Age Owner Branch data TLA Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * predicate.c
4 : : * POSTGRES predicate locking
5 : : * to support full serializable transaction isolation
6 : : *
7 : : *
8 : : * The approach taken is to implement Serializable Snapshot Isolation (SSI)
9 : : * as initially described in this paper:
10 : : *
11 : : * Michael J. Cahill, Uwe Röhm, and Alan D. Fekete. 2008.
12 : : * Serializable isolation for snapshot databases.
13 : : * In SIGMOD '08: Proceedings of the 2008 ACM SIGMOD
14 : : * international conference on Management of data,
15 : : * pages 729-738, New York, NY, USA. ACM.
16 : : * http://doi.acm.org/10.1145/1376616.1376690
17 : : *
18 : : * and further elaborated in Cahill's doctoral thesis:
19 : : *
20 : : * Michael James Cahill. 2009.
21 : : * Serializable Isolation for Snapshot Databases.
22 : : * Sydney Digital Theses.
23 : : * University of Sydney, School of Information Technologies.
24 : : * http://hdl.handle.net/2123/5353
25 : : *
26 : : *
27 : : * Predicate locks for Serializable Snapshot Isolation (SSI) are SIREAD
28 : : * locks, which are so different from normal locks that a distinct set of
29 : : * structures is required to handle them. They are needed to detect
30 : : * rw-conflicts when the read happens before the write. (When the write
31 : : * occurs first, the reading transaction can check for a conflict by
32 : : * examining the MVCC data.)
33 : : *
34 : : * (1) Besides tuples actually read, they must cover ranges of tuples
35 : : * which would have been read based on the predicate. This will
36 : : * require modelling the predicates through locks against database
37 : : * objects such as pages, index ranges, or entire tables.
38 : : *
39 : : * (2) They must be kept in RAM for quick access. Because of this, it
40 : : * isn't possible to always maintain tuple-level granularity -- when
41 : : * the space allocated to store these approaches exhaustion, a
42 : : * request for a lock may need to scan for situations where a single
43 : : * transaction holds many fine-grained locks which can be coalesced
44 : : * into a single coarser-grained lock.
45 : : *
46 : : * (3) They never block anything; they are more like flags than locks
47 : : * in that regard; although they refer to database objects and are
48 : : * used to identify rw-conflicts with normal write locks.
49 : : *
50 : : * (4) While they are associated with a transaction, they must survive
51 : : * a successful COMMIT of that transaction, and remain until all
52 : : * overlapping transactions complete. This even means that they
53 : : * must survive termination of the transaction's process. If a
54 : : * top level transaction is rolled back, however, it is immediately
55 : : * flagged so that it can be ignored, and its SIREAD locks can be
56 : : * released any time after that.
57 : : *
58 : : * (5) The only transactions which create SIREAD locks or check for
59 : : * conflicts with them are serializable transactions.
60 : : *
61 : : * (6) When a write lock for a top level transaction is found to cover
62 : : * an existing SIREAD lock for the same transaction, the SIREAD lock
63 : : * can be deleted.
64 : : *
65 : : * (7) A write from a serializable transaction must ensure that an xact
66 : : * record exists for the transaction, with the same lifespan (until
67 : : * all concurrent transaction complete or the transaction is rolled
68 : : * back) so that rw-dependencies to that transaction can be
69 : : * detected.
70 : : *
71 : : * We use an optimization for read-only transactions. Under certain
72 : : * circumstances, a read-only transaction's snapshot can be shown to
73 : : * never have conflicts with other transactions. This is referred to
74 : : * as a "safe" snapshot (and one known not to be is "unsafe").
75 : : * However, it can't be determined whether a snapshot is safe until
76 : : * all concurrent read/write transactions complete.
77 : : *
78 : : * Once a read-only transaction is known to have a safe snapshot, it
79 : : * can release its predicate locks and exempt itself from further
80 : : * predicate lock tracking. READ ONLY DEFERRABLE transactions run only
81 : : * on safe snapshots, waiting as necessary for one to be available.
82 : : *
83 : : *
84 : : * Lightweight locks to manage access to the predicate locking shared
85 : : * memory objects must be taken in this order, and should be released in
86 : : * reverse order:
87 : : *
88 : : * SerializableFinishedListLock
89 : : * - Protects the list of transactions which have completed but which
90 : : * may yet matter because they overlap still-active transactions.
91 : : *
92 : : * SerializablePredicateListLock
93 : : * - Protects the linked list of locks held by a transaction. Note
94 : : * that the locks themselves are also covered by the partition
95 : : * locks of their respective lock targets; this lock only affects
96 : : * the linked list connecting the locks related to a transaction.
97 : : * - All transactions share this single lock (with no partitioning).
98 : : * - There is never a need for a process other than the one running
99 : : * an active transaction to walk the list of locks held by that
100 : : * transaction, except parallel query workers sharing the leader's
101 : : * transaction. In the parallel case, an extra per-sxact lock is
102 : : * taken; see below.
103 : : * - It is relatively infrequent that another process needs to
104 : : * modify the list for a transaction, but it does happen for such
105 : : * things as index page splits for pages with predicate locks and
106 : : * freeing of predicate locked pages by a vacuum process. When
107 : : * removing a lock in such cases, the lock itself contains the
108 : : * pointers needed to remove it from the list. When adding a
109 : : * lock in such cases, the lock can be added using the anchor in
110 : : * the transaction structure. Neither requires walking the list.
111 : : * - Cleaning up the list for a terminated transaction is sometimes
112 : : * not done on a retail basis, in which case no lock is required.
113 : : * - Due to the above, a process accessing its active transaction's
114 : : * list always uses a shared lock, regardless of whether it is
115 : : * walking or maintaining the list. This improves concurrency
116 : : * for the common access patterns.
117 : : * - A process which needs to alter the list of a transaction other
118 : : * than its own active transaction must acquire an exclusive
119 : : * lock.
120 : : *
121 : : * SERIALIZABLEXACT's member 'perXactPredicateListLock'
122 : : * - Protects the linked list of predicate locks held by a transaction.
123 : : * Only needed for parallel mode, where multiple backends share the
124 : : * same SERIALIZABLEXACT object. Not needed if
125 : : * SerializablePredicateListLock is held exclusively.
126 : : *
127 : : * PredicateLockHashPartitionLock(hashcode)
128 : : * - The same lock protects a target, all locks on that target, and
129 : : * the linked list of locks on the target.
130 : : * - When more than one is needed, acquire in ascending address order.
131 : : * - When all are needed (rare), acquire in ascending index order with
132 : : * PredicateLockHashPartitionLockByIndex(index).
133 : : *
134 : : * SerializableXactHashLock
135 : : * - Protects both PredXact and SerializableXidHash.
