2647 [ - + ]: 1544 : Assert(!RecoveryInProgress());
2648 : :
2649 : : /*
2650 : : * Allocating space for maxProcs xids is usually overkill; numProcs would
2651 : : * be sufficient. But it seems better to do the malloc while not holding
2652 : : * the lock, so we can't look at numProcs. Likewise, we allocate much
2653 : : * more subxip storage than is probably needed.
2654 : : *
2655 : : * Should only be allocated in bgwriter, since only ever executed during
2656 : : * checkpoints.
2657 : : */
2658 [ + + ]: 1544 : if (CurrentRunningXacts->xids == NULL)
2659 : : {
2660 : : /*
2661 : : * First call
2662 : : */
2663 : 606 : CurrentRunningXacts->xids = (TransactionId *)
2664 : 606 : malloc(TOTAL_MAX_CACHED_SUBXIDS * sizeof(TransactionId));
2665 [ - + ]: 606 : if (CurrentRunningXacts->xids == NULL)
5981 simon@2ndQuadrant.co 2666 [ # # ]:UBC 0 : ereport(ERROR,
2667 : : (errcode(ERRCODE_OUT_OF_MEMORY),
2668 : : errmsg("out of memory")));
2669 : : }
2670 : :
5981 simon@2ndQuadrant.co 2671 :CBC 1544 : xids = CurrentRunningXacts->xids;
2672 : :
2673 : 1544 : count = subcount = 0;
2674 : 1544 : suboverflowed = false;
2675 : :
2676 : : /*
2677 : : * Ensure that no xids enter or leave the procarray while we obtain
2678 : : * snapshot.
2679 : : */
2680 : 1544 : LWLockAcquire(ProcArrayLock, LW_SHARED);
2681 : 1544 : LWLockAcquire(XidGenLock, LW_SHARED);
2682 : :
2093 andres@anarazel.de 2683 : 1544 : latestCompletedXid =
879 heikki.linnakangas@i 2684 : 1544 : XidFromFullTransactionId(TransamVariables->latestCompletedXid);
782 akorotkov@postgresql 2685 : 1544 : oldestDatabaseRunningXid = oldestRunningXid =
879 heikki.linnakangas@i 2686 : 1544 : XidFromFullTransactionId(TransamVariables->nextXid);
2687 : :
2688 : : /*
2689 : : * Spin over procArray collecting all xids
2690 : : */
5981 simon@2ndQuadrant.co 2691 [ + + ]: 7564 : for (index = 0; index < arrayP->numProcs; index++)
2692 : : {
2693 : : TransactionId xid;
2694 : :
2695 : : /* Fetch xid just once - see GetNewTransactionId */
2090 andres@anarazel.de 2696 : 6020 : xid = UINT32_ACCESS_ONCE(other_xids[index]);
2697 : :
2698 : : /*
2699 : : * We don't need to store transactions that don't have a TransactionId
2700 : : * yet because they will not show as running on a standby server.
2701 : : */
5981 simon@2ndQuadrant.co 2702 [ + + ]: 6020 : if (!TransactionIdIsValid(xid))
2703 : 4971 : continue;
2704 : :
2705 : : /*
2706 : : * Filter by database OID if requested.
2707 : : */
28 alvherre@kurilemu.de 2708 [ - + ]:GNC 1049 : if (OidIsValid(dbid))
2709 : : {
28 alvherre@kurilemu.de 2710 :UNC 0 : int pgprocno = arrayP->pgprocnos[index];
2711 : 0 : PGPROC *proc = &allProcs[pgprocno];
2712 : :
2713 [ # # ]: 0 : if (proc->databaseId != dbid)
2714 : 0 : continue;
2715 : : }
2716 : :
2717 : : /*
2718 : : * Be careful not to exclude any xids before calculating the values of
2719 : : * oldestRunningXid and suboverflowed, since these are used to clean
2720 : : * up transaction information held on standbys.
2721 : : */
5981 simon@2ndQuadrant.co 2722 [ + + ]:CBC 1049 : if (TransactionIdPrecedes(xid, oldestRunningXid))
2723 : 665 : oldestRunningXid = xid;
2724 : :
2725 : : /*
2726 : : * Also, update the oldest running xid within the current database. As
2727 : : * fetching pgprocno and PGPROC could cause cache misses, we do cheap
2728 : : * TransactionId comparison first.
2729 : : */
670 akorotkov@postgresql 2730 [ + - ]: 1049 : if (TransactionIdPrecedes(xid, oldestDatabaseRunningXid))
2731 : : {
2732 : 1049 : int pgprocno = arrayP->pgprocnos[index];
2733 : 1049 : PGPROC *proc = &allProcs[pgprocno];
2734 : :
2735 [ + + ]: 1049 : if (proc->databaseId == MyDatabaseId)
2736 : 237 : oldestDatabaseRunningXid = xid;
2737 : : }
2738 : :
2090 andres@anarazel.de 2739 [ + + ]: 1049 : if (ProcGlobal->subxidStates[index].overflowed)
4902 simon@2ndQuadrant.co 2740 : 2 : suboverflowed = true;
2741 : :
2742 : : /*
2743 : : * If we wished to exclude xids this would be the right place for it.
2744 : : * Procs with the PROC_IN_VACUUM flag set don't usually assign xids,
2745 : : * but they do during truncation at the end when they get the lock and
2746 : : * truncate, so it is not much of a problem to include them if they
2747 : : * are seen and it is cleaner to include them.
2748 : : */
2749 : :
2882 2750 : 1049 : xids[count++] = xid;
2751 : : }
2752 : :
2753 : : /*
2754 : : * Spin over procArray collecting all subxids, but only if there hasn't
2755 : : * been a suboverflow.
2756 : : */
4902 2757 [ + + ]: 1544 : if (!suboverflowed)
2758 : : {
2090 andres@anarazel.de 2759 : 1542 : XidCacheStatus *other_subxidstates = ProcGlobal->subxidStates;
2760 : :
4902 simon@2ndQuadrant.co 2761 [ + + ]: 7558 : for (index = 0; index < arrayP->numProcs; index++)
2762 : : {
2763 : 6016 : int pgprocno = arrayP->pgprocnos[index];
2734 andres@anarazel.de 2764 : 6016 : PGPROC *proc = &allProcs[pgprocno];
2765 : : int nsubxids;
2766 : :
2767 : : /*
2768 : : * Filter by database OID if requested.
2769 : : */
28 alvherre@kurilemu.de 2770 [ + + + + ]:GNC 6016 : if (OidIsValid(dbid) && proc->databaseId != dbid)
2771 : 6 : continue;
2772 : :
2773 : : /*
2774 : : * Save subtransaction XIDs. Other backends can't add or remove
2775 : : * entries while we're holding XidGenLock.
2776 : : */
2090 andres@anarazel.de 2777 :CBC 6010 : nsubxids = other_subxidstates[index].count;
2778 [ + + ]: 6010 : if (nsubxids > 0)
2779 : : {
2780 : : /* barrier not really required, as XidGenLock is held, but ... */
2734 2781 : 17 : pg_read_barrier(); /* pairs with GetNewTransactionId */
2782 : :
1183 peter@eisentraut.org 2783 : 17 : memcpy(&xids[count], proc->subxids.xids,
2784 : : nsubxids * sizeof(TransactionId));
2090 andres@anarazel.de 2785 : 17 : count += nsubxids;
2786 : 17 : subcount += nsubxids;
2787 : :
2788 : : /*
2789 : : * Top-level XID of a transaction is always less than any of
2790 : : * its subxids, so we don't need to check if any of the
2791 : : * subxids are smaller than oldestRunningXid
2792 : : */
2793 : : }
2794 : : }
2795 : : }
2796 : :
2797 : : /*
2798 : : * It's important *not* to include the limits set by slots here because
2799 : : * snapbuild.c uses oldestRunningXid to manage its xmin horizon. If those
2800 : : * were to be included here the initial value could never increase because
2801 : : * of a circular dependency where slots only increase their limits when
2802 : : * running xacts increases oldestRunningXid and running xacts only
2803 : : * increases if slots do.
2804 : : */
2805 : :
28 alvherre@kurilemu.de 2806 :GNC 1544 : CurrentRunningXacts->dbid = dbid;
4902 simon@2ndQuadrant.co 2807 :CBC 1544 : CurrentRunningXacts->xcnt = count - subcount;
2808 : 1544 : CurrentRunningXacts->subxcnt = subcount;
677 heikki.linnakangas@i 2809 [ + + ]: 1544 : CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
879 2810 : 1544 : CurrentRunningXacts->nextXid = XidFromFullTransactionId(TransamVariables->nextXid);
5981 simon@2ndQuadrant.co 2811 : 1544 : CurrentRunningXacts->oldestRunningXid = oldestRunningXid;
782 akorotkov@postgresql 2812 : 1544 : CurrentRunningXacts->oldestDatabaseRunningXid = oldestDatabaseRunningXid;
5836 simon@2ndQuadrant.co 2813 : 1544 : CurrentRunningXacts->latestCompletedXid = latestCompletedXid;
2814 : :
5835 2815 [ - + ]: 1544 : Assert(TransactionIdIsValid(CurrentRunningXacts->nextXid));
2816 [ - + ]: 1544 : Assert(TransactionIdIsValid(CurrentRunningXacts->oldestRunningXid));
2817 [ - + ]: 1544 : Assert(TransactionIdIsNormal(CurrentRunningXacts->latestCompletedXid));
2818 : :
2819 : : /* We don't release the locks here, the caller is responsible for that */
2820 : :
5981 2821 : 1544 : return CurrentRunningXacts;
2822 : : }
2823 : :
2824 : : /*
2825 : : * GetOldestActiveTransactionId()
2826 : : *
2827 : : * Similar to GetSnapshotData but returns just oldestActiveXid. We include
2828 : : * all PGPROCs with an assigned TransactionId, even VACUUM processes.
2829 : : *
2830 : : * If allDbs is true, we look at all databases, though there is no need to
2831 : : * include WALSender since this has no effect on hot standby conflicts. If
2832 : : * allDbs is false, skip processes attached to other databases.
2833 : : *
2834 : : * This is never executed during recovery so there is no need to look at
2835 : : * KnownAssignedXids.
2836 : : *
2837 : : * We don't worry about updating other counters, we want to keep this as
2838 : : * simple as possible and leave GetSnapshotData() as the primary code for
2839 : : * that bookkeeping.
2840 : : *
2841 : : * inCommitOnly indicates getting the oldestActiveXid among the transactions
2842 : : * in the commit critical section.
2843 : : */
2844 : : TransactionId
286 akapila@postgresql.o 2845 :GNC 2859 : GetOldestActiveTransactionId(bool inCommitOnly, bool allDbs)
2846 : : {
5298 simon@2ndQuadrant.co 2847 :CBC 2859 : ProcArrayStruct *arrayP = procArray;
2090 andres@anarazel.de 2848 : 2859 : TransactionId *other_xids = ProcGlobal->xids;
2849 : : TransactionId oldestRunningXid;
2850 : : int index;
2851 : :
5298 simon@2ndQuadrant.co 2852 [ - + ]: 2859 : Assert(!RecoveryInProgress());
2853 : :
2854 : : /*
2855 : : * Read nextXid, as the upper bound of what's still active.
2856 : : *
2857 : : * Reading a TransactionId is atomic, but we must grab the lock to make
2858 : : * sure that all XIDs < nextXid are already present in the proc array (or
2859 : : * have already completed), when we spin over it.
2860 : : */
3218 heikki.linnakangas@i 2861 : 2859 : LWLockAcquire(XidGenLock, LW_SHARED);
879 2862 : 2859 : oldestRunningXid = XidFromFullTransactionId(TransamVariables->nextXid);
3218 2863 : 2859 : LWLockRelease(XidGenLock);
2864 : :
2865 : : /*
2866 : : * Spin over procArray collecting all xids and subxids.
2867 : : */
2868 : 2859 : LWLockAcquire(ProcArrayLock, LW_SHARED);
5298 simon@2ndQuadrant.co 2869 [ + + ]: 15902 : for (index = 0; index < arrayP->numProcs; index++)
2870 : : {
2871 : : TransactionId xid;
286 akapila@postgresql.o 2872 :GNC 13043 : int pgprocno = arrayP->pgprocnos[index];
2873 : 13043 : PGPROC *proc = &allProcs[pgprocno];
2874 : :
2875 : : /* Fetch xid just once - see GetNewTransactionId */
2090 andres@anarazel.de 2876 :CBC 13043 : xid = UINT32_ACCESS_ONCE(other_xids[index]);
2877 : :
5298 simon@2ndQuadrant.co 2878 [ + + ]: 13043 : if (!TransactionIdIsNormal(xid))
2879 : 10314 : continue;
2880 : :
286 akapila@postgresql.o 2881 [ + + ]:GNC 2729 : if (inCommitOnly &&
2882 [ + + ]: 1784 : (proc->delayChkptFlags & DELAY_CHKPT_IN_COMMIT) == 0)
2883 : 1765 : continue;
2884 : :
2885 [ + + - + ]: 964 : if (!allDbs && proc->databaseId != MyDatabaseId)
286 akapila@postgresql.o 2886 :UNC 0 : continue;
2887 : :
5298 simon@2ndQuadrant.co 2888 [ + + ]:CBC 964 : if (TransactionIdPrecedes(xid, oldestRunningXid))
2889 : 621 : oldestRunningXid = xid;
2890 : :
2891 : : /*
2892 : : * Top-level XID of a transaction is always less than any of its
2893 : : * subxids, so we don't need to check if any of the subxids are
2894 : : * smaller than oldestRunningXid
2895 : : */
2896 : : }
2897 : 2859 : LWLockRelease(ProcArrayLock);
2898 : :
2899 : 2859 : return oldestRunningXid;
2900 : : }
2901 : :
2902 : : /*
2903 : : * GetOldestSafeDecodingTransactionId -- lowest xid not affected by vacuum
2904 : : *
2905 : : * Returns the oldest xid that we can guarantee not to have been affected by
2906 : : * vacuum, i.e. no rows >= that xid have been vacuumed away unless the
2907 : : * transaction aborted. Note that the value can (and most of the time will) be
2908 : : * much more conservative than what really has been affected by vacuum, but we
2909 : : * currently don't have better data available.
2910 : : *
2911 : : * This is useful to initialize the cutoff xid after which a new changeset
2912 : : * extraction replication slot can start decoding changes.
2913 : : *
2914 : : * Must be called with ProcArrayLock held either shared or exclusively,
2915 : : * although most callers will want to use exclusive mode since it is expected
2916 : : * that the caller will immediately use the xid to peg the xmin horizon.
2917 : : */
2918 : : TransactionId
3299 andres@anarazel.de 2919 : 728 : GetOldestSafeDecodingTransactionId(bool catalogOnly)
2920 : : {
4446 rhaas@postgresql.org 2921 : 728 : ProcArrayStruct *arrayP = procArray;
2922 : : TransactionId oldestSafeXid;
2923 : : int index;
2924 : 728 : bool recovery_in_progress = RecoveryInProgress();
2925 : :
2926 [ - + ]: 728 : Assert(LWLockHeldByMe(ProcArrayLock));
2927 : :
2928 : : /*
2929 : : * Acquire XidGenLock, so no transactions can acquire an xid while we're
2930 : : * running. If no transaction with xid were running concurrently a new xid
2931 : : * could influence the RecentXmin et al.
2932 : : *
2933 : : * We initialize the computation to nextXid since that's guaranteed to be
2934 : : * a safe, albeit pessimal, value.
2935 : : */
2936 : 728 : LWLockAcquire(XidGenLock, LW_SHARED);
879 heikki.linnakangas@i 2937 : 728 : oldestSafeXid = XidFromFullTransactionId(TransamVariables->nextXid);
2938 : :
2939 : : /*
2940 : : * If there's already a slot pegging the xmin horizon, we can start with
2941 : : * that value, it's guaranteed to be safe since it's computed by this
2942 : : * routine initially and has been enforced since. We can always use the
2943 : : * slot's general xmin horizon, but the catalog horizon is only usable
2944 : : * when only catalog data is going to be looked at.
2945 : : */
3299 andres@anarazel.de 2946 [ + + + + ]: 959 : if (TransactionIdIsValid(procArray->replication_slot_xmin) &&
2947 : 231 : TransactionIdPrecedes(procArray->replication_slot_xmin,
2948 : : oldestSafeXid))
2949 : 12 : oldestSafeXid = procArray->replication_slot_xmin;
2950 : :
2951 [ + + ]: 728 : if (catalogOnly &&
2952 [ + + + + ]: 370 : TransactionIdIsValid(procArray->replication_slot_catalog_xmin) &&
4446 rhaas@postgresql.org 2953 : 78 : TransactionIdPrecedes(procArray->replication_slot_catalog_xmin,
2954 : : oldestSafeXid))
2955 : 30 : oldestSafeXid = procArray->replication_slot_catalog_xmin;
2956 : :
2957 : : /*
2958 : : * If we're not in recovery, we walk over the procarray and collect the
2959 : : * lowest xid. Since we're called with ProcArrayLock held and have
2960 : : * acquired XidGenLock, no entries can vanish concurrently, since
2961 : : * ProcGlobal->xids[i] is only set with XidGenLock held and only cleared
2962 : : * with ProcArrayLock held.