136 : : *
137 : : * SerialControlLock
138 : : * - Protects SerialControlData members
139 : : *
140 : : * SLRU per-bank locks
141 : : * - Protects SerialSlruCtl
142 : : *
143 : : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
144 : : * Portions Copyright (c) 1994, Regents of the University of California
145 : : *
146 : : *
147 : : * IDENTIFICATION
148 : : * src/backend/storage/lmgr/predicate.c
149 : : *
150 : : *-------------------------------------------------------------------------
151 : : */
152 : : /*
153 : : * INTERFACE ROUTINES
154 : : *
155 : : * housekeeping for setting up shared memory predicate lock structures
156 : : * PredicateLockShmemInit(void)
157 : : * PredicateLockShmemSize(void)
158 : : *
159 : : * predicate lock reporting
160 : : * GetPredicateLockStatusData(void)
161 : : * PageIsPredicateLocked(Relation relation, BlockNumber blkno)
162 : : *
163 : : * predicate lock maintenance
164 : : * GetSerializableTransactionSnapshot(Snapshot snapshot)
165 : : * SetSerializableTransactionSnapshot(Snapshot snapshot,
166 : : * VirtualTransactionId *sourcevxid)
167 : : * RegisterPredicateLockingXid(void)
168 : : * PredicateLockRelation(Relation relation, Snapshot snapshot)
169 : : * PredicateLockPage(Relation relation, BlockNumber blkno,
170 : : * Snapshot snapshot)
171 : : * PredicateLockTID(Relation relation, const ItemPointerData *tid, Snapshot snapshot,
172 : : * TransactionId tuple_xid)
173 : : * PredicateLockPageSplit(Relation relation, BlockNumber oldblkno,
174 : : * BlockNumber newblkno)
175 : : * PredicateLockPageCombine(Relation relation, BlockNumber oldblkno,
176 : : * BlockNumber newblkno)
177 : : * TransferPredicateLocksToHeapRelation(Relation relation)
178 : : * ReleasePredicateLocks(bool isCommit, bool isReadOnlySafe)
179 : : *
180 : : * conflict detection (may also trigger rollback)
181 : : * CheckForSerializableConflictOut(Relation relation, TransactionId xid,
182 : : * Snapshot snapshot)
183 : : * CheckForSerializableConflictIn(Relation relation, const ItemPointerData *tid,
184 : : * BlockNumber blkno)
185 : : * CheckTableForSerializableConflictIn(Relation relation)
186 : : *
187 : : * final rollback checking
188 : : * PreCommit_CheckForSerializationFailure(void)
189 : : *
190 : : * two-phase commit support
191 : : * AtPrepare_PredicateLocks(void);
192 : : * PostPrepare_PredicateLocks(TransactionId xid);
193 : : * PredicateLockTwoPhaseFinish(FullTransactionId fxid, bool isCommit);
194 : : * predicatelock_twophase_recover(FullTransactionId fxid, uint16 info,
195 : : * void *recdata, uint32 len);
196 : : */
197 : :
198 : : #include "postgres.h"
199 : :
200 : : #include "access/parallel.h"
201 : : #include "access/slru.h"
202 : : #include "access/transam.h"
203 : : #include "access/twophase.h"
204 : : #include "access/twophase_rmgr.h"
205 : : #include "access/xact.h"
206 : : #include "access/xlog.h"
207 : : #include "miscadmin.h"
208 : : #include "pgstat.h"
209 : : #include "port/pg_lfind.h"
210 : : #include "storage/predicate.h"
211 : : #include "storage/predicate_internals.h"
212 : : #include "storage/proc.h"
213 : : #include "storage/procarray.h"
214 : : #include "utils/guc_hooks.h"
215 : : #include "utils/rel.h"
216 : : #include "utils/snapmgr.h"
217 : : #include "utils/wait_event.h"
218 : :
219 : : /* Uncomment the next line to test the graceful degradation code. */
220 : : /* #define TEST_SUMMARIZE_SERIAL */
221 : :
222 : : /*
223 : : * Test the most selective fields first, for performance.
224 : : *
225 : : * a is covered by b if all of the following hold:
226 : : * 1) a.database = b.database
227 : : * 2) a.relation = b.relation
228 : : * 3) b.offset is invalid (b is page-granularity or higher)
229 : : * 4) either of the following:
230 : : * 4a) a.offset is valid (a is tuple-granularity) and a.page = b.page
231 : : * or 4b) a.offset is invalid and b.page is invalid (a is
232 : : * page-granularity and b is relation-granularity
233 : : */
234 : : #define TargetTagIsCoveredBy(covered_target, covering_target) \
235 : : ((GET_PREDICATELOCKTARGETTAG_RELATION(covered_target) == /* (2) */ \
236 : : GET_PREDICATELOCKTARGETTAG_RELATION(covering_target)) \
237 : : && (GET_PREDICATELOCKTARGETTAG_OFFSET(covering_target) == \
238 : : InvalidOffsetNumber) /* (3) */ \
239 : : && (((GET_PREDICATELOCKTARGETTAG_OFFSET(covered_target) != \
240 : : InvalidOffsetNumber) /* (4a) */ \
241 : : && (GET_PREDICATELOCKTARGETTAG_PAGE(covering_target) == \
242 : : GET_PREDICATELOCKTARGETTAG_PAGE(covered_target))) \
243 : : || ((GET_PREDICATELOCKTARGETTAG_PAGE(covering_target) == \
244 : : InvalidBlockNumber) /* (4b) */ \
245 : : && (GET_PREDICATELOCKTARGETTAG_PAGE(covered_target) \
246 : : != InvalidBlockNumber))) \
247 : : && (GET_PREDICATELOCKTARGETTAG_DB(covered_target) == /* (1) */ \
248 : : GET_PREDICATELOCKTARGETTAG_DB(covering_target)))
249 : :
250 : : /*
251 : : * The predicate locking target and lock shared hash tables are partitioned to
252 : : * reduce contention. To determine which partition a given target belongs to,
253 : : * compute the tag's hash code with PredicateLockTargetTagHashCode(), then
254 : : * apply one of these macros.
255 : : * NB: NUM_PREDICATELOCK_PARTITIONS must be a power of 2!
256 : : */
257 : : #define PredicateLockHashPartition(hashcode) \
258 : : ((hashcode) % NUM_PREDICATELOCK_PARTITIONS)
259 : : #define PredicateLockHashPartitionLock(hashcode) \
260 : : (&MainLWLockArray[PREDICATELOCK_MANAGER_LWLOCK_OFFSET + \
261 : : PredicateLockHashPartition(hashcode)].lock)
262 : : #define PredicateLockHashPartitionLockByIndex(i) \
263 : : (&MainLWLockArray[PREDICATELOCK_MANAGER_LWLOCK_OFFSET + (i)].lock)
264 : :
265 : : #define NPREDICATELOCKTARGETENTS() \
266 : : mul_size(max_predicate_locks_per_xact, add_size(MaxBackends, max_prepared_xacts))
267 : :
268 : : #define SxactIsOnFinishedList(sxact) (!dlist_node_is_detached(&(sxact)->finishedLink))
269 : :
270 : : /*
271 : : * Note that a sxact is marked "prepared" once it has passed
272 : : * PreCommit_CheckForSerializationFailure, even if it isn't using
273 : : * 2PC. This is the point at which it can no longer be aborted.
274 : : *
275 : : * The PREPARED flag remains set after commit, so SxactIsCommitted
276 : : * implies SxactIsPrepared.
277 : : */
278 : : #define SxactIsCommitted(sxact) (((sxact)->flags & SXACT_FLAG_COMMITTED) != 0)
279 : : #define SxactIsPrepared(sxact) (((sxact)->flags & SXACT_FLAG_PREPARED) != 0)
280 : : #define SxactIsRolledBack(sxact) (((sxact)->flags & SXACT_FLAG_ROLLED_BACK) != 0)
281 : : #define SxactIsDoomed(sxact) (((sxact)->flags & SXACT_FLAG_DOOMED) != 0)
282 : : #define SxactIsReadOnly(sxact) (((sxact)->flags & SXACT_FLAG_READ_ONLY) != 0)
283 : : #define SxactHasSummaryConflictIn(sxact) (((sxact)->flags & SXACT_FLAG_SUMMARY_CONFLICT_IN) != 0)
284 : : #define SxactHasSummaryConflictOut(sxact) (((sxact)->flags & SXACT_FLAG_SUMMARY_CONFLICT_OUT) != 0)
285 : : /*
286 : : * The following macro actually means that the specified transaction has a
287 : : * conflict out *to a transaction which committed ahead of it*. It's hard
288 : : * to get that into a name of a reasonable length.
289 : : */
290 : : #define SxactHasConflictOut(sxact) (((sxact)->flags & SXACT_FLAG_CONFLICT_OUT) != 0)
291 : : #define SxactIsDeferrableWaiting(sxact) (((sxact)->flags & SXACT_FLAG_DEFERRABLE_WAITING) != 0)
292 : : #define SxactIsROSafe(sxact) (((sxact)->flags & SXACT_FLAG_RO_SAFE) != 0)
293 : : #define SxactIsROUnsafe(sxact) (((sxact)->flags & SXACT_FLAG_RO_UNSAFE) != 0)
294 : : #define SxactIsPartiallyReleased(sxact) (((sxact)->flags & SXACT_FLAG_PARTIALLY_RELEASED) != 0)
295 : :
296 : : /*
297 : : * Compute the hash code associated with a PREDICATELOCKTARGETTAG.
298 : : *
299 : : * To avoid unnecessary recomputations of the hash code, we try to do this
300 : : * just once per function, and then pass it around as needed. Aside from
301 : : * passing the hashcode to hash_search_with_hash_value(), we can extract
302 : : * the lock partition number from the hashcode.