2963 : : *
2964 : : * In recovery we can't lower the safe value besides what we've computed
2965 : : * above, so we'll have to wait a bit longer there. We unfortunately can
2966 : : * *not* use KnownAssignedXidsGetOldestXmin() since the KnownAssignedXids
2967 : : * machinery can miss values and return an older value than is safe.
2968 : : */
2969 [ + + ]: 728 : if (!recovery_in_progress)
2970 : : {
2090 andres@anarazel.de 2971 : 693 : TransactionId *other_xids = ProcGlobal->xids;
2972 : :
2973 : : /*
2974 : : * Spin over procArray collecting min(ProcGlobal->xids[i])
2975 : : */
4446 rhaas@postgresql.org 2976 [ + + ]: 3568 : for (index = 0; index < arrayP->numProcs; index++)
2977 : : {
2978 : : TransactionId xid;
2979 : :
2980 : : /* Fetch xid just once - see GetNewTransactionId */
2090 andres@anarazel.de 2981 : 2875 : xid = UINT32_ACCESS_ONCE(other_xids[index]);
2982 : :
4446 rhaas@postgresql.org 2983 [ + + ]: 2875 : if (!TransactionIdIsNormal(xid))
2984 : 2866 : continue;
2985 : :
2986 [ + + ]: 9 : if (TransactionIdPrecedes(xid, oldestSafeXid))
2987 : 8 : oldestSafeXid = xid;
2988 : : }
2989 : : }
2990 : :
2991 : 728 : LWLockRelease(XidGenLock);
2992 : :
2993 : 728 : return oldestSafeXid;
2994 : : }
2995 : :
2996 : : /*
2997 : : * GetVirtualXIDsDelayingChkpt -- Get the VXIDs of transactions that are
2998 : : * delaying checkpoint because they have critical actions in progress.
2999 : : *
3000 : : * Constructs an array of VXIDs of transactions that are currently in commit
3001 : : * critical sections, as shown by having specified delayChkptFlags bits set
3002 : : * in their PGPROC.
3003 : : *
3004 : : * Returns a palloc'd array that should be freed by the caller.
3005 : : * *nvxids is the number of valid entries.
3006 : : *
3007 : : * Note that because backends set or clear delayChkptFlags without holding any
3008 : : * lock, the result is somewhat indeterminate, but we don't really care. Even
3009 : : * in a multiprocessor with delayed writes to shared memory, it should be
3010 : : * certain that setting of delayChkptFlags will propagate to shared memory
3011 : : * when the backend takes a lock, so we cannot fail to see a virtual xact as
3012 : : * delayChkptFlags if it's already inserted its commit record. Whether it
3013 : : * takes a little while for clearing of delayChkptFlags to propagate is
3014 : : * unimportant for correctness.
3015 : : */
3016 : : VirtualTransactionId *
1503 3017 : 3462 : GetVirtualXIDsDelayingChkpt(int *nvxids, int type)
3018 : : {
3019 : : VirtualTransactionId *vxids;
6972 tgl@sss.pgh.pa.us 3020 : 3462 : ProcArrayStruct *arrayP = procArray;
4901 simon@2ndQuadrant.co 3021 : 3462 : int count = 0;
3022 : : int index;
3023 : :
1503 rhaas@postgresql.org 3024 [ - + ]: 3462 : Assert(type != 0);
3025 : :
3026 : : /* allocate what's certainly enough result space */
146 michael@paquier.xyz 3027 :GNC 3462 : vxids = palloc_array(VirtualTransactionId, arrayP->maxProcs);
3028 : :
6972 tgl@sss.pgh.pa.us 3029 :CBC 3462 : LWLockAcquire(ProcArrayLock, LW_SHARED);
3030 : :
3031 [ + + ]: 11080 : for (index = 0; index < arrayP->numProcs; index++)
3032 : : {
4724 bruce@momjian.us 3033 : 7618 : int pgprocno = arrayP->pgprocnos[index];
2734 andres@anarazel.de 3034 : 7618 : PGPROC *proc = &allProcs[pgprocno];
3035 : :
1488 rhaas@postgresql.org 3036 [ + + ]: 7618 : if ((proc->delayChkptFlags & type) != 0)
3037 : : {
3038 : : VirtualTransactionId vxid;
3039 : :
4901 simon@2ndQuadrant.co 3040 : 29 : GET_VXID_FROM_PGPROC(vxid, *proc);
3041 [ + - ]: 29 : if (VirtualTransactionIdIsValid(vxid))
3042 : 29 : vxids[count++] = vxid;
3043 : : }
3044 : : }
3045 : :
6972 tgl@sss.pgh.pa.us 3046 : 3462 : LWLockRelease(ProcArrayLock);
3047 : :
4901 simon@2ndQuadrant.co 3048 : 3462 : *nvxids = count;
3049 : 3462 : return vxids;
3050 : : }
3051 : :
3052 : : /*
3053 : : * HaveVirtualXIDsDelayingChkpt -- Are any of the specified VXIDs delaying?
3054 : : *
3055 : : * This is used with the results of GetVirtualXIDsDelayingChkpt to see if any
3056 : : * of the specified VXIDs are still in critical sections of code.
3057 : : *
3058 : : * Note: this is O(N^2) in the number of vxacts that are/were delaying, but
3059 : : * those numbers should be small enough for it not to be a problem.
3060 : : */
3061 : : bool
1503 rhaas@postgresql.org 3062 : 26 : HaveVirtualXIDsDelayingChkpt(VirtualTransactionId *vxids, int nvxids, int type)
3063 : : {
6746 bruce@momjian.us 3064 : 26 : bool result = false;
6972 tgl@sss.pgh.pa.us 3065 : 26 : ProcArrayStruct *arrayP = procArray;
3066 : : int index;
3067 : :
1503 rhaas@postgresql.org 3068 [ - + ]: 26 : Assert(type != 0);
3069 : :
6972 tgl@sss.pgh.pa.us 3070 : 26 : LWLockAcquire(ProcArrayLock, LW_SHARED);
3071 : :
4710 noah@leadboat.com 3072 [ + + ]: 301 : for (index = 0; index < arrayP->numProcs; index++)
3073 : : {
3074 : 275 : int pgprocno = arrayP->pgprocnos[index];
2734 andres@anarazel.de 3075 : 275 : PGPROC *proc = &allProcs[pgprocno];
3076 : : VirtualTransactionId vxid;
3077 : :
4710 noah@leadboat.com 3078 : 275 : GET_VXID_FROM_PGPROC(vxid, *proc);
3079 : :
1488 rhaas@postgresql.org 3080 [ + + ]: 275 : if ((proc->delayChkptFlags & type) != 0 &&
1503 3081 [ + - ]: 2 : VirtualTransactionIdIsValid(vxid))
3082 : : {
3083 : : int i;
3084 : :
4710 noah@leadboat.com 3085 [ + + ]: 5 : for (i = 0; i < nvxids; i++)
3086 : : {
3087 [ + + - + ]: 3 : if (VirtualTransactionIdEquals(vxid, vxids[i]))
3088 : : {
6972 tgl@sss.pgh.pa.us 3089 :LBC (10) : result = true;
3090 : (10) : break;
3091 : : }
3092 : : }
4710 noah@leadboat.com 3093 [ - + ]:CBC 2 : if (result)
4710 noah@leadboat.com 3094 :LBC (10) : break;
3095 : : }
3096 : : }
3097 : :
6972 tgl@sss.pgh.pa.us 3098 :CBC 26 : LWLockRelease(ProcArrayLock);
3099 : :
3100 : 26 : return result;
3101 : : }
3102 : :
3103 : : /*
3104 : : * ProcNumberGetProc -- get a backend's PGPROC given its proc number
3105 : : *
3106 : : * The result may be out of date arbitrarily quickly, so the caller
3107 : : * must be careful about how this information is used. NULL is
3108 : : * returned if the backend is not active.
3109 : : */
3110 : : PGPROC *
793 heikki.linnakangas@i 3111 : 702 : ProcNumberGetProc(ProcNumber procNumber)
3112 : : {
3113 : : PGPROC *result;
3114 : :
3115 [ + + - + ]: 702 : if (procNumber < 0 || procNumber >= ProcGlobal->allProcCount)
3116 : 1 : return NULL;
3117 : 701 : result = GetPGProcByNumber(procNumber);
3118 : :
3119 [ + + ]: 701 : if (result->pid == 0)
3120 : 4 : return NULL;
3121 : :
3122 : 697 : return result;
3123 : : }
3124 : :
3125 : : /*
3126 : : * ProcNumberGetTransactionIds -- get a backend's transaction status
3127 : : *
3128 : : * Get the xid, xmin, nsubxid and overflow status of the backend. The
3129 : : * result may be out of date arbitrarily quickly, so the caller must be
3130 : : * careful about how this information is used.
3131 : : */
3132 : : void
3133 : 9999 : ProcNumberGetTransactionIds(ProcNumber procNumber, TransactionId *xid,
3134 : : TransactionId *xmin, int *nsubxid, bool *overflowed)
3135 : : {
3136 : : PGPROC *proc;
3137 : :
3138 : 9999 : *xid = InvalidTransactionId;
3139 : 9999 : *xmin = InvalidTransactionId;
3140 : 9999 : *nsubxid = 0;
3141 : 9999 : *overflowed = false;
3142 : :
3143 [ + - - + ]: 9999 : if (procNumber < 0 || procNumber >= ProcGlobal->allProcCount)
793 heikki.linnakangas@i 3144 :UBC 0 : return;
793 heikki.linnakangas@i 3145 :CBC 9999 : proc = GetPGProcByNumber(procNumber);
3146 : :
3147 : : /* Need to lock out additions/removals of backends */
3148 : 9999 : LWLockAcquire(ProcArrayLock, LW_SHARED);
3149 : :
3150 [ + - ]: 9999 : if (proc->pid != 0)
3151 : : {
3152 : 9999 : *xid = proc->xid;
3153 : 9999 : *xmin = proc->xmin;
3154 : 9999 : *nsubxid = proc->subxidStatus.count;
3155 : 9999 : *overflowed = proc->subxidStatus.overflowed;
3156 : : }
3157 : :
3158 : 9999 : LWLockRelease(ProcArrayLock);
3159 : : }
3160 : :
3161 : : /*
3162 : : * BackendPidGetProc -- get a backend's PGPROC given its PID
3163 : : *
3164 : : * Returns NULL if not found. Note that it is up to the caller to be
3165 : : * sure that the question remains meaningful for long enough for the
3166 : : * answer to be used ...
3167 : : */
3168 : : PGPROC *
7656 tgl@sss.pgh.pa.us 3169 : 10536 : BackendPidGetProc(int pid)
3170 : : {
3171 : : PGPROC *result;
3172 : :
3725 3173 [ + + ]: 10536 : if (pid == 0) /* never match dummy PGPROCs */
3174 : 4 : return NULL;
3175 : :
3176 : 10532 : LWLockAcquire(ProcArrayLock, LW_SHARED);
3177 : :
3178 : 10532 : result = BackendPidGetProcWithLock(pid);
3179 : :
3180 : 10532 : LWLockRelease(ProcArrayLock);
3181 : :
3182 : 10532 : return result;
3183 : : }
3184 : :
3185 : : /*
3186 : : * BackendPidGetProcWithLock -- get a backend's PGPROC given its PID
3187 : : *
3188 : : * Same as above, except caller must be holding ProcArrayLock. The found
3189 : : * entry, if any, can be assumed to be valid as long as the lock remains held.
3190 : : */
3191 : : PGPROC *
3192 : 12610 : BackendPidGetProcWithLock(int pid)
3193 : : {
7656 3194 : 12610 : PGPROC *result = NULL;
3195 : 12610 : ProcArrayStruct *arrayP = procArray;
3196 : : int index;
3197 : :
7627 3198 [ - + ]: 12610 : if (pid == 0) /* never match dummy PGPROCs */
7627 tgl@sss.pgh.pa.us 3199 :UBC 0 : return NULL;
3200 : :
7656 tgl@sss.pgh.pa.us 3201 [ + + ]:CBC 47706 : for (index = 0; index < arrayP->numProcs; index++)
3202 : : {
5275 rhaas@postgresql.org 3203 : 43368 : PGPROC *proc = &allProcs[arrayP->pgprocnos[index]];
3204 : :
7656 tgl@sss.pgh.pa.us 3205 [ + + ]: 43368 : if (proc->pid == pid)
3206 : : {
3207 : 8272 : result = proc;
3208 : 8272 : break;
3209 : : }
3210 : : }
3211 : :
3212 : 12610 : return result;
3213 : : }
3214 : :
3215 : : /*
3216 : : * BackendXidGetPid -- get a backend's pid given its XID
3217 : : *
3218 : : * Returns 0 if not found or it's a prepared transaction. Note that
3219 : : * it is up to the caller to be sure that the question remains
3220 : : * meaningful for long enough for the answer to be used ...
3221 : : *
3222 : : * Only main transaction Ids are considered. This function is mainly
3223 : : * useful for determining what backend owns a lock.
3224 : : *
3225 : : * Beware that not every xact has an XID assigned. However, as long as you
3226 : : * only call this using an XID found on disk, you're safe.
3227 : : */
3228 : : int
7563 ishii@postgresql.org 3229 : 30 : BackendXidGetPid(TransactionId xid)
3230 : : {
7507 bruce@momjian.us 3231 : 30 : int result = 0;
7563 ishii@postgresql.org 3232 : 30 : ProcArrayStruct *arrayP = procArray;
2090 andres@anarazel.de 3233 : 30 : TransactionId *other_xids = ProcGlobal->xids;
3234 : : int index;
3235 : :
7563 ishii@postgresql.org 3236 [ - + ]: 30 : if (xid == InvalidTransactionId) /* never match invalid xid */
7563 ishii@postgresql.org 3237 :UBC 0 : return 0;
3238 : :
7563 ishii@postgresql.org 3239 :CBC 30 : LWLockAcquire(ProcArrayLock, LW_SHARED);
3240 : :
3241 [ + + ]: 92 : for (index = 0; index < arrayP->numProcs; index++)
3242 : : {
2090 andres@anarazel.de 3243 [ + + ]: 84 : if (other_xids[index] == xid)
3244 : : {
970 michael@paquier.xyz 3245 : 22 : int pgprocno = arrayP->pgprocnos[index];
3246 : 22 : PGPROC *proc = &allProcs[pgprocno];
3247 : :
7563 ishii@postgresql.org 3248 : 22 : result = proc->pid;
3249 : 22 : break;
3250 : : }
3251 : : }
3252 : :
3253 : 30 : LWLockRelease(ProcArrayLock);
3254 : :
3255 : 30 : return result;
3256 : : }
3257 : :
3258 : : /*
3259 : : * IsBackendPid -- is a given pid a running backend
3260 : : *
3261 : : * This is not called by the backend, but is called by external modules.
3262 : : */
3263 : : bool
7656 tgl@sss.pgh.pa.us 3264 : 2 : IsBackendPid(int pid)
3265 : : {
3266 : 2 : return (BackendPidGetProc(pid) != NULL);
3267 : : }
3268 : :
3269 : :
3270 : : /*
3271 : : * GetCurrentVirtualXIDs -- returns an array of currently active VXIDs.
3272 : : *
3273 : : * The array is palloc'd. The number of valid entries is returned into *nvxids.
3274 : : *
3275 : : * The arguments allow filtering the set of VXIDs returned. Our own process
3276 : : * is always skipped. In addition:
3277 : : * If limitXmin is not InvalidTransactionId, skip processes with
3278 : : * xmin > limitXmin.
3279 : : * If excludeXmin0 is true, skip processes with xmin = 0.
3280 : : * If allDbs is false, skip processes attached to other databases.
3281 : : * If excludeVacuum isn't zero, skip processes for which
3282 : : * (statusFlags & excludeVacuum) is not zero.
3283 : : *
3284 : : * Note: the purpose of the limitXmin and excludeXmin0 parameters is to
3285 : : * allow skipping backends whose oldest live snapshot is no older than
3286 : : * some snapshot we have. Since we examine the procarray with only shared
3287 : : * lock, there are race conditions: a backend could set its xmin just after
3288 : : * we look. Indeed, on multiprocessors with weak memory ordering, the
3289 : : * other backend could have set its xmin *before* we look. We know however
3290 : : * that such a backend must have held shared ProcArrayLock overlapping our
3291 : : * own hold of ProcArrayLock, else we would see its xmin update. Therefore,
3292 : : * any snapshot the other backend is taking concurrently with our scan cannot
3293 : : * consider any transactions as still running that we think are committed
3294 : : * (since backends must hold ProcArrayLock exclusive to commit).
3295 : : */
3296 : : VirtualTransactionId *
6240 3297 : 494 : GetCurrentVirtualXIDs(TransactionId limitXmin, bool excludeXmin0,
3298 : : bool allDbs, int excludeVacuum,
3299 : : int *nvxids)
3300 : : {
3301 : : VirtualTransactionId *vxids;
6817 3302 : 494 : ProcArrayStruct *arrayP = procArray;
3303 : 494 : int count = 0;
3304 : : int index;
3305 : :
3306 : : /* allocate what's certainly enough result space */
146 michael@paquier.xyz 3307 :GNC 494 : vxids = palloc_array(VirtualTransactionId, arrayP->maxProcs);
3308 : :
6817 tgl@sss.pgh.pa.us 3309 :CBC 494 : LWLockAcquire(ProcArrayLock, LW_SHARED);
3310 : :
3311 [ + + ]: 3127 : for (index = 0; index < arrayP->numProcs; index++)
3312 : : {
5077 bruce@momjian.us 3313 : 2633 : int pgprocno = arrayP->pgprocnos[index];
2734 andres@anarazel.de 3314 : 2633 : PGPROC *proc = &allProcs[pgprocno];
1996 alvherre@alvh.no-ip. 3315 : 2633 : uint8 statusFlags = ProcGlobal->statusFlags[index];
3316 : :
6817 tgl@sss.pgh.pa.us 3317 [ + + ]: 2633 : if (proc == MyProc)
3318 : 494 : continue;
3319 : :
1996 alvherre@alvh.no-ip. 3320 [ + + ]: 2139 : if (excludeVacuum & statusFlags)
6691 tgl@sss.pgh.pa.us 3321 : 11 : continue;
3322 : :
6815 3323 [ + - + + ]: 2128 : if (allDbs || proc->databaseId == MyDatabaseId)
3324 : : {
3325 : : /* Fetch xmin just once - might change on us */
2091 andres@anarazel.de 3326 : 1053 : TransactionId pxmin = UINT32_ACCESS_ONCE(proc->xmin);
3327 : :
6240 tgl@sss.pgh.pa.us 3328 [ + - + + ]: 1053 : if (excludeXmin0 && !TransactionIdIsValid(pxmin))
3329 : 510 : continue;
3330 : :
3331 : : /*
3332 : : * InvalidTransactionId precedes all other XIDs, so a proc that
3333 : : * hasn't set xmin yet will not be rejected by this test.
3334 : : */
6815 3335 [ + - + + ]: 1086 : if (!TransactionIdIsValid(limitXmin) ||
6240 3336 : 543 : TransactionIdPrecedesOrEquals(pxmin, limitXmin))
3337 : : {
3338 : : VirtualTransactionId vxid;
3339 : :
6815 3340 : 498 : GET_VXID_FROM_PGPROC(vxid, *proc);
3341 [ + - ]: 498 : if (VirtualTransactionIdIsValid(vxid))
3342 : 498 : vxids[count++] = vxid;
3343 : : }
3344 : : }
3345 : : }
3346 : :
6817 3347 : 494 : LWLockRelease(ProcArrayLock);
3348 : :
6240 3349 : 494 : *nvxids = count;
6817 3350 : 494 : return vxids;
3351 : : }
3352 : :
3353 : : /*
3354 : : * GetConflictingVirtualXIDs -- returns an array of currently active VXIDs.
3355 : : *
3356 : : * Usage is limited to conflict resolution during recovery on standby servers.
3357 : : * limitXmin is supplied as either a cutoff with snapshotConflictHorizon
3358 : : * semantics, or InvalidTransactionId in cases where caller cannot accurately
3359 : : * determine a safe snapshotConflictHorizon value.
3360 : : *
3361 : : * If limitXmin is InvalidTransactionId then we want to kill everybody,
3362 : : * so we're not worried if they have a snapshot or not, nor does it really
3363 : : * matter what type of lock we hold. Caller must avoid calling here with
3364 : : * snapshotConflictHorizon style cutoffs that were set to InvalidTransactionId
3365 : : * during original execution, since that actually indicates that there is
3366 : : * definitely no need for a recovery conflict (the snapshotConflictHorizon
3367 : : * convention for InvalidTransactionId values is the opposite of our own!).
3368 : : *
3369 : : * All callers that are checking xmins always now supply a valid and useful
3370 : : * value for limitXmin. The limitXmin is always lower than the lowest
3371 : : * numbered KnownAssignedXid that is not already a FATAL error. This is
3372 : : * because we only care about cleanup records that are cleaning up tuple
3373 : : * versions from committed transactions. In that case they will only occur
3374 : : * at the point where the record is less than the lowest running xid. That
3375 : : * allows us to say that if any backend takes a snapshot concurrently with
3376 : : * us then the conflict assessment made here would never include the snapshot
3377 : : * that is being derived. So we take LW_SHARED on the ProcArray and allow
3378 : : * concurrent snapshots when limitXmin is valid. We might think about adding
3379 : : * Assert(limitXmin < lowest(KnownAssignedXids))
3380 : : * but that would not be true in the case of FATAL errors lagging in array,
3381 : : * but we already know those are bogus anyway, so we skip that test.
3382 : : *
3383 : : * If dbOid is valid we skip backends attached to other databases.
3384 : : *
3385 : : * Be careful to *not* pfree the result from this function. We reuse
3386 : : * this array sufficiently often that we use malloc for the result.
3387 : : */
3388 : : VirtualTransactionId *
5946 simon@2ndQuadrant.co 3389 : 16088 : GetConflictingVirtualXIDs(TransactionId limitXmin, Oid dbOid)
3390 : : {
3391 : : static VirtualTransactionId *vxids;
5981 3392 : 16088 : ProcArrayStruct *arrayP = procArray;
3393 : 16088 : int count = 0;
3394 : : int index;
3395 : :
3396 : : /*
3397 : : * If first time through, get workspace to remember main XIDs in. We
3398 : : * malloc it permanently to avoid repeated palloc/pfree overhead. Allow
3399 : : * result space, remembering room for a terminator.
3400 : : */
3401 [ + + ]: 16088 : if (vxids == NULL)
3402 : : {
3403 : 19 : vxids = (VirtualTransactionId *)
3404 : 19 : malloc(sizeof(VirtualTransactionId) * (arrayP->maxProcs + 1));
3405 [ - + ]: 19 : if (vxids == NULL)
5981 simon@2ndQuadrant.co 3406 [ # # ]:UBC 0 : ereport(ERROR,
3407 : : (errcode(ERRCODE_OUT_OF_MEMORY),
3408 : : errmsg("out of memory")));
3409 : : }
3410 : :
5858 simon@2ndQuadrant.co 3411 :CBC 16088 : LWLockAcquire(ProcArrayLock, LW_SHARED);
3412 : :
5981 3413 [ + + ]: 17259 : for (index = 0; index < arrayP->numProcs; index++)
3414 : : {
5077 bruce@momjian.us 3415 : 1171 : int pgprocno = arrayP->pgprocnos[index];
2734 andres@anarazel.de 3416 : 1171 : PGPROC *proc = &allProcs[pgprocno];
3417 : :
3418 : : /* Exclude prepared transactions */
5981 simon@2ndQuadrant.co 3419 [ - + ]: 1171 : if (proc->pid == 0)
5981 simon@2ndQuadrant.co 3420 :UBC 0 : continue;
3421 : :
5981 simon@2ndQuadrant.co 3422 [ + + ]:CBC 1171 : if (!OidIsValid(dbOid) ||
3423 [ + + ]: 1165 : proc->databaseId == dbOid)
3424 : : {
3425 : : /* Fetch xmin just once - can't change on us, but good coding */
2091 andres@anarazel.de 3426 : 498 : TransactionId pxmin = UINT32_ACCESS_ONCE(proc->xmin);
3427 : :
3428 : : /*
3429 : : * We ignore an invalid pxmin because this means that backend has
3430 : : * no snapshot currently. We hold a Share lock to avoid contention
3431 : : * with users taking snapshots. That is not a problem because the
3432 : : * current xmin is always at least one higher than the latest
3433 : : * removed xid, so any new snapshot would never conflict with the
3434 : : * test here.
3435 : : */
5858 simon@2ndQuadrant.co 3436 [ + + + + ]: 498 : if (!TransactionIdIsValid(limitXmin) ||
3437 [ + + ]: 483 : (TransactionIdIsValid(pxmin) && !TransactionIdFollows(pxmin, limitXmin)))
3438 : : {
3439 : : VirtualTransactionId vxid;
3440 : :
5981 3441 : 3 : GET_VXID_FROM_PGPROC(vxid, *proc);
3442 [ + - ]: 3 : if (VirtualTransactionIdIsValid(vxid))
3443 : 3 : vxids[count++] = vxid;
3444 : : }
3445 : : }
3446 : : }
3447 : :
3448 : 16088 : LWLockRelease(ProcArrayLock);
3449 : :
3450 : : /* add the terminator */
793 heikki.linnakangas@i 3451 : 16088 : vxids[count].procNumber = INVALID_PROC_NUMBER;
5981 simon@2ndQuadrant.co 3452 : 16088 : vxids[count].localTransactionId = InvalidLocalTransactionId;
3453 : :
3454 : 16088 : return vxids;
3455 : : }
3456 : :
3457 : : /*
3458 : : * SignalRecoveryConflict -- signal that a process is blocking recovery
3459 : : *
3460 : : * The 'pid' is redundant with 'proc', but it acts as a cross-check to
3461 : : * detect process had exited and the PGPROC entry was reused for a different
3462 : : * process.
3463 : : *
3464 : : * Returns true if the process was signaled, or false if not found.
3465 : : */
3466 : : bool
84 heikki.linnakangas@i 3467 :GNC 5 : SignalRecoveryConflict(PGPROC *proc, pid_t pid, RecoveryConflictReason reason)
3468 : : {
3469 : 5 : bool found = false;
3470 : :
3471 : 5 : LWLockAcquire(ProcArrayLock, LW_SHARED);
3472 : :
3473 : : /*
3474 : : * Kill the pid if it's still here. If not, that's what we wanted so
3475 : : * ignore any errors.
3476 : : */
3477 [ + - ]: 5 : if (proc->pid == pid)
3478 : : {
3479 : 5 : (void) pg_atomic_fetch_or_u32(&proc->pendingRecoveryConflicts, (1 << reason));
3480 : :
3481 : : /* wake up the process */
3482 : 5 : (void) SendProcSignal(pid, PROCSIG_RECOVERY_CONFLICT, GetNumberFromPGProc(proc));
3483 : 5 : found = true;
3484 : : }
3485 : :
3486 : 5 : LWLockRelease(ProcArrayLock);
3487 : :
3488 : 5 : return found;
3489 : : }
3490 : :
3491 : : /*
3492 : : * SignalRecoveryConflictWithVirtualXID -- signal that a VXID is blocking recovery
3493 : : *
3494 : : * Like SignalRecoveryConflict, but the target is identified by VXID
3495 : : */
3496 : : bool
3497 : 5 : SignalRecoveryConflictWithVirtualXID(VirtualTransactionId vxid, RecoveryConflictReason reason)
3498 : : {
5981 simon@2ndQuadrant.co 3499 :CBC 5 : ProcArrayStruct *arrayP = procArray;
3500 : : int index;
3501 : 5 : pid_t pid = 0;
3502 : :
3503 : 5 : LWLockAcquire(ProcArrayLock, LW_SHARED);
3504 : :
3505 [ + - ]: 5 : for (index = 0; index < arrayP->numProcs; index++)
3506 : : {
5077 bruce@momjian.us 3507 : 5 : int pgprocno = arrayP->pgprocnos[index];
2734 andres@anarazel.de 3508 : 5 : PGPROC *proc = &allProcs[pgprocno];
3509 : : VirtualTransactionId procvxid;
3510 : :
5981 simon@2ndQuadrant.co 3511 : 5 : GET_VXID_FROM_PGPROC(procvxid, *proc);
3512 : :
793 heikki.linnakangas@i 3513 [ + - ]: 5 : if (procvxid.procNumber == vxid.procNumber &&
5981 simon@2ndQuadrant.co 3514 [ + - ]: 5 : procvxid.localTransactionId == vxid.localTransactionId)
3515 : : {
3516 : 5 : pid = proc->pid;
5953 3517 [ + - ]: 5 : if (pid != 0)
3518 : : {
84 heikki.linnakangas@i 3519 :GNC 5 : (void) pg_atomic_fetch_or_u32(&proc->pendingRecoveryConflicts, (1 << reason));
3520 : :
3521 : : /*
3522 : : * Kill the pid if it's still here. If not, that's what we
3523 : : * wanted so ignore any errors.
3524 : : */
3525 : 5 : (void) SendProcSignal(pid, PROCSIG_RECOVERY_CONFLICT, vxid.procNumber);
3526 : : }
5981 simon@2ndQuadrant.co 3527 :CBC 5 : break;
3528 : : }
3529 : : }
3530 : :
3531 : 5 : LWLockRelease(ProcArrayLock);
3532 : :
84 heikki.linnakangas@i 3533 :GNC 5 : return pid != 0;
3534 : : }
3535 : :
3536 : : /*
3537 : : * SignalRecoveryConflictWithDatabase -- signal backends using specified database
3538 : : *
3539 : : * Like SignalRecoveryConflict, but signals all backends using the database.
3540 : : */
3541 : : void
3542 : 10 : SignalRecoveryConflictWithDatabase(Oid databaseid, RecoveryConflictReason reason)
3543 : : {
3544 : 10 : ProcArrayStruct *arrayP = procArray;
3545 : : int index;
3546 : :
3547 : : /* tell all backends to die */
3548 : 10 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
3549 : :
3550 [ + + ]: 20 : for (index = 0; index < arrayP->numProcs; index++)
3551 : : {
3552 : 10 : int pgprocno = arrayP->pgprocnos[index];
3553 : 10 : PGPROC *proc = &allProcs[pgprocno];
3554 : :
3555 [ + + + - ]: 10 : if (databaseid == InvalidOid || proc->databaseId == databaseid)
3556 : : {
3557 : : VirtualTransactionId procvxid;
3558 : : pid_t pid;
3559 : :
3560 : 10 : GET_VXID_FROM_PGPROC(procvxid, *proc);
3561 : :
3562 : 10 : pid = proc->pid;
3563 [ + - ]: 10 : if (pid != 0)
3564 : : {
3565 : 10 : (void) pg_atomic_fetch_or_u32(&proc->pendingRecoveryConflicts, (1 << reason));
3566 : :
3567 : : /*
3568 : : * Kill the pid if it's still here. If not, that's what we
3569 : : * wanted so ignore any errors.
3570 : : */
3571 : 10 : (void) SendProcSignal(pid, PROCSIG_RECOVERY_CONFLICT, procvxid.procNumber);
3572 : : }
3573 : : }
3574 : : }
3575 : :
3576 : 10 : LWLockRelease(ProcArrayLock);
5981 simon@2ndQuadrant.co 3577 :GIC 10 : }
3578 : :
3579 : : /*
3580 : : * MinimumActiveBackends --- count backends (other than myself) that are
3581 : : * in active transactions. Return true if the count exceeds the
3582 : : * minimum threshold passed. This is used as a heuristic to decide if
3583 : : * a pre-XLOG-flush delay is worthwhile during commit.
3584 : : *
3585 : : * Do not count backends that are blocked waiting for locks, since they are
3586 : : * not going to get to run until someone else commits.
3587 : : */
3588 : : bool
5627 simon@2ndQuadrant.co 3589 :UBC 0 : MinimumActiveBackends(int min)
3590 : : {
7656 tgl@sss.pgh.pa.us 3591 : 0 : ProcArrayStruct *arrayP = procArray;
3592 : 0 : int count = 0;
3593 : : int index;
3594 : :
3595 : : /* Quick short-circuit if no minimum is specified */
5627 simon@2ndQuadrant.co 3596 [ # # ]: 0 : if (min == 0)
3597 : 0 : return true;
3598 : :
3599 : : /*
3600 : : * Note: for speed, we don't acquire ProcArrayLock. This is a little bit
3601 : : * bogus, but since we are only testing fields for zero or nonzero, it
3602 : : * should be OK. The result is only used for heuristic purposes anyway...
3603 : : */
7656 tgl@sss.pgh.pa.us 3604 [ # # ]: 0 : for (index = 0; index < arrayP->numProcs; index++)
3605 : : {
5077 bruce@momjian.us 3606 : 0 : int pgprocno = arrayP->pgprocnos[index];
2734 andres@anarazel.de 3607 : 0 : PGPROC *proc = &allProcs[pgprocno];
3608 : :
3609 : : /*
3610 : : * Since we're not holding a lock, need to be prepared to deal with
3611 : : * garbage, as someone could have incremented numProcs but not yet
3612 : : * filled the structure.
3613 : : *
3614 : : * If someone just decremented numProcs, 'proc' could also point to a
3615 : : * PGPROC entry that's no longer in the array. It still points to a
3616 : : * PGPROC struct, though, because freed PGPROC entries just go to the
3617 : : * free list and are recycled. Its contents are nonsense in that case,
3618 : : * but that's acceptable for this function.