303 : : */
304 : : #define PredicateLockTargetTagHashCode(predicatelocktargettag) \
305 : : get_hash_value(PredicateLockTargetHash, predicatelocktargettag)
306 : :
307 : : /*
308 : : * Given a predicate lock tag, and the hash for its target,
309 : : * compute the lock hash.
310 : : *
311 : : * To make the hash code also depend on the transaction, we xor the sxid
312 : : * struct's address into the hash code, left-shifted so that the
313 : : * partition-number bits don't change. Since this is only a hash, we
314 : : * don't care if we lose high-order bits of the address; use an
315 : : * intermediate variable to suppress cast-pointer-to-int warnings.
316 : : */
317 : : #define PredicateLockHashCodeFromTargetHashCode(predicatelocktag, targethash) \
318 : : ((targethash) ^ ((uint32) PointerGetDatum((predicatelocktag)->myXact)) \
319 : : << LOG2_NUM_PREDICATELOCK_PARTITIONS)
320 : :
321 : :
322 : : /*
323 : : * The SLRU buffer area through which we access the old xids.
324 : : */
325 : : static SlruCtlData SerialSlruCtlData;
326 : :
327 : : #define SerialSlruCtl (&SerialSlruCtlData)
328 : :
329 : : #define SERIAL_PAGESIZE BLCKSZ
330 : : #define SERIAL_ENTRYSIZE sizeof(SerCommitSeqNo)
331 : : #define SERIAL_ENTRIESPERPAGE (SERIAL_PAGESIZE / SERIAL_ENTRYSIZE)
332 : :
333 : : /*
334 : : * Set maximum pages based on the number needed to track all transactions.
335 : : */
336 : : #define SERIAL_MAX_PAGE (MaxTransactionId / SERIAL_ENTRIESPERPAGE)
337 : :
338 : : #define SerialNextPage(page) (((page) >= SERIAL_MAX_PAGE) ? 0 : (page) + 1)
339 : :
340 : : #define SerialValue(slotno, xid) (*((SerCommitSeqNo *) \
341 : : (SerialSlruCtl->shared->page_buffer[slotno] + \
342 : : ((((uint32) (xid)) % SERIAL_ENTRIESPERPAGE) * SERIAL_ENTRYSIZE))))
343 : :
344 : : #define SerialPage(xid) (((uint32) (xid)) / SERIAL_ENTRIESPERPAGE)
345 : :
346 : : typedef struct SerialControlData
347 : : {
348 : : int64 headPage; /* newest initialized page */
349 : : TransactionId headXid; /* newest valid Xid in the SLRU */
350 : : TransactionId tailXid; /* oldest xmin we might be interested in */
351 : : } SerialControlData;
352 : :
353 : : typedef struct SerialControlData *SerialControl;
354 : :
355 : : static SerialControl serialControl;
356 : :
357 : : /*
358 : : * When the oldest committed transaction on the "finished" list is moved to
359 : : * SLRU, its predicate locks will be moved to this "dummy" transaction,
360 : : * collapsing duplicate targets. When a duplicate is found, the later
361 : : * commitSeqNo is used.
362 : : */
363 : : static SERIALIZABLEXACT *OldCommittedSxact;
364 : :
365 : :
366 : : /*
367 : : * These configuration variables are used to set the predicate lock table size
368 : : * and to control promotion of predicate locks to coarser granularity in an
369 : : * attempt to degrade performance (mostly as false positive serialization
370 : : * failure) gracefully in the face of memory pressure.
371 : : */
372 : : int max_predicate_locks_per_xact; /* in guc_tables.c */
373 : : int max_predicate_locks_per_relation; /* in guc_tables.c */
374 : : int max_predicate_locks_per_page; /* in guc_tables.c */
375 : :
376 : : /*
377 : : * This provides a list of objects in order to track transactions
378 : : * participating in predicate locking. Entries in the list are fixed size,
379 : : * and reside in shared memory. The memory address of an entry must remain
380 : : * fixed during its lifetime. The list will be protected from concurrent
381 : : * update externally; no provision is made in this code to manage that. The
382 : : * number of entries in the list, and the size allowed for each entry is
383 : : * fixed upon creation.
384 : : */
385 : : static PredXactList PredXact;
386 : :
387 : : /*
388 : : * This provides a pool of RWConflict data elements to use in conflict lists
389 : : * between transactions.
390 : : */
391 : : static RWConflictPoolHeader RWConflictPool;
392 : :
393 : : /*
394 : : * The predicate locking hash tables are in shared memory.
395 : : * Each backend keeps pointers to them.
396 : : */
397 : : static HTAB *SerializableXidHash;
398 : : static HTAB *PredicateLockTargetHash;
399 : : static HTAB *PredicateLockHash;
400 : : static dlist_head *FinishedSerializableTransactions;
401 : :
402 : : /*
403 : : * Tag for a dummy entry in PredicateLockTargetHash. By temporarily removing
404 : : * this entry, you can ensure that there's enough scratch space available for
405 : : * inserting one entry in the hash table. This is an otherwise-invalid tag.
406 : : */
407 : : static const PREDICATELOCKTARGETTAG ScratchTargetTag = {0, 0, 0, 0};
408 : : static uint32 ScratchTargetTagHash;
409 : : static LWLock *ScratchPartitionLock;
410 : :
411 : : /*
412 : : * The local hash table used to determine when to combine multiple fine-
413 : : * grained locks into a single courser-grained lock.
414 : : */
415 : : static HTAB *LocalPredicateLockHash = NULL;
416 : :
417 : : /*
418 : : * Keep a pointer to the currently-running serializable transaction (if any)
419 : : * for quick reference. Also, remember if we have written anything that could
420 : : * cause a rw-conflict.
421 : : */
422 : : static SERIALIZABLEXACT *MySerializableXact = InvalidSerializableXact;
423 : : static bool MyXactDidWrite = false;
424 : :
425 : : /*
426 : : * The SXACT_FLAG_RO_UNSAFE optimization might lead us to release
427 : : * MySerializableXact early. If that happens in a parallel query, the leader
428 : : * needs to defer the destruction of the SERIALIZABLEXACT until end of
429 : : * transaction, because the workers still have a reference to it. In that
430 : : * case, the leader stores it here.
431 : : */
432 : : static SERIALIZABLEXACT *SavedSerializableXact = InvalidSerializableXact;
433 : :
434 : : /* local functions */
435 : :
436 : : static SERIALIZABLEXACT *CreatePredXact(void);
437 : : static void ReleasePredXact(SERIALIZABLEXACT *sxact);
438 : :
439 : : static bool RWConflictExists(const SERIALIZABLEXACT *reader, const SERIALIZABLEXACT *writer);
440 : : static void SetRWConflict(SERIALIZABLEXACT *reader, SERIALIZABLEXACT *writer);
441 : : static void SetPossibleUnsafeConflict(SERIALIZABLEXACT *roXact, SERIALIZABLEXACT *activeXact);
442 : : static void ReleaseRWConflict(RWConflict conflict);
443 : : static void FlagSxactUnsafe(SERIALIZABLEXACT *sxact);
444 : :
445 : : static bool SerialPagePrecedesLogically(int64 page1, int64 page2);
446 : : static int serial_errdetail_for_io_error(const void *opaque_data);
447 : : static void SerialInit(void);
448 : : static void SerialAdd(TransactionId xid, SerCommitSeqNo minConflictCommitSeqNo);
449 : : static SerCommitSeqNo SerialGetMinConflictCommitSeqNo(TransactionId xid);
450 : : static void SerialSetActiveSerXmin(TransactionId xid);
451 : :
452 : : static uint32 predicatelock_hash(const void *key, Size keysize);
453 : : static void SummarizeOldestCommittedSxact(void);
454 : : static Snapshot GetSafeSnapshot(Snapshot origSnapshot);
455 : : static Snapshot GetSerializableTransactionSnapshotInt(Snapshot snapshot,
456 : : VirtualTransactionId *sourcevxid,
457 : : int sourcepid);
458 : : static bool PredicateLockExists(const PREDICATELOCKTARGETTAG *targettag);
459 : : static bool GetParentPredicateLockTag(const PREDICATELOCKTARGETTAG *tag,
460 : : PREDICATELOCKTARGETTAG *parent);
461 : : static bool CoarserLockCovers(const PREDICATELOCKTARGETTAG *newtargettag);
462 : : static void RemoveScratchTarget(bool lockheld);
463 : : static void RestoreScratchTarget(bool lockheld);
464 : : static void RemoveTargetIfNoLongerUsed(PREDICATELOCKTARGET *target,
465 : : uint32 targettaghash);
466 : : static void DeleteChildTargetLocks(const PREDICATELOCKTARGETTAG *newtargettag);
467 : : static int MaxPredicateChildLocks(const PREDICATELOCKTARGETTAG *tag);
468 : : static bool CheckAndPromotePredicateLockRequest(const PREDICATELOCKTARGETTAG *reqtag);
469 : : static void DecrementParentLocks(const PREDICATELOCKTARGETTAG *targettag);
470 : : static void CreatePredicateLock(const PREDICATELOCKTARGETTAG *targettag,
471 : : uint32 targettaghash,
472 : : SERIALIZABLEXACT *sxact);
473 : : static void DeleteLockTarget(PREDICATELOCKTARGET *target, uint32 targettaghash);
474 : : static bool TransferPredicateLocksToNewTarget(PREDICATELOCKTARGETTAG oldtargettag,
475 : : PREDICATELOCKTARGETTAG newtargettag,
476 : : bool removeOld);
477 : : static void PredicateLockAcquire(const PREDICATELOCKTARGETTAG *targettag);
478 : : static void DropAllPredicateLocksFromTable(Relation relation,
479 : : bool transfer);
480 : : static void SetNewSxactGlobalXmin(void);
481 : : static void ClearOldPredicateLocks(void);
482 : : static void ReleaseOneSerializableXact(SERIALIZABLEXACT *sxact, bool partial,
483 : : bool summarize);
484 : : static bool XidIsConcurrent(TransactionId xid);
485 : : static void CheckTargetForConflictsIn(PREDICATELOCKTARGETTAG *targettag);
486 : : static void FlagRWConflict(SERIALIZABLEXACT *reader, SERIALIZABLEXACT *writer);
487 : : static void OnConflict_CheckForSerializationFailure(const SERIALIZABLEXACT *reader,
488 : : SERIALIZABLEXACT *writer);
489 : : static void CreateLocalPredicateLockHash(void);
490 : : static void ReleasePredicateLocksLocal(void);
491 : :
492 : :
493 : : /*------------------------------------------------------------------------*/
494 : :
495 : : /*
496 : : * Does this relation participate in predicate locking? Temporary and system
497 : : * relations are exempt.
498 : : */
499 : : static inline bool
5387 heikki.linnakangas@i 500 :CBC 160452 : PredicateLockingNeededForRelation(Relation relation)
501 : : {
1704 tgl@sss.pgh.pa.us 502 [ + + ]: 217636 : return !(relation->rd_id < FirstUnpinnedObjectId ||
1200 michael@paquier.xyz 503 [ + + ]: 57184 : RelationUsesLocalBuffers(relation));
504 : : }
505 : :
506 : : /*
507 : : * When a public interface method is called for a read, this is the test to
508 : : * see if we should do a quick return.
509 : : *
510 : : * Note: this function has side-effects! If this transaction has been flagged
511 : : * as RO-safe since the last call, we release all predicate locks and reset
512 : : * MySerializableXact. That makes subsequent calls to return quickly.
513 : : *
514 : : * This is marked as 'inline' to eliminate the function call overhead in the
515 : : * common case that serialization is not needed.
516 : : */
517 : : static inline bool
5387 heikki.linnakangas@i 518 : 62277264 : SerializationNeededForRead(Relation relation, Snapshot snapshot)
519 : : {
520 : : /* Nothing to do if this is not a serializable transaction */
521 [ + + ]: 62277264 : if (MySerializableXact == InvalidSerializableXact)
522 : 62123028 : return false;
523 : :
524 : : /*
525 : : * Don't acquire locks or conflict when scanning with a special snapshot.
526 : : * This excludes things like CLUSTER and REINDEX. They use the wholesale
527 : : * functions TransferPredicateLocksToHeapRelation() and
528 : : * CheckTableForSerializableConflictIn() to participate in serialization,
529 : : * but the scans involved don't need serialization.
530 : : */
531 [ + + + - ]: 154236 : if (!IsMVCCSnapshot(snapshot))
532 : 1718 : return false;
533 : :
534 : : /*
535 : : * Check if we have just become "RO-safe". If we have, immediately release
536 : : * all locks as they're not needed anymore. This also resets
537 : : * MySerializableXact, so that subsequent calls to this function can exit
538 : : * quickly.
539 : : *
540 : : * A transaction is flagged as RO_SAFE if all concurrent R/W transactions
541 : : * commit without having conflicts out to an earlier snapshot, thus
542 : : * ensuring that no conflicts are possible for this transaction.
543 : : */
544 [ + + ]: 152518 : if (SxactIsROSafe(MySerializableXact))
545 : : {
2557 tmunro@postgresql.or 546 : 33 : ReleasePredicateLocks(false, true);
5387 heikki.linnakangas@i 547 : 33 : return false;
548 : : }
549 : :
550 : : /* Check if the relation doesn't participate in predicate locking */
551 [ + + ]: 152485 : if (!PredicateLockingNeededForRelation(relation))
552 : 99999 : return false;
553 : :
5386 554 : 52486 : return true; /* no excuse to skip predicate locking */
555 : : }
556 : :
557 : : /*
558 : : * Like SerializationNeededForRead(), but called on writes.
559 : : * The logic is the same, but there is no snapshot and we can't be RO-safe.
560 : : */
561 : : static inline bool
5387 562 : 18053852 : SerializationNeededForWrite(Relation relation)
563 : : {
564 : : /* Nothing to do if this is not a serializable transaction */
565 [ + + ]: 18053852 : if (MySerializableXact == InvalidSerializableXact)
566 : 18045970 : return false;
567 : :
568 : : /* Check if the relation doesn't participate in predicate locking */
569 [ + + ]: 7882 : if (!PredicateLockingNeededForRelation(relation))
570 : 3394 : return false;
571 : :
5386 572 : 4488 : return true; /* no excuse to skip predicate locking */
573 : : }
574 : :
575 : :
576 : : /*------------------------------------------------------------------------*/
577 : :
578 : : /*
579 : : * These functions are a simple implementation of a list for this specific
580 : : * type of struct. If there is ever a generalized shared memory list, we
581 : : * should probably switch to that.
582 : : */
583 : : static SERIALIZABLEXACT *
5515 584 : 2817 : CreatePredXact(void)
585 : : {
586 : : SERIALIZABLEXACT *sxact;
587 : :
1151 andres@anarazel.de 588 [ - + ]: 2817 : if (dlist_is_empty(&PredXact->availableList))
5515 heikki.linnakangas@i 589 :UBC 0 : return NULL;
590 : :
1151 andres@anarazel.de 591 :CBC 2817 : sxact = dlist_container(SERIALIZABLEXACT, xactLink,
592 : : dlist_pop_head_node(&PredXact->availableList));
593 : 2817 : dlist_push_tail(&PredXact->activeList, &sxact->xactLink);
594 : 2817 : return sxact;
595 : : }
596 : :
597 : : static void
5515 heikki.linnakangas@i 598 : 1667 : ReleasePredXact(SERIALIZABLEXACT *sxact)
599 : : {
600 [ - + ]: 1667 : Assert(ShmemAddrIsValid(sxact));
601 : :
1151 andres@anarazel.de 602 : 1667 : dlist_delete(&sxact->xactLink);
603 : 1667 : dlist_push_tail(&PredXact->availableList, &sxact->xactLink);
5515 heikki.linnakangas@i 604 : 1667 : }
605 : :
606 : : /*------------------------------------------------------------------------*/
607 : :
608 : : /*
609 : : * These functions manage primitive access to the RWConflict pool and lists.