3619 : : */
4131 sfrost@snowman.net 3620 [ # # ]: 0 : if (pgprocno == -1)
3621 : 0 : continue; /* do not count deleted entries */
7656 tgl@sss.pgh.pa.us 3622 [ # # ]: 0 : if (proc == MyProc)
3623 : 0 : continue; /* do not count myself */
2090 andres@anarazel.de 3624 [ # # ]: 0 : if (proc->xid == InvalidTransactionId)
5275 rhaas@postgresql.org 3625 : 0 : continue; /* do not count if no XID assigned */
7627 tgl@sss.pgh.pa.us 3626 [ # # ]: 0 : if (proc->pid == 0)
3627 : 0 : continue; /* do not count prepared xacts */
7656 3628 [ # # ]: 0 : if (proc->waitLock != NULL)
3629 : 0 : continue; /* do not count if blocked on a lock */
3630 : 0 : count++;
5627 simon@2ndQuadrant.co 3631 [ # # ]: 0 : if (count >= min)
3632 : 0 : break;
3633 : : }
3634 : :
3635 : 0 : return count >= min;
3636 : : }
3637 : :
3638 : : /*
3639 : : * CountDBBackends --- count backends that are using specified database
3640 : : */
3641 : : int
7583 tgl@sss.pgh.pa.us 3642 :CBC 16 : CountDBBackends(Oid databaseid)
3643 : : {
3644 : 16 : ProcArrayStruct *arrayP = procArray;
3645 : 16 : int count = 0;
3646 : : int index;
3647 : :
3648 : 16 : LWLockAcquire(ProcArrayLock, LW_SHARED);
3649 : :
3650 [ + + ]: 23 : for (index = 0; index < arrayP->numProcs; index++)
3651 : : {
5077 bruce@momjian.us 3652 : 7 : int pgprocno = arrayP->pgprocnos[index];
2734 andres@anarazel.de 3653 : 7 : PGPROC *proc = &allProcs[pgprocno];
3654 : :
7583 tgl@sss.pgh.pa.us 3655 [ - + ]: 7 : if (proc->pid == 0)
7583 tgl@sss.pgh.pa.us 3656 :UBC 0 : continue; /* do not count prepared xacts */
5857 simon@2ndQuadrant.co 3657 [ + - ]:CBC 7 : if (!OidIsValid(databaseid) ||
3658 [ + + ]: 7 : proc->databaseId == databaseid)
7583 tgl@sss.pgh.pa.us 3659 : 2 : count++;
3660 : : }
3661 : :
3662 : 16 : LWLockRelease(ProcArrayLock);
3663 : :
3664 : 16 : return count;
3665 : : }
3666 : :
3667 : : /*
3668 : : * CountDBConnections --- counts database backends (only regular backends)
3669 : : */
3670 : : int
3380 andrew@dunslane.net 3671 :UBC 0 : CountDBConnections(Oid databaseid)
3672 : : {
3673 : 0 : ProcArrayStruct *arrayP = procArray;
3674 : 0 : int count = 0;
3675 : : int index;
3676 : :
3677 : 0 : LWLockAcquire(ProcArrayLock, LW_SHARED);
3678 : :
3679 [ # # ]: 0 : for (index = 0; index < arrayP->numProcs; index++)
3680 : : {
3681 : 0 : int pgprocno = arrayP->pgprocnos[index];
2734 andres@anarazel.de 3682 : 0 : PGPROC *proc = &allProcs[pgprocno];
3683 : :
3380 andrew@dunslane.net 3684 [ # # ]: 0 : if (proc->pid == 0)
3685 : 0 : continue; /* do not count prepared xacts */
90 heikki.linnakangas@i 3686 [ # # ]:UNC 0 : if (proc->backendType != B_BACKEND)
493 tgl@sss.pgh.pa.us 3687 :UBC 0 : continue; /* count only regular backend processes */
3380 andrew@dunslane.net 3688 [ # # ]: 0 : if (!OidIsValid(databaseid) ||
3689 [ # # ]: 0 : proc->databaseId == databaseid)
3690 : 0 : count++;
3691 : : }
3692 : :
3693 : 0 : LWLockRelease(ProcArrayLock);
3694 : :
3695 : 0 : return count;
3696 : : }
3697 : :
3698 : : /*
3699 : : * CountUserBackends --- count backends that are used by specified user
3700 : : * (only regular backends, not any type of background worker)
3701 : : */
3702 : : int
7583 tgl@sss.pgh.pa.us 3703 : 0 : CountUserBackends(Oid roleid)
3704 : : {
3705 : 0 : ProcArrayStruct *arrayP = procArray;
3706 : 0 : int count = 0;
3707 : : int index;
3708 : :
3709 : 0 : LWLockAcquire(ProcArrayLock, LW_SHARED);
3710 : :
3711 [ # # ]: 0 : for (index = 0; index < arrayP->numProcs; index++)
3712 : : {
5077 bruce@momjian.us 3713 : 0 : int pgprocno = arrayP->pgprocnos[index];
2734 andres@anarazel.de 3714 : 0 : PGPROC *proc = &allProcs[pgprocno];
3715 : :
7583 tgl@sss.pgh.pa.us 3716 [ # # ]: 0 : if (proc->pid == 0)
3717 : 0 : continue; /* do not count prepared xacts */
90 heikki.linnakangas@i 3718 [ # # ]:UNC 0 : if (proc->backendType != B_BACKEND)
493 tgl@sss.pgh.pa.us 3719 :UBC 0 : continue; /* count only regular backend processes */
7583 3720 [ # # ]: 0 : if (proc->roleId == roleid)
3721 : 0 : count++;
3722 : : }
3723 : :
3724 : 0 : LWLockRelease(ProcArrayLock);
3725 : :
3726 : 0 : return count;
3727 : : }
3728 : :
3729 : : /*
3730 : : * CountOtherDBBackends -- check for other backends running in the given DB
3731 : : *
3732 : : * If there are other backends in the DB, we will wait a maximum of 5 seconds
3733 : : * for them to exit (or 0.3s for testing purposes). Autovacuum backends are
3734 : : * encouraged to exit early by sending them SIGTERM, but normal user backends
3735 : : * are just waited for. If background workers connected to this database are
3736 : : * marked as interruptible, they are terminated.
3737 : : *
3738 : : * The current backend is always ignored; it is caller's responsibility to
3739 : : * check whether the current backend uses the given DB, if it's important.
3740 : : *
3741 : : * Returns true if there are (still) other backends in the DB, false if not.
3742 : : * Also, *nbackends and *nprepared are set to the number of other backends
3743 : : * and prepared transactions in the DB, respectively.
3744 : : *
3745 : : * This function is used to interlock DROP DATABASE and related commands
3746 : : * against there being any active backends in the target DB --- dropping the
3747 : : * DB while active backends remain would be a Bad Thing. Note that we cannot
3748 : : * detect here the possibility of a newly-started backend that is trying to
3749 : : * connect to the doomed database, so additional interlocking is needed during
3750 : : * backend startup. The caller should normally hold an exclusive lock on the
3751 : : * target DB before calling this, which is one reason we mustn't wait
3752 : : * indefinitely.
3753 : : */
3754 : : bool
6483 tgl@sss.pgh.pa.us 3755 :CBC 499 : CountOtherDBBackends(Oid databaseId, int *nbackends, int *nprepared)
3756 : : {
6913 3757 : 499 : ProcArrayStruct *arrayP = procArray;
3758 : :
3759 : : #define MAXAUTOVACPIDS 10 /* max autovacs to SIGTERM per iteration */
3760 : : int autovac_pids[MAXAUTOVACPIDS];
3761 : :
3762 : : /*
3763 : : * Retry up to 50 times with 100ms between attempts (max 5s total). Can be
3764 : : * reduced to 3 attempts (max 0.3s total) to speed up tests.
3765 : : */
119 michael@paquier.xyz 3766 :GNC 499 : int ntries = 50;
3767 : :
3768 : : #ifdef USE_INJECTION_POINTS
3769 [ + + ]: 499 : if (IS_INJECTION_POINT_ATTACHED("procarray-reduce-count"))
3770 : 1 : ntries = 3;
3771 : : #endif
3772 : :
3773 [ + + ]: 506 : for (int tries = 0; tries < ntries; tries++)
3774 : : {
6483 tgl@sss.pgh.pa.us 3775 :CBC 505 : int nautovacs = 0;
6913 3776 : 505 : bool found = false;
3777 : : int index;
3778 : :
3779 [ - + ]: 505 : CHECK_FOR_INTERRUPTS();
3780 : :
6483 3781 : 505 : *nbackends = *nprepared = 0;
3782 : :
6913 3783 : 505 : LWLockAcquire(ProcArrayLock, LW_SHARED);
3784 : :
3785 [ + + ]: 1940 : for (index = 0; index < arrayP->numProcs; index++)
3786 : : {
5077 bruce@momjian.us 3787 : 1435 : int pgprocno = arrayP->pgprocnos[index];
2734 andres@anarazel.de 3788 : 1435 : PGPROC *proc = &allProcs[pgprocno];
1996 alvherre@alvh.no-ip. 3789 : 1435 : uint8 statusFlags = ProcGlobal->statusFlags[index];
3790 : :
6913 tgl@sss.pgh.pa.us 3791 [ + + ]: 1435 : if (proc->databaseId != databaseId)
3792 : 1313 : continue;
3793 [ + + ]: 122 : if (proc == MyProc)
3794 : 115 : continue;
3795 : :
6913 tgl@sss.pgh.pa.us 3796 :GBC 7 : found = true;
3797 : :
6483 3798 [ - + ]: 7 : if (proc->pid == 0)
6483 tgl@sss.pgh.pa.us 3799 :UBC 0 : (*nprepared)++;
3800 : : else
3801 : : {
6483 tgl@sss.pgh.pa.us 3802 :GBC 7 : (*nbackends)++;
1996 alvherre@alvh.no-ip. 3803 [ - + - - ]: 7 : if ((statusFlags & PROC_IS_AUTOVACUUM) &&
3804 : : nautovacs < MAXAUTOVACPIDS)
6483 tgl@sss.pgh.pa.us 3805 :UBC 0 : autovac_pids[nautovacs++] = proc->pid;
3806 : : }
3807 : : }
3808 : :
6483 tgl@sss.pgh.pa.us 3809 :CBC 505 : LWLockRelease(ProcArrayLock);
3810 : :
6913 3811 [ + + ]: 505 : if (!found)
6746 bruce@momjian.us 3812 : 498 : return false; /* no conflicting backends, so done */
3813 : :
3814 : : /*
3815 : : * Send SIGTERM to any conflicting autovacuums before sleeping. We
3816 : : * postpone this step until after the loop because we don't want to
3817 : : * hold ProcArrayLock while issuing kill(). We have no idea what might
3818 : : * block kill() inside the kernel...
3819 : : */
6483 tgl@sss.pgh.pa.us 3820 [ - + ]:GBC 7 : for (index = 0; index < nautovacs; index++)
6483 tgl@sss.pgh.pa.us 3821 :UBC 0 : (void) kill(autovac_pids[index], SIGTERM); /* ignore any error */
3822 : :
3823 : : /*
3824 : : * Terminate all background workers for this database, if they have
3825 : : * requested it (BGWORKER_INTERRUPTIBLE).
3826 : : */
119 michael@paquier.xyz 3827 :GNC 7 : TerminateBackgroundWorkersForDatabase(databaseId);
3828 : :
3829 : : /* sleep, then try again */
6746 bruce@momjian.us 3830 :GBC 7 : pg_usleep(100 * 1000L); /* 100ms */
3831 : : }
3832 : :
3833 : 1 : return true; /* timed out, still conflicts */
3834 : : }
3835 : :
3836 : : /*
3837 : : * Terminate existing connections to the specified database. This routine
3838 : : * is used by the DROP DATABASE command when user has asked to forcefully
3839 : : * drop the database.
3840 : : *
3841 : : * The current backend is always ignored; it is caller's responsibility to
3842 : : * check whether the current backend uses the given DB, if it's important.
3843 : : *
3844 : : * If the target database has a prepared transaction or permissions checks
3845 : : * fail for a connection, this fails without terminating anything.
3846 : : */
3847 : : void
2366 akapila@postgresql.o 3848 :CBC 1 : TerminateOtherDBBackends(Oid databaseId)
3849 : : {
3850 : 1 : ProcArrayStruct *arrayP = procArray;
3851 : 1 : List *pids = NIL;
3852 : 1 : int nprepared = 0;
3853 : : int i;
3854 : :
3855 : 1 : LWLockAcquire(ProcArrayLock, LW_SHARED);
3856 : :
3857 [ + + ]: 4 : for (i = 0; i < procArray->numProcs; i++)
3858 : : {
3859 : 3 : int pgprocno = arrayP->pgprocnos[i];
3860 : 3 : PGPROC *proc = &allProcs[pgprocno];
3861 : :
3862 [ + - ]: 3 : if (proc->databaseId != databaseId)
3863 : 3 : continue;
2366 akapila@postgresql.o 3864 [ # # ]:UBC 0 : if (proc == MyProc)
3865 : 0 : continue;
3866 : :
3867 [ # # ]: 0 : if (proc->pid != 0)
3868 : 0 : pids = lappend_int(pids, proc->pid);
3869 : : else
3870 : 0 : nprepared++;
3871 : : }
3872 : :
2366 akapila@postgresql.o 3873 :CBC 1 : LWLockRelease(ProcArrayLock);
3874 : :
3875 [ - + ]: 1 : if (nprepared > 0)
2366 akapila@postgresql.o 3876 [ # # ]:UBC 0 : ereport(ERROR,
3877 : : (errcode(ERRCODE_OBJECT_IN_USE),
3878 : : errmsg("database \"%s\" is being used by prepared transactions",
3879 : : get_database_name(databaseId)),
3880 : : errdetail_plural("There is %d prepared transaction using the database.",
3881 : : "There are %d prepared transactions using the database.",
3882 : : nprepared,
3883 : : nprepared)));
3884 : :
2366 akapila@postgresql.o 3885 [ - + ]:CBC 1 : if (pids)
3886 : : {
3887 : : ListCell *lc;
3888 : :
3889 : : /*
3890 : : * Permissions checks relax the pg_terminate_backend checks in two
3891 : : * ways, both by omitting the !OidIsValid(proc->roleId) check:
3892 : : *
3893 : : * - Accept terminating autovacuum workers, since DROP DATABASE
3894 : : * without FORCE terminates them.
3895 : : *
3896 : : * - Accept terminating bgworkers. For bgworker authors, it's
3897 : : * convenient to be able to recommend FORCE if a worker is blocking
3898 : : * DROP DATABASE unexpectedly.
3899 : : *
3900 : : * Unlike pg_terminate_backend, we don't raise some warnings - like
3901 : : * "PID %d is not a PostgreSQL server process", because for us already
3902 : : * finished session is not a problem.
3903 : : */
2366 akapila@postgresql.o 3904 [ # # # # :UBC 0 : foreach(lc, pids)
# # ]
3905 : : {
3906 : 0 : int pid = lfirst_int(lc);
3907 : 0 : PGPROC *proc = BackendPidGetProc(pid);
3908 : :
3909 [ # # ]: 0 : if (proc != NULL)
3910 : : {
3911 [ # # # # ]: 0 : if (superuser_arg(proc->roleId) && !superuser())
3912 [ # # ]: 0 : ereport(ERROR,
3913 : : (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
3914 : : errmsg("permission denied to terminate process"),
3915 : : errdetail("Only roles with the %s attribute may terminate processes of roles with the %s attribute.",
3916 : : "SUPERUSER", "SUPERUSER")));
3917 : :
3918 [ # # ]: 0 : if (!has_privs_of_role(GetUserId(), proc->roleId) &&
1860 sfrost@snowman.net 3919 [ # # ]: 0 : !has_privs_of_role(GetUserId(), ROLE_PG_SIGNAL_BACKEND))
2366 akapila@postgresql.o 3920 [ # # ]: 0 : ereport(ERROR,
3921 : : (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
3922 : : errmsg("permission denied to terminate process"),
3923 : : errdetail("Only roles with privileges of the role whose process is being terminated or with privileges of the \"%s\" role may terminate this process.",
3924 : : "pg_signal_backend")));
3925 : : }
3926 : : }
3927 : :
3928 : : /*
3929 : : * There's a race condition here: once we release the ProcArrayLock,
3930 : : * it's possible for the session to exit before we issue kill. That
3931 : : * race condition possibility seems too unlikely to worry about. See
3932 : : * pg_signal_backend.
3933 : : */
3934 [ # # # # : 0 : foreach(lc, pids)
# # ]
3935 : : {
3936 : 0 : int pid = lfirst_int(lc);
3937 : 0 : PGPROC *proc = BackendPidGetProc(pid);
3938 : :
3939 [ # # ]: 0 : if (proc != NULL)
3940 : : {
3941 : : /*
3942 : : * If we have setsid(), signal the backend's whole process
3943 : : * group
3944 : : */
3945 : : #ifdef HAVE_SETSID
3946 : 0 : (void) kill(-pid, SIGTERM);
3947 : : #else
3948 : : (void) kill(pid, SIGTERM);
3949 : : #endif
3950 : : }
3951 : : }
3952 : : }
2366 akapila@postgresql.o 3953 :CBC 1 : }
3954 : :
3955 : : /*
3956 : : * ProcArraySetReplicationSlotXmin
3957 : : *
3958 : : * Install limits to future computations of the xmin horizon to prevent vacuum
3959 : : * and HOT pruning from removing affected rows still needed by clients with
3960 : : * replication slots.