610 : : */
611 : : static bool
5380 tgl@sss.pgh.pa.us 612 : 2682 : RWConflictExists(const SERIALIZABLEXACT *reader, const SERIALIZABLEXACT *writer)
613 : : {
614 : : dlist_iter iter;
615 : :
5515 heikki.linnakangas@i 616 [ - + ]: 2682 : Assert(reader != writer);
617 : :
618 : : /* Check the ends of the purported conflict first. */
5387 619 [ + - ]: 2682 : if (SxactIsDoomed(reader)
620 [ + + ]: 2682 : || SxactIsDoomed(writer)
1151 andres@anarazel.de 621 [ + + ]: 2674 : || dlist_is_empty(&reader->outConflicts)
622 [ + + ]: 609 : || dlist_is_empty(&writer->inConflicts))
5515 heikki.linnakangas@i 623 : 2145 : return false;
624 : :
625 : : /*
626 : : * A conflict is possible; walk the list to find out.
627 : : *
628 : : * The unconstify is needed as we have no const version of
629 : : * dlist_foreach().
630 : : */
1151 andres@anarazel.de 631 [ + - + + ]: 561 : dlist_foreach(iter, &unconstify(SERIALIZABLEXACT *, reader)->outConflicts)
632 : : {
633 : 537 : RWConflict conflict =
1031 tgl@sss.pgh.pa.us 634 :ECB (535) : dlist_container(RWConflictData, outLink, iter.cur);
635 : :
5515 heikki.linnakangas@i 636 [ + + ]:CBC 537 : if (conflict->sxactIn == writer)
637 : 513 : return true;
638 : : }
639 : :
640 : : /* No conflict found. */
641 : 24 : return false;
642 : : }
643 : :
644 : : static void
645 : 792 : SetRWConflict(SERIALIZABLEXACT *reader, SERIALIZABLEXACT *writer)
646 : : {
647 : : RWConflict conflict;
648 : :
649 [ - + ]: 792 : Assert(reader != writer);
650 [ - + ]: 792 : Assert(!RWConflictExists(reader, writer));
651 : :
1151 andres@anarazel.de 652 [ - + ]: 792 : if (dlist_is_empty(&RWConflictPool->availableList))
5515 heikki.linnakangas@i 653 [ # # ]:UBC 0 : ereport(ERROR,
654 : : (errcode(ERRCODE_OUT_OF_MEMORY),
655 : : errmsg("not enough elements in RWConflictPool to record a read/write conflict"),
656 : : errhint("You might need to run fewer transactions at a time or increase \"max_connections\".")));
657 : :
1151 andres@anarazel.de 658 :CBC 792 : conflict = dlist_head_element(RWConflictData, outLink, &RWConflictPool->availableList);
659 : 792 : dlist_delete(&conflict->outLink);
660 : :
5515 heikki.linnakangas@i 661 : 792 : conflict->sxactOut = reader;
662 : 792 : conflict->sxactIn = writer;
1151 andres@anarazel.de 663 : 792 : dlist_push_tail(&reader->outConflicts, &conflict->outLink);
664 : 792 : dlist_push_tail(&writer->inConflicts, &conflict->inLink);
5515 heikki.linnakangas@i 665 : 792 : }
666 : :
667 : : static void
668 : 135 : SetPossibleUnsafeConflict(SERIALIZABLEXACT *roXact,
669 : : SERIALIZABLEXACT *activeXact)
670 : : {
671 : : RWConflict conflict;
672 : :
673 [ - + ]: 135 : Assert(roXact != activeXact);
674 [ - + ]: 135 : Assert(SxactIsReadOnly(roXact));
675 [ - + ]: 135 : Assert(!SxactIsReadOnly(activeXact));
676 : :
1151 andres@anarazel.de 677 [ - + ]: 135 : if (dlist_is_empty(&RWConflictPool->availableList))
5515 heikki.linnakangas@i 678 [ # # ]:UBC 0 : ereport(ERROR,
679 : : (errcode(ERRCODE_OUT_OF_MEMORY),
680 : : errmsg("not enough elements in RWConflictPool to record a potential read/write conflict"),
681 : : errhint("You might need to run fewer transactions at a time or increase \"max_connections\".")));
682 : :
1151 andres@anarazel.de 683 :CBC 135 : conflict = dlist_head_element(RWConflictData, outLink, &RWConflictPool->availableList);
684 : 135 : dlist_delete(&conflict->outLink);
685 : :
5515 heikki.linnakangas@i 686 : 135 : conflict->sxactOut = activeXact;
687 : 135 : conflict->sxactIn = roXact;
1151 andres@anarazel.de 688 : 135 : dlist_push_tail(&activeXact->possibleUnsafeConflicts, &conflict->outLink);
689 : 135 : dlist_push_tail(&roXact->possibleUnsafeConflicts, &conflict->inLink);
5515 heikki.linnakangas@i 690 : 135 : }
691 : :
692 : : static void
693 : 927 : ReleaseRWConflict(RWConflict conflict)
694 : : {
1151 andres@anarazel.de 695 : 927 : dlist_delete(&conflict->inLink);
696 : 927 : dlist_delete(&conflict->outLink);
697 : 927 : dlist_push_tail(&RWConflictPool->availableList, &conflict->outLink);
5515 heikki.linnakangas@i 698 : 927 : }
699 : :
700 : : static void
701 : 3 : FlagSxactUnsafe(SERIALIZABLEXACT *sxact)
702 : : {
703 : : dlist_mutable_iter iter;
704 : :
705 [ - + ]: 3 : Assert(SxactIsReadOnly(sxact));
706 [ - + ]: 3 : Assert(!SxactIsROSafe(sxact));
707 : :
708 : 3 : sxact->flags |= SXACT_FLAG_RO_UNSAFE;
709 : :
710 : : /*
711 : : * We know this isn't a safe snapshot, so we can stop looking for other
712 : : * potential conflicts.
713 : : */
1151 andres@anarazel.de 714 [ + - + + ]: 6 : dlist_foreach_modify(iter, &sxact->possibleUnsafeConflicts)
715 : : {
716 : 3 : RWConflict conflict =
1031 tgl@sss.pgh.pa.us 717 : 3 : dlist_container(RWConflictData, inLink, iter.cur);
718 : :
5515 heikki.linnakangas@i 719 [ - + ]: 3 : Assert(!SxactIsReadOnly(conflict->sxactOut));
720 [ - + ]: 3 : Assert(sxact == conflict->sxactIn);
721 : :
722 : 3 : ReleaseRWConflict(conflict);
723 : : }
724 : 3 : }
725 : :
726 : : /*------------------------------------------------------------------------*/
727 : :
728 : : /*
729 : : * Decide whether a Serial page number is "older" for truncation purposes.
730 : : * Analogous to CLOGPagePrecedes().
731 : : */
732 : : static bool
837 akorotkov@postgresql 733 : 47150 : SerialPagePrecedesLogically(int64 page1, int64 page2)
734 : : {
735 : : TransactionId xid1;
736 : : TransactionId xid2;
737 : :
1884 noah@leadboat.com 738 : 47150 : xid1 = ((TransactionId) page1) * SERIAL_ENTRIESPERPAGE;
739 : 47150 : xid1 += FirstNormalTransactionId + 1;
740 : 47150 : xid2 = ((TransactionId) page2) * SERIAL_ENTRIESPERPAGE;
741 : 47150 : xid2 += FirstNormalTransactionId + 1;
742 : :
743 [ + + + + ]: 79350 : return (TransactionIdPrecedes(xid1, xid2) &&
744 : 32200 : TransactionIdPrecedes(xid1, xid2 + SERIAL_ENTRIESPERPAGE - 1));
745 : : }
746 : :
747 : : static int
2 heikki.linnakangas@i 748 :UNC 0 : serial_errdetail_for_io_error(const void *opaque_data)
749 : : {
750 : 0 : TransactionId xid = *(const TransactionId *) opaque_data;
751 : :
752 : 0 : return errdetail("Could not access serializable CSN of transaction %u.", xid);
753 : : }
754 : :
755 : : #ifdef USE_ASSERT_CHECKING
756 : : static void
1884 noah@leadboat.com 757 :CBC 1150 : SerialPagePrecedesLogicallyUnitTests(void)
758 : : {
759 : 1150 : int per_page = SERIAL_ENTRIESPERPAGE,
760 : 1150 : offset = per_page / 2;
761 : : int64 newestPage,
762 : : oldestPage,
763 : : headPage,
764 : : targetPage;
765 : : TransactionId newestXact,
766 : : oldestXact;
767 : :
768 : : /* GetNewTransactionId() has assigned the last XID it can safely use. */
769 : 1150 : newestPage = 2 * SLRU_PAGES_PER_SEGMENT - 1; /* nothing special */
770 : 1150 : newestXact = newestPage * per_page + offset;
771 [ - + ]: 1150 : Assert(newestXact / per_page == newestPage);
772 : 1150 : oldestXact = newestXact + 1;
773 : 1150 : oldestXact -= 1U << 31;
774 : 1150 : oldestPage = oldestXact / per_page;
775 : :
776 : : /*
777 : : * In this scenario, the SLRU headPage pertains to the last ~1000 XIDs
778 : : * assigned. oldestXact finishes, ~2B XIDs having elapsed since it
779 : : * started. Further transactions cause us to summarize oldestXact to
780 : : * tailPage. Function must return false so SerialAdd() doesn't zero
781 : : * tailPage (which may contain entries for other old, recently-finished
782 : : * XIDs) and half the SLRU. Reaching this requires burning ~2B XIDs in
783 : : * single-user mode, a negligible possibility.
784 : : */
785 : 1150 : headPage = newestPage;
786 : 1150 : targetPage = oldestPage;
787 [ - + ]: 1150 : Assert(!SerialPagePrecedesLogically(headPage, targetPage));
788 : :
789 : : /*
790 : : * In this scenario, the SLRU headPage pertains to oldestXact. We're
791 : : * summarizing an XID near newestXact. (Assume few other XIDs used
792 : : * SERIALIZABLE, hence the minimal headPage advancement. Assume
793 : : * oldestXact was long-running and only recently reached the SLRU.)
794 : : * Function must return true to make SerialAdd() create targetPage.
795 : : *
796 : : * Today's implementation mishandles this case, but it doesn't matter
797 : : * enough to fix. Verify that the defect affects just one page by
798 : : * asserting correct treatment of its prior page. Reaching this case
799 : : * requires burning ~2B XIDs in single-user mode, a negligible
800 : : * possibility. Moreover, if it does happen, the consequence would be
801 : : * mild, namely a new transaction failing in SimpleLruReadPage().
802 : : */
803 : 1150 : headPage = oldestPage;
804 : 1150 : targetPage = newestPage;
805 [ - + ]: 1150 : Assert(SerialPagePrecedesLogically(headPage, targetPage - 1));
806 : : #if 0
807 : : Assert(SerialPagePrecedesLogically(headPage, targetPage));
808 : : #endif
5515 heikki.linnakangas@i 809 : 1150 : }
810 : : #endif
811 : :
812 : : /*
813 : : * Initialize for the tracking of old serializable committed xids.
814 : : */
815 : : static void
2130 tgl@sss.pgh.pa.us 816 : 1150 : SerialInit(void)
817 : : {
818 : : bool found;
819 : :
820 : : /*
821 : : * Set up SLRU management of the pg_serial data.
822 : : */
823 : 1150 : SerialSlruCtl->PagePrecedes = SerialPagePrecedesLogically;
2 heikki.linnakangas@i 824 :GNC 1150 : SerialSlruCtl->errdetail_for_io_error = serial_errdetail_for_io_error;
746 alvherre@alvh.no-ip. 825 :CBC 1150 : SimpleLruInit(SerialSlruCtl, "serializable",
826 : : serializable_buffers, 0, "pg_serial",
827 : : LWTRANCHE_SERIAL_BUFFER, LWTRANCHE_SERIAL_SLRU,
828 : : SYNC_HANDLER_NONE, false);
829 : : #ifdef USE_ASSERT_CHECKING
1884 noah@leadboat.com 830 : 1150 : SerialPagePrecedesLogicallyUnitTests();
831 : : #endif
832 : 1150 : SlruPagePrecedesUnitTests(SerialSlruCtl, SERIAL_ENTRIESPERPAGE);
833 : :
834 : : /*
835 : : * Create or attach to the SerialControl structure.
836 : : */
2130 tgl@sss.pgh.pa.us 837 : 1150 : serialControl = (SerialControl)
838 : 1150 : ShmemInitStruct("SerialControlData", sizeof(SerialControlData), &found);
839 : :
3156 840 [ - + ]: 1150 : Assert(found == IsUnderPostmaster);
5515 heikki.linnakangas@i 841 [ + - ]: 1150 : if (!found)
842 : : {
843 : : /*
844 : : * Set control information to reflect empty SLRU.
845 : : */
775 alvherre@alvh.no-ip. 846 : 1150 : LWLockAcquire(SerialControlLock, LW_EXCLUSIVE);
2130 tgl@sss.pgh.pa.us 847 : 1150 : serialControl->headPage = -1;
848 : 1150 : serialControl->headXid = InvalidTransactionId;
849 : 1150 : serialControl->tailXid = InvalidTransactionId;
775 alvherre@alvh.no-ip. 850 : 1150 : LWLockRelease(SerialControlLock);
851 : : }
5515 heikki.linnakangas@i 852 : 1150 : }
853 : :
854 : : /*
855 : : * GUC check_hook for serializable_buffers
856 : : */
857 : : bool
746 alvherre@alvh.no-ip. 858 : 1184 : check_serial_buffers(int *newval, void **extra, GucSource source)
859 : : {
860 : 1184 : return check_slru_buffers("serializable_buffers", newval);
861 : : }
862 : :
863 : : /*
864 : : * Record a committed read write serializable xid and the minimum
865 : : * commitSeqNo of any transactions to which this xid had a rw-conflict out.
866 : : * An invalid commitSeqNo means that there were no conflicts out from xid.
867 : : */
868 : : static void
2130 tgl@sss.pgh.pa.us 869 :UBC 0 : SerialAdd(TransactionId xid, SerCommitSeqNo minConflictCommitSeqNo)
870 : : {
871 : : TransactionId tailXid;
872 : : int64 targetPage;
873 : : int slotno;
874 : : int64 firstZeroPage;
875 : : bool isNewPage;
876 : : LWLock *lock;
877 : :
5515 heikki.linnakangas@i 878 [ # # ]: 0 : Assert(TransactionIdIsValid(xid));
879 : :
2130 tgl@sss.pgh.pa.us 880 : 0 : targetPage = SerialPage(xid);
746 alvherre@alvh.no-ip. 881 : 0 : lock = SimpleLruGetBankLock(SerialSlruCtl, targetPage);
882 : :
883 : : /*
884 : : * In this routine, we must hold both SerialControlLock and the SLRU bank
885 : : * lock simultaneously while making the SLRU data catch up with the new
886 : : * state that we determine.
887 : : */
775 888 : 0 : LWLockAcquire(SerialControlLock, LW_EXCLUSIVE);
889 : :
890 : : /*
891 : : * If 'xid' is older than the global xmin (== tailXid), there's no need to
892 : : * store it, after all. This can happen if the oldest transaction holding
893 : : * back the global xmin just finished, making 'xid' uninteresting, but
894 : : * ClearOldPredicateLocks() has not yet run.
895 : : */
2130 tgl@sss.pgh.pa.us 896 : 0 : tailXid = serialControl->tailXid;
510 heikki.linnakangas@i 897 [ # # # # ]: 0 : if (!TransactionIdIsValid(tailXid) || TransactionIdPrecedes(xid, tailXid))
898 : : {
899 : 0 : LWLockRelease(SerialControlLock);
900 : 0 : return;
901 : : }
902 : :
903 : : /*
904 : : * If the SLRU is currently unused, zero out the whole active region from
905 : : * tailXid to headXid before taking it into use. Otherwise zero out only
906 : : * any new pages that enter the tailXid-headXid range as we advance
907 : : * headXid.