3961 : : */
3962 : : void
4446 rhaas@postgresql.org 3963 : 2612 : ProcArraySetReplicationSlotXmin(TransactionId xmin, TransactionId catalog_xmin,
3964 : : bool already_locked)
3965 : : {
3966 [ + + - + ]: 2612 : Assert(!already_locked || LWLockHeldByMe(ProcArrayLock));
3967 : :
3968 [ + + ]: 2612 : if (!already_locked)
3969 : 2100 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
3970 : :
4477 3971 : 2612 : procArray->replication_slot_xmin = xmin;
4446 3972 : 2612 : procArray->replication_slot_catalog_xmin = catalog_xmin;
3973 : :
3974 [ + + ]: 2612 : if (!already_locked)
3975 : 2100 : LWLockRelease(ProcArrayLock);
3976 : :
1260 alvherre@alvh.no-ip. 3977 [ + + ]: 2612 : elog(DEBUG1, "xmin required by slots: data %u, catalog %u",
3978 : : xmin, catalog_xmin);
4446 rhaas@postgresql.org 3979 : 2612 : }
3980 : :
3981 : : /*
3982 : : * ProcArrayGetReplicationSlotXmin
3983 : : *
3984 : : * Return the current slot xmin limits. That's useful to be able to remove
3985 : : * data that's older than those limits.
3986 : : */
3987 : : void
3988 : 25 : ProcArrayGetReplicationSlotXmin(TransactionId *xmin,
3989 : : TransactionId *catalog_xmin)
3990 : : {
3991 : 25 : LWLockAcquire(ProcArrayLock, LW_SHARED);
3992 : :
3993 [ - + ]: 25 : if (xmin != NULL)
4446 rhaas@postgresql.org 3994 :UBC 0 : *xmin = procArray->replication_slot_xmin;
3995 : :
4446 rhaas@postgresql.org 3996 [ + - ]:CBC 25 : if (catalog_xmin != NULL)
3997 : 25 : *catalog_xmin = procArray->replication_slot_catalog_xmin;
3998 : :
4477 3999 : 25 : LWLockRelease(ProcArrayLock);
4000 : 25 : }
4001 : :
4002 : : /*
4003 : : * XidCacheRemoveRunningXids
4004 : : *
4005 : : * Remove a bunch of TransactionIds from the list of known-running
4006 : : * subtransactions for my backend. Both the specified xid and those in
4007 : : * the xids[] array (of length nxids) are removed from the subxids cache.
4008 : : * latestXid must be the latest XID among the group.
4009 : : */
4010 : : void
6814 tgl@sss.pgh.pa.us 4011 : 855 : XidCacheRemoveRunningXids(TransactionId xid,
4012 : : int nxids, const TransactionId *xids,
4013 : : TransactionId latestXid)
4014 : : {
4015 : : int i,
4016 : : j;
4017 : : XidCacheStatus *mysubxidstat;
4018 : :
7184 4019 [ - + ]: 855 : Assert(TransactionIdIsValid(xid));
4020 : :
4021 : : /*
4022 : : * We must hold ProcArrayLock exclusively in order to remove transactions
4023 : : * from the PGPROC array. (See src/backend/access/transam/README.) It's
4024 : : * possible this could be relaxed since we know this routine is only used
4025 : : * to abort subtransactions, but pending closer analysis we'd best be
4026 : : * conservative.
4027 : : *
4028 : : * Note that we do not have to be careful about memory ordering of our own
4029 : : * reads wrt. GetNewTransactionId() here - only this process can modify
4030 : : * relevant fields of MyProc/ProcGlobal->xids[]. But we do have to be
4031 : : * careful about our own writes being well ordered.
4032 : : */
7656 4033 : 855 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
4034 : :
2090 andres@anarazel.de 4035 : 855 : mysubxidstat = &ProcGlobal->subxidStates[MyProc->pgxactoff];
4036 : :
4037 : : /*
4038 : : * Under normal circumstances xid and xids[] will be in increasing order,
4039 : : * as will be the entries in subxids. Scan backwards to avoid O(N^2)
4040 : : * behavior when removing a lot of xids.
4041 : : */
7656 tgl@sss.pgh.pa.us 4042 [ + + ]: 886 : for (i = nxids - 1; i >= 0; i--)
4043 : : {
4044 : 31 : TransactionId anxid = xids[i];
4045 : :
2090 andres@anarazel.de 4046 [ + - ]: 31 : for (j = MyProc->subxidStatus.count - 1; j >= 0; j--)
4047 : : {
7656 tgl@sss.pgh.pa.us 4048 [ + - ]: 31 : if (TransactionIdEquals(MyProc->subxids.xids[j], anxid))
4049 : : {
2090 andres@anarazel.de 4050 : 31 : MyProc->subxids.xids[j] = MyProc->subxids.xids[MyProc->subxidStatus.count - 1];
4051 : 31 : pg_write_barrier();
4052 : 31 : mysubxidstat->count--;
4053 : 31 : MyProc->subxidStatus.count--;
7656 tgl@sss.pgh.pa.us 4054 : 31 : break;
4055 : : }
4056 : : }
4057 : :
4058 : : /*
4059 : : * Ordinarily we should have found it, unless the cache has
4060 : : * overflowed. However it's also possible for this routine to be
4061 : : * invoked multiple times for the same subtransaction, in case of an
4062 : : * error during AbortSubTransaction. So instead of Assert, emit a
4063 : : * debug warning.
4064 : : */
2090 andres@anarazel.de 4065 [ - + - - ]: 31 : if (j < 0 && !MyProc->subxidStatus.overflowed)
7656 tgl@sss.pgh.pa.us 4066 [ # # ]:UBC 0 : elog(WARNING, "did not find subXID %u in MyProc", anxid);
4067 : : }
4068 : :
2090 andres@anarazel.de 4069 [ + + ]:CBC 1047 : for (j = MyProc->subxidStatus.count - 1; j >= 0; j--)
4070 : : {
7656 tgl@sss.pgh.pa.us 4071 [ + + ]: 1044 : if (TransactionIdEquals(MyProc->subxids.xids[j], xid))
4072 : : {
2090 andres@anarazel.de 4073 : 852 : MyProc->subxids.xids[j] = MyProc->subxids.xids[MyProc->subxidStatus.count - 1];
4074 : 852 : pg_write_barrier();
4075 : 852 : mysubxidstat->count--;
4076 : 852 : MyProc->subxidStatus.count--;
7656 tgl@sss.pgh.pa.us 4077 : 852 : break;
4078 : : }
4079 : : }
4080 : : /* Ordinarily we should have found it, unless the cache has overflowed */
2090 andres@anarazel.de 4081 [ + + - + ]: 855 : if (j < 0 && !MyProc->subxidStatus.overflowed)
7656 tgl@sss.pgh.pa.us 4082 [ # # ]:UBC 0 : elog(WARNING, "did not find subXID %u in MyProc", xid);
4083 : :
4084 : : /* Also advance global latestCompletedXid while holding the lock */
2093 andres@anarazel.de 4085 :CBC 855 : MaintainLatestCompletedXid(latestXid);
4086 : :
4087 : : /* ... and xactCompletionCount */
879 heikki.linnakangas@i 4088 : 855 : TransamVariables->xactCompletionCount++;
4089 : :
7656 tgl@sss.pgh.pa.us 4090 : 855 : LWLockRelease(ProcArrayLock);
4091 : 855 : }
4092 : :
4093 : : #ifdef XIDCACHE_DEBUG
4094 : :
4095 : : /*
4096 : : * Print stats about effectiveness of XID cache
4097 : : */
4098 : : static void
4099 : : DisplayXidCache(void)
4100 : : {
4101 : : fprintf(stderr,
4102 : : "XidCache: xmin: %ld, known: %ld, myxact: %ld, latest: %ld, mainxid: %ld, childxid: %ld, knownassigned: %ld, nooflo: %ld, slow: %ld\n",
4103 : : xc_by_recent_xmin,
4104 : : xc_by_known_xact,
4105 : : xc_by_my_xact,
4106 : : xc_by_latest_xid,
4107 : : xc_by_main_xid,
4108 : : xc_by_child_xid,
4109 : : xc_by_known_assigned,
4110 : : xc_no_overflow,
4111 : : xc_slow_answer);
4112 : : }
4113 : : #endif /* XIDCACHE_DEBUG */
4114 : :
4115 : : /*
4116 : : * If rel != NULL, return test state appropriate for relation, otherwise
4117 : : * return state usable for all relations. The latter may consider XIDs as
4118 : : * not-yet-visible-to-everyone that a state for a specific relation would
4119 : : * already consider visible-to-everyone.
4120 : : *
4121 : : * This needs to be called while a snapshot is active or registered, otherwise
4122 : : * there are wraparound and other dangers.
4123 : : *
4124 : : * See comment for GlobalVisState for details.
4125 : : */
4126 : : GlobalVisState *
2092 andres@anarazel.de 4127 : 20000508 : GlobalVisTestFor(Relation rel)
4128 : : {
1746 4129 : 20000508 : GlobalVisState *state = NULL;
4130 : :
4131 : : /* XXX: we should assert that a snapshot is pushed or registered */
2092 4132 [ - + ]: 20000508 : Assert(RecentXmin);
4133 : :
1746 4134 [ + + + + : 20000508 : switch (GlobalVisHorizonKindForRel(rel))
- ]
4135 : : {
4136 : 129725 : case VISHORIZON_SHARED:
4137 : 129725 : state = &GlobalVisSharedRels;
4138 : 129725 : break;
4139 : 4363140 : case VISHORIZON_CATALOG:
4140 : 4363140 : state = &GlobalVisCatalogRels;
4141 : 4363140 : break;
4142 : 15401258 : case VISHORIZON_DATA:
4143 : 15401258 : state = &GlobalVisDataRels;
4144 : 15401258 : break;
4145 : 106385 : case VISHORIZON_TEMP:
4146 : 106385 : state = &GlobalVisTempRels;
4147 : 106385 : break;
4148 : : }
4149 : :
2092 4150 [ + - - + ]: 20000508 : Assert(FullTransactionIdIsValid(state->definitely_needed) &&
4151 : : FullTransactionIdIsValid(state->maybe_needed));
4152 : :
4153 : 20000508 : return state;
4154 : : }
4155 : :
4156 : : /*
4157 : : * Return true if it's worth updating the accurate maybe_needed boundary.
4158 : : *
4159 : : * As it is somewhat expensive to determine xmin horizons, we don't want to
4160 : : * repeatedly do so when there is a low likelihood of it being beneficial.
4161 : : *
4162 : : * The current heuristic is that we update only if RecentXmin has changed
4163 : : * since the last update. If the oldest currently running transaction has not
4164 : : * finished, it is unlikely that recomputing the horizon would be useful.
4165 : : */
4166 : : static bool
4167 : 756783 : GlobalVisTestShouldUpdate(GlobalVisState *state)
4168 : : {
4169 : : /* hasn't been updated yet */
4170 [ + + ]: 756783 : if (!TransactionIdIsValid(ComputeXidHorizonsResultLastXmin))
4171 : 16812 : return true;
4172 : :
4173 : : /*
4174 : : * If the maybe_needed/definitely_needed boundaries are the same, it's
4175 : : * unlikely to be beneficial to refresh boundaries.
4176 : : */
4177 [ - + ]: 739971 : if (FullTransactionIdFollowsOrEquals(state->maybe_needed,
4178 : : state->definitely_needed))
2092 andres@anarazel.de 4179 :UBC 0 : return false;
4180 : :
4181 : : /* does the last snapshot built have a different xmin? */
2092 andres@anarazel.de 4182 :CBC 739971 : return RecentXmin != ComputeXidHorizonsResultLastXmin;
4183 : : }
4184 : :
4185 : : static void
4186 : 127882 : GlobalVisUpdateApply(ComputeXidHorizonsResult *horizons)
4187 : : {
4188 : : GlobalVisSharedRels.maybe_needed =
4189 : 127882 : FullXidRelativeTo(horizons->latest_completed,
4190 : : horizons->shared_oldest_nonremovable);
4191 : : GlobalVisCatalogRels.maybe_needed =
4192 : 127882 : FullXidRelativeTo(horizons->latest_completed,
4193 : : horizons->catalog_oldest_nonremovable);
4194 : : GlobalVisDataRels.maybe_needed =
4195 : 127882 : FullXidRelativeTo(horizons->latest_completed,
4196 : : horizons->data_oldest_nonremovable);
4197 : : GlobalVisTempRels.maybe_needed =
2015 4198 : 127882 : FullXidRelativeTo(horizons->latest_completed,
4199 : : horizons->temp_oldest_nonremovable);
4200 : :
4201 : : /*
4202 : : * In longer running transactions it's possible that transactions we
4203 : : * previously needed to treat as running aren't around anymore. So update
4204 : : * definitely_needed to not be earlier than maybe_needed.
4205 : : */
4206 : : GlobalVisSharedRels.definitely_needed =
2092 4207 : 127882 : FullTransactionIdNewer(GlobalVisSharedRels.maybe_needed,
4208 : : GlobalVisSharedRels.definitely_needed);
4209 : : GlobalVisCatalogRels.definitely_needed =
4210 : 127882 : FullTransactionIdNewer(GlobalVisCatalogRels.maybe_needed,
4211 : : GlobalVisCatalogRels.definitely_needed);
4212 : : GlobalVisDataRels.definitely_needed =
4213 : 127882 : FullTransactionIdNewer(GlobalVisDataRels.maybe_needed,
4214 : : GlobalVisDataRels.definitely_needed);
2015 4215 : 127882 : GlobalVisTempRels.definitely_needed = GlobalVisTempRels.maybe_needed;
4216 : :
2092 4217 : 127882 : ComputeXidHorizonsResultLastXmin = RecentXmin;
4218 : 127882 : }
4219 : :
4220 : : /*
4221 : : * Update boundaries in GlobalVis{Shared,Catalog, Data}Rels
4222 : : * using ComputeXidHorizons().
4223 : : */
4224 : : static void
4225 : 81990 : GlobalVisUpdate(void)
4226 : : {
4227 : : ComputeXidHorizonsResult horizons;
4228 : :
4229 : : /* updates the horizons as a side-effect */
4230 : 81990 : ComputeXidHorizons(&horizons);
4231 : 81990 : }
4232 : :
4233 : : /*
4234 : : * Return true if no snapshot still considers fxid to be running.
4235 : : *
4236 : : * The state passed needs to have been initialized for the relation fxid is
4237 : : * from (NULL is also OK), otherwise the result may not be correct.
4238 : : *
4239 : : * If allow_update is false, the GlobalVisState boundaries will not be updated
4240 : : * even if it would otherwise be beneficial. This is useful for callers that
4241 : : * do not want GlobalVisState to advance at all, for example because they need
4242 : : * a conservative answer based on the current boundaries.
4243 : : *
4244 : : * See comment for GlobalVisState for details.
4245 : : */
4246 : : bool
4247 : 15756937 : GlobalVisTestIsRemovableFullXid(GlobalVisState *state,
4248 : : FullTransactionId fxid,
4249 : : bool allow_update)
4250 : : {
4251 : : /*
4252 : : * If fxid is older than maybe_needed bound, it definitely is visible to
4253 : : * everyone.
4254 : : */
4255 [ + + ]: 15756937 : if (FullTransactionIdPrecedes(fxid, state->maybe_needed))
4256 : 5747401 : return true;
4257 : :
4258 : : /*
4259 : : * If fxid is >= definitely_needed bound, it is very likely to still be
4260 : : * considered running.
4261 : : */
4262 [ + + ]: 10009536 : if (FullTransactionIdFollowsOrEquals(fxid, state->definitely_needed))
4263 : 9252753 : return false;
4264 : :
4265 : : /*
4266 : : * fxid is between maybe_needed and definitely_needed, i.e. there might or
4267 : : * might not exist a snapshot considering fxid running. If it makes sense,
4268 : : * update boundaries and recheck.
4269 : : */
42 melanieplageman@gmai 4270 [ + - + + ]:GNC 756783 : if (allow_update && GlobalVisTestShouldUpdate(state))
4271 : : {
2092 andres@anarazel.de 4272 :CBC 81990 : GlobalVisUpdate();
4273 : :
4274 [ - + ]: 81990 : Assert(FullTransactionIdPrecedes(fxid, state->definitely_needed));
4275 : :
4276 : 81990 : return FullTransactionIdPrecedes(fxid, state->maybe_needed);
4277 : : }
4278 : : else
4279 : 674793 : return false;
4280 : : }
4281 : :
4282 : : /*
4283 : : * Wrapper around GlobalVisTestIsRemovableFullXid() for 32bit xids.
4284 : : *
4285 : : * It is crucial that this only gets called for xids from a source that
4286 : : * protects against xid wraparounds (e.g. from a table and thus protected by
4287 : : * relfrozenxid).
4288 : : */
4289 : : bool
42 melanieplageman@gmai 4290 :GNC 15755482 : GlobalVisTestIsRemovableXid(GlobalVisState *state, TransactionId xid,
4291 : : bool allow_update)
4292 : : {
4293 : : FullTransactionId fxid;
4294 : :
4295 : : /*
4296 : : * Convert 32 bit argument to FullTransactionId. We can do so safely
4297 : : * because we know the xid has to, at the very least, be between
4298 : : * [oldestXid, nextXid), i.e. within 2 billion of xid. To avoid taking a
4299 : : * lock to determine either, we can just compare with
4300 : : * state->definitely_needed, which was based on those value at the time
4301 : : * the current snapshot was built.