908 : : */
2130 tgl@sss.pgh.pa.us 909 [ # # ]: 0 : if (serialControl->headPage < 0)
910 : : {
911 : 0 : firstZeroPage = SerialPage(tailXid);
5515 heikki.linnakangas@i 912 : 0 : isNewPage = true;
913 : : }
914 : : else
915 : : {
2130 tgl@sss.pgh.pa.us 916 [ # # ]: 0 : firstZeroPage = SerialNextPage(serialControl->headPage);
917 : 0 : isNewPage = SerialPagePrecedesLogically(serialControl->headPage,
918 : : targetPage);
919 : : }
920 : :
921 [ # # ]: 0 : if (!TransactionIdIsValid(serialControl->headXid)
922 [ # # ]: 0 : || TransactionIdFollows(xid, serialControl->headXid))
923 : 0 : serialControl->headXid = xid;
5486 heikki.linnakangas@i 924 [ # # ]: 0 : if (isNewPage)
2130 tgl@sss.pgh.pa.us 925 : 0 : serialControl->headPage = targetPage;
926 : :
5515 heikki.linnakangas@i 927 [ # # ]: 0 : if (isNewPage)
928 : : {
929 : : /* Initialize intervening pages; might involve trading locks */
930 : : for (;;)
931 : : {
711 alvherre@alvh.no-ip. 932 : 0 : lock = SimpleLruGetBankLock(SerialSlruCtl, firstZeroPage);
933 : 0 : LWLockAcquire(lock, LW_EXCLUSIVE);
934 : 0 : slotno = SimpleLruZeroPage(SerialSlruCtl, firstZeroPage);
935 [ # # ]: 0 : if (firstZeroPage == targetPage)
936 : 0 : break;
2130 tgl@sss.pgh.pa.us 937 [ # # ]: 0 : firstZeroPage = SerialNextPage(firstZeroPage);
711 alvherre@alvh.no-ip. 938 : 0 : LWLockRelease(lock);
939 : : }
940 : : }
941 : : else
942 : : {
943 : 0 : LWLockAcquire(lock, LW_EXCLUSIVE);
2 heikki.linnakangas@i 944 :UNC 0 : slotno = SimpleLruReadPage(SerialSlruCtl, targetPage, true, &xid);
945 : : }
946 : :
2130 tgl@sss.pgh.pa.us 947 :UBC 0 : SerialValue(slotno, xid) = minConflictCommitSeqNo;
948 : 0 : SerialSlruCtl->shared->page_dirty[slotno] = true;
949 : :
746 alvherre@alvh.no-ip. 950 : 0 : LWLockRelease(lock);
775 951 : 0 : LWLockRelease(SerialControlLock);
952 : : }
953 : :
954 : : /*
955 : : * Get the minimum commitSeqNo for any conflict out for the given xid. For
956 : : * a transaction which exists but has no conflict out, InvalidSerCommitSeqNo
957 : : * will be returned.
958 : : */
959 : : static SerCommitSeqNo
2130 tgl@sss.pgh.pa.us 960 :CBC 24 : SerialGetMinConflictCommitSeqNo(TransactionId xid)
961 : : {
962 : : TransactionId headXid;
963 : : TransactionId tailXid;
964 : : SerCommitSeqNo val;
965 : : int slotno;
966 : :
5515 heikki.linnakangas@i 967 [ - + ]: 24 : Assert(TransactionIdIsValid(xid));
968 : :
775 alvherre@alvh.no-ip. 969 : 24 : LWLockAcquire(SerialControlLock, LW_SHARED);
2130 tgl@sss.pgh.pa.us 970 : 24 : headXid = serialControl->headXid;
971 : 24 : tailXid = serialControl->tailXid;
775 alvherre@alvh.no-ip. 972 : 24 : LWLockRelease(SerialControlLock);
973 : :
5515 heikki.linnakangas@i 974 [ + - ]: 24 : if (!TransactionIdIsValid(headXid))
975 : 24 : return 0;
976 : :
5515 heikki.linnakangas@i 977 [ # # ]:UBC 0 : Assert(TransactionIdIsValid(tailXid));
978 : :
979 [ # # ]: 0 : if (TransactionIdPrecedes(xid, tailXid)
980 [ # # ]: 0 : || TransactionIdFollows(xid, headXid))
981 : 0 : return 0;
982 : :
983 : : /*
984 : : * The following function must be called without holding SLRU bank lock,
985 : : * but will return with that lock held, which must then be released.
986 : : */
2130 tgl@sss.pgh.pa.us 987 : 0 : slotno = SimpleLruReadPage_ReadOnly(SerialSlruCtl,
2 heikki.linnakangas@i 988 :UNC 0 : SerialPage(xid), &xid);
2130 tgl@sss.pgh.pa.us 989 :UBC 0 : val = SerialValue(slotno, xid);
746 alvherre@alvh.no-ip. 990 : 0 : LWLockRelease(SimpleLruGetBankLock(SerialSlruCtl, SerialPage(xid)));
5515 heikki.linnakangas@i 991 : 0 : return val;
992 : : }
993 : :
994 : : /*
995 : : * Call this whenever there is a new xmin for active serializable
996 : : * transactions. We don't need to keep information on transactions which
997 : : * precede that. InvalidTransactionId means none active, so everything in
998 : : * the SLRU can be discarded.
999 : : */
1000 : : static void
2130 tgl@sss.pgh.pa.us 1001 :CBC 1715 : SerialSetActiveSerXmin(TransactionId xid)
1002 : : {
775 alvherre@alvh.no-ip. 1003 : 1715 : LWLockAcquire(SerialControlLock, LW_EXCLUSIVE);
1004 : :
1005 : : /*
1006 : : * When no sxacts are active, nothing overlaps, set the xid values to
1007 : : * invalid to show that there are no valid entries. Don't clear headPage,
1008 : : * though. A new xmin might still land on that page, and we don't want to
1009 : : * repeatedly zero out the same page.
1010 : : */
5515 heikki.linnakangas@i 1011 [ + + ]: 1715 : if (!TransactionIdIsValid(xid))
1012 : : {
2130 tgl@sss.pgh.pa.us 1013 : 847 : serialControl->tailXid = InvalidTransactionId;
1014 : 847 : serialControl->headXid = InvalidTransactionId;
775 alvherre@alvh.no-ip. 1015 : 847 : LWLockRelease(SerialControlLock);
5515 heikki.linnakangas@i 1016 : 847 : return;
1017 : : }
1018 : :
1019 : : /*
1020 : : * When we're recovering prepared transactions, the global xmin might move
1021 : : * backwards depending on the order they're recovered. Normally that's not
1022 : : * OK, but during recovery no serializable transactions will commit, so
1023 : : * the SLRU is empty and we can get away with it.
1024 : : */
1025 [ - + ]: 868 : if (RecoveryInProgress())
1026 : : {
2130 tgl@sss.pgh.pa.us 1027 [ # # ]:UBC 0 : Assert(serialControl->headPage < 0);
1028 [ # # ]: 0 : if (!TransactionIdIsValid(serialControl->tailXid)
1029 [ # # ]: 0 : || TransactionIdPrecedes(xid, serialControl->tailXid))
1030 : : {
1031 : 0 : serialControl->tailXid = xid;
1032 : : }
775 alvherre@alvh.no-ip. 1033 : 0 : LWLockRelease(SerialControlLock);
5515 heikki.linnakangas@i 1034 : 0 : return;
1035 : : }
1036 : :
2130 tgl@sss.pgh.pa.us 1037 [ + + - + ]:CBC 868 : Assert(!TransactionIdIsValid(serialControl->tailXid)
1038 : : || TransactionIdFollows(xid, serialControl->tailXid));
1039 : :
1040 : 868 : serialControl->tailXid = xid;
1041 : :
775 alvherre@alvh.no-ip. 1042 : 868 : LWLockRelease(SerialControlLock);
1043 : : }
1044 : :
1045 : : /*
1046 : : * Perform a checkpoint --- either during shutdown, or on-the-fly
1047 : : *
1048 : : * We don't have any data that needs to survive a restart, but this is a
1049 : : * convenient place to truncate the SLRU.
1050 : : */
1051 : : void
5486 heikki.linnakangas@i 1052 : 1802 : CheckPointPredicate(void)
1053 : : {
1054 : : int64 truncateCutoffPage;
1055 : :
775 alvherre@alvh.no-ip. 1056 : 1802 : LWLockAcquire(SerialControlLock, LW_EXCLUSIVE);
1057 : :
1058 : : /* Exit quickly if the SLRU is currently not in use. */
2130 tgl@sss.pgh.pa.us 1059 [ + - ]: 1802 : if (serialControl->headPage < 0)
1060 : : {
775 alvherre@alvh.no-ip. 1061 : 1802 : LWLockRelease(SerialControlLock);
5515 heikki.linnakangas@i 1062 : 1802 : return;
1063 : : }
1064 : :
2130 tgl@sss.pgh.pa.us 1065 [ # # ]:UBC 0 : if (TransactionIdIsValid(serialControl->tailXid))
1066 : : {
1067 : : int64 tailPage;
1068 : :
1069 : 0 : tailPage = SerialPage(serialControl->tailXid);
1070 : :
1071 : : /*
1072 : : * It is possible for the tailXid to be ahead of the headXid. This
1073 : : * occurs if we checkpoint while there are in-progress serializable
1074 : : * transaction(s) advancing the tail but we are yet to summarize the
1075 : : * transactions. In this case, we cutoff up to the headPage and the
1076 : : * next summary will advance the headXid.