4302 : : */
2092 andres@anarazel.de 4303 :CBC 15755482 : fxid = FullXidRelativeTo(state->definitely_needed, xid);
4304 : :
42 melanieplageman@gmai 4305 :GNC 15755482 : return GlobalVisTestIsRemovableFullXid(state, fxid, allow_update);
4306 : : }
4307 : :
4308 : : /*
4309 : : * Wrapper around GlobalVisTestIsRemovableXid() for use when examining live
4310 : : * tuples. Returns true if the given XID may be considered running by at least
4311 : : * one snapshot.
4312 : : *
4313 : : * This function alone is insufficient to determine tuple visibility; callers
4314 : : * must also consider the XID's commit status. Its purpose is purely semantic:
4315 : : * when applied to live tuples, GlobalVisTestIsRemovableXid() is checking
4316 : : * whether the inserting transaction is still considered running, not whether
4317 : : * the tuple is removable. Live tuples are, by definition, not removable, but
4318 : : * the snapshot criteria for "transaction still running" are identical to
4319 : : * those used for removal XIDs.
4320 : : *
4321 : : * If allow_update is true, the GlobalVisState boundaries may be updated. If
4322 : : * it is false, they definitely will not be updated.
4323 : : *
4324 : : * See the comment above GlobalVisTestIsRemovable[Full]Xid() for details on
4325 : : * the required preconditions for calling this function.
4326 : : */
4327 : : bool
4328 : 120468 : GlobalVisTestXidConsideredRunning(GlobalVisState *state, TransactionId xid,
4329 : : bool allow_update)
4330 : : {
4331 : 120468 : return !GlobalVisTestIsRemovableXid(state, xid, allow_update);
4332 : : }
4333 : :
4334 : : /*
4335 : : * Convenience wrapper around GlobalVisTestFor() and
4336 : : * GlobalVisTestIsRemovableFullXid(), see their comments.
4337 : : */
4338 : : bool
1913 pg@bowt.ie 4339 :CBC 1455 : GlobalVisCheckRemovableFullXid(Relation rel, FullTransactionId fxid)
4340 : : {
4341 : : GlobalVisState *state;
4342 : :
2092 andres@anarazel.de 4343 : 1455 : state = GlobalVisTestFor(rel);
4344 : :
42 melanieplageman@gmai 4345 :GNC 1455 : return GlobalVisTestIsRemovableFullXid(state, fxid, true);
4346 : : }
4347 : :
4348 : : /*
4349 : : * Convenience wrapper around GlobalVisTestFor() and
4350 : : * GlobalVisTestIsRemovableXid(), see their comments.
4351 : : */
4352 : : bool
2092 andres@anarazel.de 4353 :CBC 8 : GlobalVisCheckRemovableXid(Relation rel, TransactionId xid)
4354 : : {
4355 : : GlobalVisState *state;
4356 : :
4357 : 8 : state = GlobalVisTestFor(rel);
4358 : :
42 melanieplageman@gmai 4359 :GNC 8 : return GlobalVisTestIsRemovableXid(state, xid, true);
4360 : : }
4361 : :
4362 : : /*
4363 : : * Convert a 32 bit transaction id into 64 bit transaction id, by assuming it
4364 : : * is within MaxTransactionId / 2 of XidFromFullTransactionId(rel).
4365 : : *
4366 : : * Be very careful about when to use this function. It can only safely be used
4367 : : * when there is a guarantee that xid is within MaxTransactionId / 2 xids of
4368 : : * rel. That e.g. can be guaranteed if the caller assures a snapshot is
4369 : : * held by the backend and xid is from a table (where vacuum/freezing ensures
4370 : : * the xid has to be within that range), or if xid is from the procarray and
4371 : : * prevents xid wraparound that way.
4372 : : */
4373 : : static inline FullTransactionId
2093 andres@anarazel.de 4374 :CBC 18040814 : FullXidRelativeTo(FullTransactionId rel, TransactionId xid)
4375 : : {
4376 : 18040814 : TransactionId rel_xid = XidFromFullTransactionId(rel);
4377 : :
4378 [ - + ]: 18040814 : Assert(TransactionIdIsValid(xid));
4379 [ - + ]: 18040814 : Assert(TransactionIdIsValid(rel_xid));
4380 : :
4381 : : /* not guaranteed to find issues, but likely to catch mistakes */
4382 : 18040814 : AssertTransactionIdInAllowableRange(xid);
4383 : :
4384 : 36081628 : return FullTransactionIdFromU64(U64FromFullTransactionId(rel)
4385 : 18040814 : + (int32) (xid - rel_xid));
4386 : : }
4387 : :
4388 : :
4389 : : /* ----------------------------------------------
4390 : : * KnownAssignedTransactionIds sub-module
4391 : : * ----------------------------------------------
4392 : : */
4393 : :
4394 : : /*
4395 : : * In Hot Standby mode, we maintain a list of transactions that are (or were)
4396 : : * running on the primary at the current point in WAL. These XIDs must be
4397 : : * treated as running by standby transactions, even though they are not in
4398 : : * the standby server's PGPROC array.
4399 : : *
4400 : : * We record all XIDs that we know have been assigned. That includes all the
4401 : : * XIDs seen in WAL records, plus all unobserved XIDs that we can deduce have
4402 : : * been assigned. We can deduce the existence of unobserved XIDs because we
4403 : : * know XIDs are assigned in sequence, with no gaps. The KnownAssignedXids
4404 : : * list expands as new XIDs are observed or inferred, and contracts when
4405 : : * transaction completion records arrive.
4406 : : *
4407 : : * During hot standby we do not fret too much about the distinction between
4408 : : * top-level XIDs and subtransaction XIDs. We store both together in the
4409 : : * KnownAssignedXids list. In backends, this is copied into snapshots in
4410 : : * GetSnapshotData(), taking advantage of the fact that XidInMVCCSnapshot()
4411 : : * doesn't care about the distinction either. Subtransaction XIDs are
4412 : : * effectively treated as top-level XIDs and in the typical case pg_subtrans
4413 : : * links are *not* maintained (which does not affect visibility).
4414 : : *
4415 : : * We have room in KnownAssignedXids and in snapshots to hold maxProcs *
4416 : : * (1 + PGPROC_MAX_CACHED_SUBXIDS) XIDs, so every primary transaction must
4417 : : * report its subtransaction XIDs in a WAL XLOG_XACT_ASSIGNMENT record at
4418 : : * least every PGPROC_MAX_CACHED_SUBXIDS. When we receive one of these
4419 : : * records, we mark the subXIDs as children of the top XID in pg_subtrans,
4420 : : * and then remove them from KnownAssignedXids. This prevents overflow of
4421 : : * KnownAssignedXids and snapshots, at the cost that status checks for these
4422 : : * subXIDs will take a slower path through TransactionIdIsInProgress().
4423 : : * This means that KnownAssignedXids is not necessarily complete for subXIDs,
4424 : : * though it should be complete for top-level XIDs; this is the same situation
4425 : : * that holds with respect to the PGPROC entries in normal running.
4426 : : *
4427 : : * When we throw away subXIDs from KnownAssignedXids, we need to keep track of
4428 : : * that, similarly to tracking overflow of a PGPROC's subxids array. We do
4429 : : * that by remembering the lastOverflowedXid, ie the last thrown-away subXID.
4430 : : * As long as that is within the range of interesting XIDs, we have to assume
4431 : : * that subXIDs are missing from snapshots. (Note that subXID overflow occurs
4432 : : * on primary when 65th subXID arrives, whereas on standby it occurs when 64th
4433 : : * subXID arrives - that is not an error.)
4434 : : *
4435 : : * Should a backend on primary somehow disappear before it can write an abort
4436 : : * record, then we just leave those XIDs in KnownAssignedXids. They actually
4437 : : * aborted but we think they were running; the distinction is irrelevant
4438 : : * because either way any changes done by the transaction are not visible to
4439 : : * backends in the standby. We prune KnownAssignedXids when
4440 : : * XLOG_RUNNING_XACTS arrives, to forestall possible overflow of the
4441 : : * array due to such dead XIDs.
4442 : : */
4443 : :
4444 : : /*
4445 : : * RecordKnownAssignedTransactionIds
4446 : : * Record the given XID in KnownAssignedXids, as well as any preceding
4447 : : * unobserved XIDs.
4448 : : *
4449 : : * RecordKnownAssignedTransactionIds() should be run for *every* WAL record
4450 : : * associated with a transaction. Must be called for each record after we
4451 : : * have executed StartupCLOG() et al, since we must ExtendCLOG() etc..
4452 : : *
4453 : : * Called during recovery in analogy with and in place of GetNewTransactionId()
4454 : : */
4455 : : void
5981 simon@2ndQuadrant.co 4456 : 2601051 : RecordKnownAssignedTransactionIds(TransactionId xid)
4457 : : {
5836 4458 [ - + ]: 2601051 : Assert(standbyState >= STANDBY_INITIALIZED);
5835 4459 [ - + ]: 2601051 : Assert(TransactionIdIsValid(xid));
4547 heikki.linnakangas@i 4460 [ - + ]: 2601051 : Assert(TransactionIdIsValid(latestObservedXid));
4461 : :
876 michael@paquier.xyz 4462 [ - + ]: 2601051 : elog(DEBUG4, "record known xact %u latestObservedXid %u",
4463 : : xid, latestObservedXid);
4464 : :
4465 : : /*
4466 : : * When a newly observed xid arrives, it is frequently the case that it is
4467 : : * *not* the next xid in sequence. When this occurs, we must treat the
4468 : : * intervening xids as running also.
4469 : : */
5981 simon@2ndQuadrant.co 4470 [ + + ]: 2601051 : if (TransactionIdFollows(xid, latestObservedXid))
4471 : : {
4472 : : TransactionId next_expected_xid;
4473 : :
4474 : : /*
4475 : : * Extend subtrans like we do in GetNewTransactionId() during normal
4476 : : * operation using individual extend steps. Note that we do not need
4477 : : * to extend clog since its extensions are WAL logged.
4478 : : *
4479 : : * This part has to be done regardless of standbyState since we
4480 : : * immediately start assigning subtransactions to their toplevel
4481 : : * transactions.
4482 : : */
5851 tgl@sss.pgh.pa.us 4483 : 24030 : next_expected_xid = latestObservedXid;
4547 heikki.linnakangas@i 4484 [ + + ]: 49190 : while (TransactionIdPrecedes(next_expected_xid, xid))
4485 : : {
4486 [ - + ]: 25160 : TransactionIdAdvance(next_expected_xid);
5981 simon@2ndQuadrant.co 4487 : 25160 : ExtendSUBTRANS(next_expected_xid);
4488 : : }
4547 heikki.linnakangas@i 4489 [ - + ]: 24030 : Assert(next_expected_xid == xid);
4490 : :
4491 : : /*
4492 : : * If the KnownAssignedXids machinery isn't up yet, there's nothing
4493 : : * more to do since we don't track assigned xids yet.
4494 : : */
4495 [ - + ]: 24030 : if (standbyState <= STANDBY_INITIALIZED)
4496 : : {
4547 heikki.linnakangas@i 4497 :UBC 0 : latestObservedXid = xid;
4498 : 0 : return;
4499 : : }
4500 : :
4501 : : /*
4502 : : * Add (latestObservedXid, xid] onto the KnownAssignedXids array.
4503 : : */
5851 tgl@sss.pgh.pa.us 4504 :CBC 24030 : next_expected_xid = latestObservedXid;
4505 [ - + ]: 24030 : TransactionIdAdvance(next_expected_xid);
4506 : 24030 : KnownAssignedXidsAdd(next_expected_xid, xid, false);
4507 : :
4508 : : /*
4509 : : * Now we can advance latestObservedXid
4510 : : */
5981 simon@2ndQuadrant.co 4511 : 24030 : latestObservedXid = xid;
4512 : :
4513 : : /* TransamVariables->nextXid must be beyond any observed xid */
2595 tmunro@postgresql.or 4514 : 24030 : AdvanceNextFullTransactionIdPastXid(latestObservedXid);
4515 : : }
4516 : : }
4517 : :
4518 : : /*
4519 : : * ExpireTreeKnownAssignedTransactionIds
4520 : : * Remove the given XIDs from KnownAssignedXids.
4521 : : *
4522 : : * Called during recovery in analogy with and in place of ProcArrayEndTransaction()
4523 : : */
4524 : : void
5981 simon@2ndQuadrant.co 4525 : 23630 : ExpireTreeKnownAssignedTransactionIds(TransactionId xid, int nsubxids,
4526 : : TransactionId *subxids, TransactionId max_xid)
4527 : : {
5836 4528 [ - + ]: 23630 : Assert(standbyState >= STANDBY_INITIALIZED);
4529 : :
4530 : : /*
4531 : : * Uses same locking as transaction commit
4532 : : */
5981 4533 : 23630 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
4534 : :
5851 tgl@sss.pgh.pa.us 4535 : 23630 : KnownAssignedXidsRemoveTree(xid, nsubxids, subxids);
4536 : :
4537 : : /* As in ProcArrayEndTransaction, advance latestCompletedXid */
2093 andres@anarazel.de 4538 : 23630 : MaintainLatestCompletedXidRecovery(max_xid);
4539 : :
4540 : : /* ... and xactCompletionCount */
879 heikki.linnakangas@i 4541 : 23630 : TransamVariables->xactCompletionCount++;
4542 : :
5981 simon@2ndQuadrant.co 4543 : 23630 : LWLockRelease(ProcArrayLock);
4544 : 23630 : }
4545 : :
4546 : : /*
4547 : : * ExpireAllKnownAssignedTransactionIds
4548 : : * Remove all entries in KnownAssignedXids and reset lastOverflowedXid.
4549 : : */
4550 : : void
4551 : 121 : ExpireAllKnownAssignedTransactionIds(void)
4552 : : {
4553 : : FullTransactionId latestXid;
4554 : :
4555 : 121 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
5851 tgl@sss.pgh.pa.us 4556 : 121 : KnownAssignedXidsRemovePreceding(InvalidTransactionId);
4557 : :
4558 : : /* Reset latestCompletedXid to nextXid - 1 */
408 heikki.linnakangas@i 4559 [ - + ]: 121 : Assert(FullTransactionIdIsValid(TransamVariables->nextXid));
4560 : 121 : latestXid = TransamVariables->nextXid;
4561 : 121 : FullTransactionIdRetreat(&latestXid);
4562 : 121 : TransamVariables->latestCompletedXid = latestXid;
4563 : :
4564 : : /*
4565 : : * Any transactions that were in-progress were effectively aborted, so
4566 : : * advance xactCompletionCount.
4567 : : */
4568 : 121 : TransamVariables->xactCompletionCount++;
4569 : :
4570 : : /*
4571 : : * Reset lastOverflowedXid. Currently, lastOverflowedXid has no use after
4572 : : * the call of this function. But do this for unification with what
4573 : : * ExpireOldKnownAssignedTransactionIds() do.
4574 : : */
1641 akorotkov@postgresql 4575 : 121 : procArray->lastOverflowedXid = InvalidTransactionId;
5981 simon@2ndQuadrant.co 4576 : 121 : LWLockRelease(ProcArrayLock);
4577 : 121 : }
4578 : :
4579 : : /*
4580 : : * ExpireOldKnownAssignedTransactionIds
4581 : : * Remove KnownAssignedXids entries preceding the given XID and
4582 : : * potentially reset lastOverflowedXid.
4583 : : */
4584 : : void
4585 : 843 : ExpireOldKnownAssignedTransactionIds(TransactionId xid)
4586 : : {
4587 : : TransactionId latestXid;
4588 : :
4589 : 843 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
4590 : :
4591 : : /* As in ProcArrayEndTransaction, advance latestCompletedXid */
408 heikki.linnakangas@i 4592 : 843 : latestXid = xid;
4593 [ - + ]: 843 : TransactionIdRetreat(latestXid);
4594 : 843 : MaintainLatestCompletedXidRecovery(latestXid);
4595 : :
4596 : : /* ... and xactCompletionCount */
4597 : 843 : TransamVariables->xactCompletionCount++;
4598 : :
4599 : : /*
4600 : : * Reset lastOverflowedXid if we know all transactions that have been
4601 : : * possibly running are being gone. Not doing so could cause an incorrect
4602 : : * lastOverflowedXid value, which makes extra snapshots be marked as
4603 : : * suboverflowed.
4604 : : */
1641 akorotkov@postgresql 4605 [ + + ]: 843 : if (TransactionIdPrecedes(procArray->lastOverflowedXid, xid))
4606 : 836 : procArray->lastOverflowedXid = InvalidTransactionId;
5851 tgl@sss.pgh.pa.us 4607 : 843 : KnownAssignedXidsRemovePreceding(xid);
5981 simon@2ndQuadrant.co 4608 : 843 : LWLockRelease(ProcArrayLock);
4609 : 843 : }
4610 : :
4611 : : /*
4612 : : * KnownAssignedTransactionIdsIdleMaintenance
4613 : : * Opportunistically do maintenance work when the startup process
4614 : : * is about to go idle.
4615 : : */
4616 : : void
1253 tgl@sss.pgh.pa.us 4617 : 14658 : KnownAssignedTransactionIdsIdleMaintenance(void)
4618 : : {
4619 : 14658 : KnownAssignedXidsCompress(KAX_STARTUP_PROCESS_IDLE, false);
4620 : 14658 : }
4621 : :
4622 : :
4623 : : /*
4624 : : * Private module functions to manipulate KnownAssignedXids
4625 : : *
4626 : : * There are 5 main uses of the KnownAssignedXids data structure:
4627 : : *
4628 : : * * backends taking snapshots - all valid XIDs need to be copied out
4629 : : * * backends seeking to determine presence of a specific XID
4630 : : * * startup process adding new known-assigned XIDs
4631 : : * * startup process removing specific XIDs as transactions end
4632 : : * * startup process pruning array when special WAL records arrive
4633 : : *
4634 : : * This data structure is known to be a hot spot during Hot Standby, so we
4635 : : * go to some lengths to make these operations as efficient and as concurrent
4636 : : * as possible.
4637 : : *
4638 : : * The XIDs are stored in an array in sorted order --- TransactionIdPrecedes
4639 : : * order, to be exact --- to allow binary search for specific XIDs. Note:
4640 : : * in general TransactionIdPrecedes would not provide a total order, but
4641 : : * we know that the entries present at any instant should not extend across
4642 : : * a large enough fraction of XID space to wrap around (the primary would
4643 : : * shut down for fear of XID wrap long before that happens). So it's OK to
4644 : : * use TransactionIdPrecedes as a binary-search comparator.
4645 : : *
4646 : : * It's cheap to maintain the sortedness during insertions, since new known
4647 : : * XIDs are always reported in XID order; we just append them at the right.
4648 : : *
4649 : : * To keep individual deletions cheap, we need to allow gaps in the array.
4650 : : * This is implemented by marking array elements as valid or invalid using
4651 : : * the parallel boolean array KnownAssignedXidsValid[]. A deletion is done
4652 : : * by setting KnownAssignedXidsValid[i] to false, *without* clearing the
4653 : : * XID entry itself. This preserves the property that the XID entries are
4654 : : * sorted, so we can do binary searches easily. Periodically we compress
4655 : : * out the unused entries; that's much cheaper than having to compress the
4656 : : * array immediately on every deletion.
4657 : : *
4658 : : * The actually valid items in KnownAssignedXids[] and KnownAssignedXidsValid[]
4659 : : * are those with indexes tail <= i < head; items outside this subscript range
4660 : : * have unspecified contents. When head reaches the end of the array, we
4661 : : * force compression of unused entries rather than wrapping around, since
4662 : : * allowing wraparound would greatly complicate the search logic. We maintain
4663 : : * an explicit tail pointer so that pruning of old XIDs can be done without
4664 : : * immediately moving the array contents. In most cases only a small fraction
4665 : : * of the array contains valid entries at any instant.
4666 : : *
4667 : : * Although only the startup process can ever change the KnownAssignedXids
4668 : : * data structure, we still need interlocking so that standby backends will
4669 : : * not observe invalid intermediate states. The convention is that backends
4670 : : * must hold shared ProcArrayLock to examine the array. To remove XIDs from
4671 : : * the array, the startup process must hold ProcArrayLock exclusively, for
4672 : : * the usual transactional reasons (compare commit/abort of a transaction
4673 : : * during normal running). Compressing unused entries out of the array
4674 : : * likewise requires exclusive lock. To add XIDs to the array, we just insert
4675 : : * them into slots to the right of the head pointer and then advance the head
4676 : : * pointer. This doesn't require any lock at all, but on machines with weak
4677 : : * memory ordering, we need to be careful that other processors see the array
4678 : : * element changes before they see the head pointer change. We handle this by
4679 : : * using memory barriers when reading or writing the head/tail pointers (unless
4680 : : * the caller holds ProcArrayLock exclusively).
4681 : : *
4682 : : * Algorithmic analysis:
4683 : : *
4684 : : * If we have a maximum of M slots, with N XIDs currently spread across
4685 : : * S elements then we have N <= S <= M always.
4686 : : *
4687 : : * * Adding a new XID is O(1) and needs no lock (unless compression must
4688 : : * happen)
4689 : : * * Compressing the array is O(S) and requires exclusive lock
4690 : : * * Removing an XID is O(logS) and requires exclusive lock
4691 : : * * Taking a snapshot is O(S) and requires shared lock
4692 : : * * Checking for an XID is O(logS) and requires shared lock
4693 : : *
4694 : : * In comparison, using a hash table for KnownAssignedXids would mean that
4695 : : * taking snapshots would be O(M). If we can maintain S << M then the
4696 : : * sorted array technique will deliver significantly faster snapshots.
4697 : : * If we try to keep S too small then we will spend too much time compressing,
4698 : : * so there is an optimal point for any workload mix. We use a heuristic to
4699 : : * decide when to compress the array, though trimming also helps reduce
4700 : : * frequency of compressing. The heuristic requires us to track the number of
4701 : : * currently valid XIDs in the array (N). Except in special cases, we'll
4702 : : * compress when S >= 2N. Bounding S at 2N in turn bounds the time for
4703 : : * taking a snapshot to be O(N), which it would have to be anyway.
4704 : : */
4705 : :
4706 : :
4707 : : /*
4708 : : * Compress KnownAssignedXids by shifting valid data down to the start of the
4709 : : * array, removing any gaps.
4710 : : *
4711 : : * A compression step is forced if "reason" is KAX_NO_SPACE, otherwise
4712 : : * we do it only if a heuristic indicates it's a good time to do it.
4713 : : *
4714 : : * Compression requires holding ProcArrayLock in exclusive mode.
4715 : : * Caller must pass haveLock = true if it already holds the lock.
4716 : : */
4717 : : static void
4718 : 39152 : KnownAssignedXidsCompress(KAXCompressReason reason, bool haveLock)
4719 : : {
2734 andres@anarazel.de 4720 : 39152 : ProcArrayStruct *pArray = procArray;
4721 : : int head,
4722 : : tail,
4723 : : nelements;
4724 : : int compress_index;
4725 : : int i;
4726 : :
4727 : : /* Counters for compression heuristics */
4728 : : static unsigned int transactionEndsCounter;
4729 : : static TimestampTz lastCompressTs;
4730 : :
4731 : : /* Tuning constants */
4732 : : #define KAX_COMPRESS_FREQUENCY 128 /* in transactions */
4733 : : #define KAX_COMPRESS_IDLE_INTERVAL 1000 /* in ms */
4734 : :
4735 : : /*
4736 : : * Since only the startup process modifies the head/tail pointers, we
4737 : : * don't need a lock to read them here.
4738 : : */
5851 tgl@sss.pgh.pa.us 4739 : 39152 : head = pArray->headKnownAssignedXids;
4740 : 39152 : tail = pArray->tailKnownAssignedXids;
1253 4741 : 39152 : nelements = head - tail;
4742 : :
4743 : : /*
4744 : : * If we can choose whether to compress, use a heuristic to avoid
4745 : : * compressing too often or not often enough. "Compress" here simply
4746 : : * means moving the values to the beginning of the array, so it is not as
4747 : : * complex or costly as typical data compression algorithms.
4748 : : */
4749 [ + + ]: 39152 : if (nelements == pArray->numKnownAssignedXids)
4750 : : {
4751 : : /*
4752 : : * When there are no gaps between head and tail, don't bother to
4753 : : * compress, except in the KAX_NO_SPACE case where we must compress to
4754 : : * create some space after the head.
4755 : : */
4756 [ + - ]: 18531 : if (reason != KAX_NO_SPACE)
4757 : 18531 : return;
4758 : : }
4759 [ + + ]: 20621 : else if (reason == KAX_TRANSACTION_END)
4760 : : {
4761 : : /*
4762 : : * Consider compressing only once every so many commits. Frequency
4763 : : * determined by benchmarks.
4764 : : */
4765 [ + + ]: 14933 : if ((transactionEndsCounter++) % KAX_COMPRESS_FREQUENCY != 0)
4766 : 14807 : return;
4767 : :
4768 : : /*
4769 : : * Furthermore, compress only if the used part of the array is less
4770 : : * than 50% full (see comments above).
4771 : : */
4772 [ + + ]: 126 : if (nelements < 2 * pArray->numKnownAssignedXids)
5851 4773 : 7 : return;
4774 : : }
1253 4775 [ + + ]: 5688 : else if (reason == KAX_STARTUP_PROCESS_IDLE)
4776 : : {
4777 : : /*
4778 : : * We're about to go idle for lack of new WAL, so we might as well
4779 : : * compress. But not too often, to avoid ProcArray lock contention
4780 : : * with readers.
4781 : : */
4782 [ + + ]: 5559 : if (lastCompressTs != 0)
4783 : : {
4784 : : TimestampTz compress_after;
4785 : :
4786 : 5558 : compress_after = TimestampTzPlusMilliseconds(lastCompressTs,
4787 : : KAX_COMPRESS_IDLE_INTERVAL);
4788 [ + + ]: 5558 : if (GetCurrentTimestamp() < compress_after)
4789 : 5517 : return;
4790 : : }
4791 : : }
4792 : :
4793 : : /* Need to compress, so get the lock if we don't have it. */
4794 [ + + ]: 290 : if (!haveLock)
4795 : 42 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
4796 : :
4797 : : /*
4798 : : * We compress the array by reading the valid values from tail to head,
4799 : : * re-aligning data to 0th element.
4800 : : */
5851 4801 : 290 : compress_index = 0;
4802 [ + + ]: 10735 : for (i = tail; i < head; i++)
4803 : : {
4804 [ + + ]: 10445 : if (KnownAssignedXidsValid[i])
4805 : : {
4806 : 1219 : KnownAssignedXids[compress_index] = KnownAssignedXids[i];
4807 : 1219 : KnownAssignedXidsValid[compress_index] = true;
4808 : 1219 : compress_index++;
4809 : : }
4810 : : }
1253 4811 [ - + ]: 290 : Assert(compress_index == pArray->numKnownAssignedXids);
4812 : :
5851 4813 : 290 : pArray->tailKnownAssignedXids = 0;
4814 : 290 : pArray->headKnownAssignedXids = compress_index;
4815 : :
1253 4816 [ + + ]: 290 : if (!haveLock)
4817 : 42 : LWLockRelease(ProcArrayLock);
4818 : :
4819 : : /* Update timestamp for maintenance. No need to hold lock for this. */
4820 : 290 : lastCompressTs = GetCurrentTimestamp();
4821 : : }
4822 : :
4823 : : /*
4824 : : * Add xids into KnownAssignedXids at the head of the array.
4825 : : *
4826 : : * xids from from_xid to to_xid, inclusive, are added to the array.
4827 : : *
4828 : : * If exclusive_lock is true then caller already holds ProcArrayLock in
4829 : : * exclusive mode, so we need no extra locking here. Else caller holds no
4830 : : * lock, so we need to be sure we maintain sufficient interlocks against
4831 : : * concurrent readers. (Only the startup process ever calls this, so no need
4832 : : * to worry about concurrent writers.)
4833 : : */
4834 : : static void
5851 4835 : 24035 : KnownAssignedXidsAdd(TransactionId from_xid, TransactionId to_xid,
4836 : : bool exclusive_lock)
4837 : : {
2734 andres@anarazel.de 4838 : 24035 : ProcArrayStruct *pArray = procArray;
4839 : : TransactionId next_xid;
4840 : : int head,
4841 : : tail;
4842 : : int nxids;
4843 : : int i;
4844 : :
5851 tgl@sss.pgh.pa.us 4845 [ - + ]: 24035 : Assert(TransactionIdPrecedesOrEquals(from_xid, to_xid));
4846 : :
4847 : : /*
4848 : : * Calculate how many array slots we'll need. Normally this is cheap; in
4849 : : * the unusual case where the XIDs cross the wrap point, we do it the hard
4850 : : * way.
4851 : : */
4852 [ + - ]: 24035 : if (to_xid >= from_xid)
4853 : 24035 : nxids = to_xid - from_xid + 1;
4854 : : else
4855 : : {
5851 tgl@sss.pgh.pa.us 4856 :UBC 0 : nxids = 1;
4857 : 0 : next_xid = from_xid;
4858 [ # # ]: 0 : while (TransactionIdPrecedes(next_xid, to_xid))
4859 : : {
4860 : 0 : nxids++;
4861 [ # # ]: 0 : TransactionIdAdvance(next_xid);
4862 : : }
4863 : : }
4864 : :
4865 : : /*
4866 : : * Since only the startup process modifies the head/tail pointers, we
4867 : : * don't need a lock to read them here.
4868 : : */
5851 tgl@sss.pgh.pa.us 4869 :CBC 24035 : head = pArray->headKnownAssignedXids;
4870 : 24035 : tail = pArray->tailKnownAssignedXids;
4871 : :
4872 [ + - - + ]: 24035 : Assert(head >= 0 && head <= pArray->maxKnownAssignedXids);
4873 [ + - - + ]: 24035 : Assert(tail >= 0 && tail < pArray->maxKnownAssignedXids);
4874 : :
4875 : : /*
4876 : : * Verify that insertions occur in TransactionId sequence. Note that even
4877 : : * if the last existing element is marked invalid, it must still have a
4878 : : * correctly sequenced XID value.
4879 : : */
4880 [ + + - + ]: 40066 : if (head > tail &&
4881 : 16031 : TransactionIdFollowsOrEquals(KnownAssignedXids[head - 1], from_xid))
4882 : : {
5851 tgl@sss.pgh.pa.us 4883 :UBC 0 : KnownAssignedXidsDisplay(LOG);
4884 [ # # ]: 0 : elog(ERROR, "out-of-order XID insertion in KnownAssignedXids");
4885 : : }
4886 : :
4887 : : /*
4888 : : * If our xids won't fit in the remaining space, compress out free space
4889 : : */
5851 tgl@sss.pgh.pa.us 4890 [ - + ]:CBC 24035 : if (head + nxids > pArray->maxKnownAssignedXids)
4891 : : {
1253 tgl@sss.pgh.pa.us 4892 :UBC 0 : KnownAssignedXidsCompress(KAX_NO_SPACE, exclusive_lock);
4893 : :
5851 4894 : 0 : head = pArray->headKnownAssignedXids;
4895 : : /* note: we no longer care about the tail pointer */
4896 : :
4897 : : /*
4898 : : * If it still won't fit then we're out of memory
4899 : : */
4900 [ # # ]: 0 : if (head + nxids > pArray->maxKnownAssignedXids)
2222 peter@eisentraut.org 4901 [ # # ]: 0 : elog(ERROR, "too many KnownAssignedXids");
4902 : : }
4903 : :
4904 : : /* Now we can insert the xids into the space starting at head */
5851 tgl@sss.pgh.pa.us 4905 :CBC 24035 : next_xid = from_xid;
4906 [ + + ]: 49200 : for (i = 0; i < nxids; i++)
4907 : : {
4908 : 25165 : KnownAssignedXids[head] = next_xid;
4909 : 25165 : KnownAssignedXidsValid[head] = true;
4910 [ - + ]: 25165 : TransactionIdAdvance(next_xid);
4911 : 25165 : head++;
4912 : : }
4913 : :
4914 : : /* Adjust count of number of valid entries */
4915 : 24035 : pArray->numKnownAssignedXids += nxids;
4916 : :
4917 : : /*
4918 : : * Now update the head pointer. We use a write barrier to ensure that
4919 : : * other processors see the above array updates before they see the head
4920 : : * pointer change. The barrier isn't required if we're holding
4921 : : * ProcArrayLock exclusively.
4922 : : */
973 nathan@postgresql.or 4923 [ + + ]: 24035 : if (!exclusive_lock)
4924 : 24030 : pg_write_barrier();
4925 : :
4926 : 24035 : pArray->headKnownAssignedXids = head;
5851 tgl@sss.pgh.pa.us 4927 : 24035 : }
4928 : :
4929 : : /*
4930 : : * KnownAssignedXidsSearch
4931 : : *
4932 : : * Searches KnownAssignedXids for a specific xid and optionally removes it.
4933 : : * Returns true if it was found, false if not.
4934 : : *
4935 : : * Caller must hold ProcArrayLock in shared or exclusive mode.
4936 : : * Exclusive lock must be held for remove = true.
4937 : : */
4938 : : static bool
4939 : 26299 : KnownAssignedXidsSearch(TransactionId xid, bool remove)
4940 : : {
2734 andres@anarazel.de 4941 : 26299 : ProcArrayStruct *pArray = procArray;
4942 : : int first,
4943 : : last;
4944 : : int head;
4945 : : int tail;
5782 bruce@momjian.us 4946 : 26299 : int result_index = -1;
4947 : :
973 nathan@postgresql.or 4948 : 26299 : tail = pArray->tailKnownAssignedXids;
4949 : 26299 : head = pArray->headKnownAssignedXids;
4950 : :
4951 : : /*
4952 : : * Only the startup process removes entries, so we don't need the read
4953 : : * barrier in that case.
4954 : : */
4955 [ - + ]: 26299 : if (!remove)
973 nathan@postgresql.or 4956 :LBC (1) : pg_read_barrier(); /* pairs with KnownAssignedXidsAdd */
4957 : :
4958 : : /*
4959 : : * Standard binary search. Note we can ignore the KnownAssignedXidsValid
4960 : : * array here, since even invalid entries will contain sorted XIDs.
4961 : : */
5851 tgl@sss.pgh.pa.us 4962 :CBC 26299 : first = tail;
4963 : 26299 : last = head - 1;
4964 [ + + ]: 91556 : while (first <= last)
4965 : : {
4966 : : int mid_index;
4967 : : TransactionId mid_xid;
4968 : :
4969 : 90344 : mid_index = (first + last) / 2;
4970 : 90344 : mid_xid = KnownAssignedXids[mid_index];
4971 : :
4972 [ + + ]: 90344 : if (xid == mid_xid)
4973 : : {
4974 : 25087 : result_index = mid_index;
4975 : 25087 : break;
4976 : : }
4977 [ + + ]: 65257 : else if (TransactionIdPrecedes(xid, mid_xid))
4978 : 14031 : last = mid_index - 1;
4979 : : else
4980 : 51226 : first = mid_index + 1;
4981 : : }
4982 : :
4983 [ + + ]: 26299 : if (result_index < 0)
4984 : 1212 : return false; /* not in array */
4985 : :
4986 [ + + ]: 25087 : if (!KnownAssignedXidsValid[result_index])
4987 : 8 : return false; /* in array, but invalid */
4988 : :
4989 [ + - ]: 25079 : if (remove)
4990 : : {
4991 : 25079 : KnownAssignedXidsValid[result_index] = false;
4992 : :
4993 : 25079 : pArray->numKnownAssignedXids--;
4994 [ - + ]: 25079 : Assert(pArray->numKnownAssignedXids >= 0);
4995 : :
4996 : : /*
4997 : : * If we're removing the tail element then advance tail pointer over
4998 : : * any invalid elements. This will speed future searches.
4999 : : */
5000 [ + + ]: 25079 : if (result_index == tail)
5001 : : {
5002 : 9279 : tail++;
5003 [ + + + + ]: 15853 : while (tail < head && !KnownAssignedXidsValid[tail])
5004 : 6574 : tail++;
5005 [ + + ]: 9279 : if (tail >= head)
5006 : : {
5007 : : /* Array is empty, so we can reset both pointers */
5008 : 7994 : pArray->headKnownAssignedXids = 0;
5009 : 7994 : pArray->tailKnownAssignedXids = 0;
5010 : : }
5011 : : else
5012 : : {
5013 : 1285 : pArray->tailKnownAssignedXids = tail;
5014 : : }
5015 : : }
5016 : : }
5017 : :
5018 : 25079 : return true;
5019 : : }
5020 : :
5021 : : /*
5022 : : * Is the specified XID present in KnownAssignedXids[]?
5023 : : *
5024 : : * Caller must hold ProcArrayLock in shared or exclusive mode.
5025 : : */
5026 : : static bool
5851 tgl@sss.pgh.pa.us 5027 :LBC (1) : KnownAssignedXidExists(TransactionId xid)
5028 : : {
5029 [ # # ]: (1) : Assert(TransactionIdIsValid(xid));
5030 : :
5031 : (1) : return KnownAssignedXidsSearch(xid, false);
5032 : : }
5033 : :
5034 : : /*
5035 : : * Remove the specified XID from KnownAssignedXids[].
5036 : : *
5037 : : * Caller must hold ProcArrayLock in exclusive mode.
5038 : : */
5039 : : static void
5981 simon@2ndQuadrant.co 5040 :CBC 26299 : KnownAssignedXidsRemove(TransactionId xid)
5041 : : {
5042 [ - + ]: 26299 : Assert(TransactionIdIsValid(xid));
5043 : :
876 michael@paquier.xyz 5044 [ - + ]: 26299 : elog(DEBUG4, "remove KnownAssignedXid %u", xid);
5045 : :
5046 : : /*
5047 : : * Note: we cannot consider it an error to remove an XID that's not
5048 : : * present. We intentionally remove subxact IDs while processing
5049 : : * XLOG_XACT_ASSIGNMENT, to avoid array overflow. Then those XIDs will be
5050 : : * removed again when the top-level xact commits or aborts.
5051 : : *
5052 : : * It might be possible to track such XIDs to distinguish this case from
5053 : : * actual errors, but it would be complicated and probably not worth it.
5054 : : * So, just ignore the search result.
5055 : : */
5851 tgl@sss.pgh.pa.us 5056 : 26299 : (void) KnownAssignedXidsSearch(xid, true);
5981 simon@2ndQuadrant.co 5057 : 26299 : }
5058 : :
5059 : : /*
5060 : : * KnownAssignedXidsRemoveTree
5061 : : * Remove xid (if it's not InvalidTransactionId) and all the subxids.
5062 : : *
5063 : : * Caller must hold ProcArrayLock in exclusive mode.
5064 : : */
5065 : : static void
5851 tgl@sss.pgh.pa.us 5066 : 23651 : KnownAssignedXidsRemoveTree(TransactionId xid, int nsubxids,
5067 : : TransactionId *subxids)
5068 : : {
5069 : : int i;
5070 : :
5071 [ + + ]: 23651 : if (TransactionIdIsValid(xid))
5072 : 23630 : KnownAssignedXidsRemove(xid);
5073 : :
5074 [ + + ]: 26320 : for (i = 0; i < nsubxids; i++)
5075 : 2669 : KnownAssignedXidsRemove(subxids[i]);
5076 : :
5077 : : /* Opportunistically compress the array */
1253 5078 : 23651 : KnownAssignedXidsCompress(KAX_TRANSACTION_END, true);
5981 simon@2ndQuadrant.co 5079 : 23651 : }
5080 : :
5081 : : /*
5082 : : * Prune KnownAssignedXids up to, but *not* including xid. If xid is invalid
5083 : : * then clear the whole table.
5084 : : *
5085 : : * Caller must hold ProcArrayLock in exclusive mode.
5086 : : */
5087 : : static void
5851 tgl@sss.pgh.pa.us 5088 : 964 : KnownAssignedXidsRemovePreceding(TransactionId removeXid)
5089 : : {
2734 andres@anarazel.de 5090 : 964 : ProcArrayStruct *pArray = procArray;
5782 bruce@momjian.us 5091 : 964 : int count = 0;
5092 : : int head,
5093 : : tail,
5094 : : i;
5095 : :
5851 tgl@sss.pgh.pa.us 5096 [ + + ]: 964 : if (!TransactionIdIsValid(removeXid))
5097 : : {
876 michael@paquier.xyz 5098 [ - + ]: 121 : elog(DEBUG4, "removing all KnownAssignedXids");
5851 tgl@sss.pgh.pa.us 5099 : 121 : pArray->numKnownAssignedXids = 0;
5100 : 121 : pArray->headKnownAssignedXids = pArray->tailKnownAssignedXids = 0;
5101 : 121 : return;
5102 : : }
5103 : :
876 michael@paquier.xyz 5104 [ - + ]: 843 : elog(DEBUG4, "prune KnownAssignedXids to %u", removeXid);
5105 : :
5106 : : /*
5107 : : * Mark entries invalid starting at the tail. Since array is sorted, we
5108 : : * can stop as soon as we reach an entry >= removeXid.
5109 : : */
5851 tgl@sss.pgh.pa.us 5110 : 843 : tail = pArray->tailKnownAssignedXids;
5111 : 843 : head = pArray->headKnownAssignedXids;
5112 : :
5113 [ + + ]: 843 : for (i = tail; i < head; i++)
5114 : : {
5115 [ + - ]: 213 : if (KnownAssignedXidsValid[i])
5116 : : {
5782 bruce@momjian.us 5117 : 213 : TransactionId knownXid = KnownAssignedXids[i];
5118 : :
5851 tgl@sss.pgh.pa.us 5119 [ + - ]: 213 : if (TransactionIdFollowsOrEquals(knownXid, removeXid))
5120 : 213 : break;
5121 : :
5851 tgl@sss.pgh.pa.us 5122 [ # # ]:UBC 0 : if (!StandbyTransactionIdIsPrepared(knownXid))
5123 : : {
5124 : 0 : KnownAssignedXidsValid[i] = false;
5125 : 0 : count++;
5126 : : }
5127 : : }
5128 : : }
5129 : :
5851 tgl@sss.pgh.pa.us 5130 :CBC 843 : pArray->numKnownAssignedXids -= count;
5131 [ - + ]: 843 : Assert(pArray->numKnownAssignedXids >= 0);
5132 : :
5133 : : /*
5134 : : * Advance the tail pointer if we've marked the tail item invalid.
5135 : : */
5136 [ + + ]: 843 : for (i = tail; i < head; i++)
5137 : : {
5138 [ + - ]: 213 : if (KnownAssignedXidsValid[i])
5139 : 213 : break;
5140 : : }
5141 [ + + ]: 843 : if (i >= head)
5142 : : {
5143 : : /* Array is empty, so we can reset both pointers */
5144 : 630 : pArray->headKnownAssignedXids = 0;
5145 : 630 : pArray->tailKnownAssignedXids = 0;
5146 : : }
5147 : : else
5148 : : {
5149 : 213 : pArray->tailKnownAssignedXids = i;
5150 : : }
5151 : :
5152 : : /* Opportunistically compress the array */
1253 5153 : 843 : KnownAssignedXidsCompress(KAX_PRUNE, true);
5154 : : }
5155 : :
5156 : : /*
5157 : : * KnownAssignedXidsGet - Get an array of xids by scanning KnownAssignedXids.
5158 : : * We filter out anything >= xmax.
5159 : : *
5160 : : * Returns the number of XIDs stored into xarray[]. Caller is responsible
5161 : : * that array is large enough.
5162 : : *
5163 : : * Caller must hold ProcArrayLock in (at least) shared mode.
5164 : : */
5165 : : static int
5851 tgl@sss.pgh.pa.us 5166 :UBC 0 : KnownAssignedXidsGet(TransactionId *xarray, TransactionId xmax)
5167 : : {
5168 : 0 : TransactionId xtmp = InvalidTransactionId;
5169 : :
5170 : 0 : return KnownAssignedXidsGetAndSetXmin(xarray, &xtmp, xmax);
5171 : : }
5172 : :
5173 : : /*
5174 : : * KnownAssignedXidsGetAndSetXmin - as KnownAssignedXidsGet, plus
5175 : : * we reduce *xmin to the lowest xid value seen if not already lower.
5176 : : *
5177 : : * Caller must hold ProcArrayLock in (at least) shared mode.
5178 : : */
5179 : : static int
5851 tgl@sss.pgh.pa.us 5180 :CBC 1549 : KnownAssignedXidsGetAndSetXmin(TransactionId *xarray, TransactionId *xmin,
5181 : : TransactionId xmax)
5182 : : {
5183 : 1549 : int count = 0;
5184 : : int head,
5185 : : tail;
5186 : : int i;
5187 : :
5188 : : /*
5189 : : * Fetch head just once, since it may change while we loop. We can stop
5190 : : * once we reach the initially seen head, since we are certain that an xid
5191 : : * cannot enter and then leave the array while we hold ProcArrayLock. We
5192 : : * might miss newly-added xids, but they should be >= xmax so irrelevant
5193 : : * anyway.
5194 : : */
3854 rhaas@postgresql.org 5195 : 1549 : tail = procArray->tailKnownAssignedXids;
5196 : 1549 : head = procArray->headKnownAssignedXids;
5197 : :
973 nathan@postgresql.or 5198 : 1549 : pg_read_barrier(); /* pairs with KnownAssignedXidsAdd */
5199 : :
5851 tgl@sss.pgh.pa.us 5200 [ + + ]: 1580 : for (i = tail; i < head; i++)
5201 : : {
5202 : : /* Skip any gaps in the array */
5203 [ + + ]: 169 : if (KnownAssignedXidsValid[i])
5204 : : {
5205 : 156 : TransactionId knownXid = KnownAssignedXids[i];
5206 : :
5207 : : /*
5208 : : * Update xmin if required. Only the first XID need be checked,
5209 : : * since the array is sorted.
5210 : : */
5211 [ + - + + ]: 312 : if (count == 0 &&
5212 : 156 : TransactionIdPrecedes(knownXid, *xmin))
5213 : 18 : *xmin = knownXid;
5214 : :
5215 : : /*
5216 : : * Filter out anything >= xmax, again relying on sorted property
5217 : : * of array.
5218 : : */
5836 simon@2ndQuadrant.co 5219 [ + - + + ]: 312 : if (TransactionIdIsValid(xmax) &&
5220 : 156 : TransactionIdFollowsOrEquals(knownXid, xmax))
5851 tgl@sss.pgh.pa.us 5221 : 138 : break;
5222 : :
5223 : : /* Add knownXid into output array */
5224 : 18 : xarray[count++] = knownXid;
5225 : : }
5226 : : }
5227 : :
5228 : 1549 : return count;
5229 : : }
5230 : :
5231 : : /*
5232 : : * Get oldest XID in the KnownAssignedXids array, or InvalidTransactionId
5233 : : * if nothing there.
5234 : : */
5235 : : static TransactionId
5727 simon@2ndQuadrant.co 5236 : 382 : KnownAssignedXidsGetOldestXmin(void)
5237 : : {
5238 : : int head,
5239 : : tail;
5240 : : int i;
5241 : :
5242 : : /*
5243 : : * Fetch head just once, since it may change while we loop.
5244 : : */
3854 rhaas@postgresql.org 5245 : 382 : tail = procArray->tailKnownAssignedXids;
5246 : 382 : head = procArray->headKnownAssignedXids;
5247 : :
973 nathan@postgresql.or 5248 : 382 : pg_read_barrier(); /* pairs with KnownAssignedXidsAdd */
5249 : :
5727 simon@2ndQuadrant.co 5250 [ + + ]: 382 : for (i = tail; i < head; i++)
5251 : : {
5252 : : /* Skip any gaps in the array */
5253 [ + - ]: 155 : if (KnownAssignedXidsValid[i])
5254 : 155 : return KnownAssignedXids[i];
5255 : : }
5256 : :
5257 : 227 : return InvalidTransactionId;
5258 : : }
5259 : :
5260 : : /*
5261 : : * Display KnownAssignedXids to provide debug trail
5262 : : *
5263 : : * Currently this is only called within startup process, so we need no
5264 : : * special locking.
5265 : : *
5266 : : * Note this is pretty expensive, and much of the expense will be incurred
5267 : : * even if the elog message will get discarded. It's not currently called
5268 : : * in any performance-critical places, however, so no need to be tenser.
5269 : : */
5270 : : static void
5981 5271 : 126 : KnownAssignedXidsDisplay(int trace_level)
5272 : : {
2734 andres@anarazel.de 5273 : 126 : ProcArrayStruct *pArray = procArray;
5274 : : StringInfoData buf;
5275 : : int head,
5276 : : tail,
5277 : : i;
5782 bruce@momjian.us 5278 : 126 : int nxids = 0;
5279 : :
5851 tgl@sss.pgh.pa.us 5280 : 126 : tail = pArray->tailKnownAssignedXids;
5281 : 126 : head = pArray->headKnownAssignedXids;
5282 : :
5981 simon@2ndQuadrant.co 5283 : 126 : initStringInfo(&buf);
5284 : :
5851 tgl@sss.pgh.pa.us 5285 [ + + ]: 136 : for (i = tail; i < head; i++)
5286 : : {
5287 [ + - ]: 10 : if (KnownAssignedXidsValid[i])
5288 : : {
5289 : 10 : nxids++;
5745 rhaas@postgresql.org 5290 : 10 : appendStringInfo(&buf, "[%d]=%u ", i, KnownAssignedXids[i]);
5291 : : }
5292 : : }
5293 : :
5294 [ - + ]: 126 : elog(trace_level, "%d KnownAssignedXids (num=%d tail=%d head=%d) %s",
5295 : : nxids,
5296 : : pArray->numKnownAssignedXids,
5297 : : pArray->tailKnownAssignedXids,
5298 : : pArray->headKnownAssignedXids,
5299 : : buf.data);
5300 : :
5981 simon@2ndQuadrant.co 5301 : 126 : pfree(buf.data);
5302 : 126 : }
5303 : :
5304 : : /*
5305 : : * KnownAssignedXidsReset
5306 : : * Resets KnownAssignedXids to be empty
5307 : : */
5308 : : static void
5079 simon@2ndQuadrant.co 5309 :UBC 0 : KnownAssignedXidsReset(void)
5310 : : {
2734 andres@anarazel.de 5311 : 0 : ProcArrayStruct *pArray = procArray;
5312 : :
5079 simon@2ndQuadrant.co 5313 : 0 : LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
5314 : :
5315 : 0 : pArray->numKnownAssignedXids = 0;
5316 : 0 : pArray->tailKnownAssignedXids = 0;
5317 : 0 : pArray->headKnownAssignedXids = 0;
5318 : :
5319 : 0 : LWLockRelease(ProcArrayLock);
5320 : 0 : }
|