1077 : : */
880 michael@paquier.xyz 1078 [ # # ]: 0 : if (SerialPagePrecedesLogically(tailPage, serialControl->headPage))
1079 : : {
1080 : : /* We can truncate the SLRU up to the page containing tailXid */
1081 : 0 : truncateCutoffPage = tailPage;
1082 : : }
1083 : : else
1084 : 0 : truncateCutoffPage = serialControl->headPage;
1085 : : }
1086 : : else
1087 : : {
1088 : : /*----------
1089 : : * The SLRU is no longer needed. Truncate to head before we set head
1090 : : * invalid.
1091 : : *
1092 : : * XXX: It's possible that the SLRU is not needed again until XID
1093 : : * wrap-around has happened, so that the segment containing headPage
1094 : : * that we leave behind will appear to be new again. In that case it
1095 : : * won't be removed until XID horizon advances enough to make it
1096 : : * current again.
1097 : : *
1098 : : * XXX: This should happen in vac_truncate_clog(), not in checkpoints.
1099 : : * Consider this scenario, starting from a system with no in-progress
1100 : : * transactions and VACUUM FREEZE having maximized oldestXact:
1101 : : * - Start a SERIALIZABLE transaction.
1102 : : * - Start, finish, and summarize a SERIALIZABLE transaction, creating
1103 : : * one SLRU page.
1104 : : * - Consume XIDs to reach xidStopLimit.
1105 : : * - Finish all transactions. Due to the long-running SERIALIZABLE
1106 : : * transaction, earlier checkpoints did not touch headPage. The
1107 : : * next checkpoint will change it, but that checkpoint happens after
1108 : : * the end of the scenario.
1109 : : * - VACUUM to advance XID limits.
1110 : : * - Consume ~2M XIDs, crossing the former xidWrapLimit.
1111 : : * - Start, finish, and summarize a SERIALIZABLE transaction.
1112 : : * SerialAdd() declines to create the targetPage, because headPage
1113 : : * is not regarded as in the past relative to that targetPage. The
1114 : : * transaction instigating the summarize fails in
1115 : : * SimpleLruReadPage().
1116 : : */
1117 : 0 : truncateCutoffPage = serialControl->headPage;
2130 tgl@sss.pgh.pa.us 1118 : 0 : serialControl->headPage = -1;
1119 : : }
1120 : :
775 alvherre@alvh.no-ip. 1121 : 0 : LWLockRelease(SerialControlLock);
1122 : :
1123 : : /*
1124 : : * Truncate away pages that are no longer required. Note that no
1125 : : * additional locking is required, because this is only called as part of
1126 : : * a checkpoint, and the validity limits have already been determined.
1127 : : */
880 michael@paquier.xyz 1128 : 0 : SimpleLruTruncate(SerialSlruCtl, truncateCutoffPage);
1129 : :
1130 : : /*
1131 : : * Write dirty SLRU pages to disk
1132 : : *
1133 : : * This is not actually necessary from a correctness point of view. We do
1134 : : * it merely as a debugging aid.
1135 : : *
1136 : : * We're doing this after the truncation to avoid writing pages right
1137 : : * before deleting the file in which they sit, which would be completely
1138 : : * pointless.
1139 : : */
1997 tmunro@postgresql.or 1140 : 0 : SimpleLruWriteAll(SerialSlruCtl, true);
1141 : : }
1142 : :
1143 : : /*------------------------------------------------------------------------*/
1144 : :
1145 : : /*
1146 : : * PredicateLockShmemInit -- Initialize the predicate locking data structures.
1147 : : *
1148 : : * This is called from CreateSharedMemoryAndSemaphores(), which see for
1149 : : * more comments. In the normal postmaster case, the shared hash tables
1150 : : * are created here. Backends inherit the pointers
1151 : : * to the shared tables via fork(). In the EXEC_BACKEND case, each
1152 : : * backend re-executes this code to obtain pointers to the already existing
1153 : : * shared hash tables.
1154 : : */
1155 : : void
563 heikki.linnakangas@i 1156 :CBC 1150 : PredicateLockShmemInit(void)
1157 : : {
1158 : : HASHCTL info;
1159 : : int64 max_table_size;
1160 : : Size requestSize;
1161 : : bool found;
1162 : :
1163 : : #ifndef EXEC_BACKEND
3156 tgl@sss.pgh.pa.us 1164 [ - + ]: 1150 : Assert(!IsUnderPostmaster);
1165 : : #endif
1166 : :
1167 : : /*
1168 : : * Compute size of predicate lock target hashtable. Note these
1169 : : * calculations must agree with PredicateLockShmemSize!
1170 : : */
5515 heikki.linnakangas@i 1171 : 1150 : max_table_size = NPREDICATELOCKTARGETENTS();
1172 : :
1173 : : /*
1174 : : * Allocate hash table for PREDICATELOCKTARGET structs. This stores
1175 : : * per-predicate-lock-target information.
1176 : : */
1177 : 1150 : info.keysize = sizeof(PREDICATELOCKTARGETTAG);
1178 : 1150 : info.entrysize = sizeof(PREDICATELOCKTARGET);
1179 : 1150 : info.num_partitions = NUM_PREDICATELOCK_PARTITIONS;
1180 : :
1181 : 1150 : PredicateLockTargetHash = ShmemInitHash("PREDICATELOCKTARGET hash",
1182 : : max_table_size,
1183 : : max_table_size,
1184 : : &info,
1185 : : HASH_ELEM | HASH_BLOBS |
1186 : : HASH_PARTITION | HASH_FIXED_SIZE);
1187 : :
1188 : : /*
1189 : : * Reserve a dummy entry in the hash table; we use it to make sure there's
1190 : : * always one entry available when we need to split or combine a page,
1191 : : * because running out of space there could mean aborting a
1192 : : * non-serializable transaction.
1193 : : */
3156 tgl@sss.pgh.pa.us 1194 [ + - ]: 1150 : if (!IsUnderPostmaster)
1195 : : {
1196 : 1150 : (void) hash_search(PredicateLockTargetHash, &ScratchTargetTag,
1197 : : HASH_ENTER, &found);
1198 [ - + ]: 1150 : Assert(!found);
1199 : : }
1200 : :
1201 : : /* Pre-calculate the hash and partition lock of the scratch entry */
1202 : 1150 : ScratchTargetTagHash = PredicateLockTargetTagHashCode(&ScratchTargetTag);
1203 : 1150 : ScratchPartitionLock = PredicateLockHashPartitionLock(ScratchTargetTagHash);
1204 : :
1205 : : /*
1206 : : * Allocate hash table for PREDICATELOCK structs. This stores per
1207 : : * xact-lock-of-a-target information.
1208 : : */
5515 heikki.linnakangas@i 1209 : 1150 : info.keysize = sizeof(PREDICATELOCKTAG);
1210 : 1150 : info.entrysize = sizeof(PREDICATELOCK);
1211 : 1150 : info.hash = predicatelock_hash;
1212 : 1150 : info.num_partitions = NUM_PREDICATELOCK_PARTITIONS;
1213 : :
1214 : : /* Assume an average of 2 xacts per target */
3156 tgl@sss.pgh.pa.us 1215 : 1150 : max_table_size *= 2;
1216 : :
5515 heikki.linnakangas@i 1217 : 1150 : PredicateLockHash = ShmemInitHash("PREDICATELOCK hash",
1218 : : max_table_size,
1219 : : max_table_size,
1220 : : &info,
1221 : : HASH_ELEM | HASH_FUNCTION |
1222 : : HASH_PARTITION | HASH_FIXED_SIZE);
1223 : :
1224 : : /*
1225 : : * Compute size for serializable transaction hashtable. Note these
1226 : : * calculations must agree with PredicateLockShmemSize!
1227 : : */
1433 rhaas@postgresql.org 1228 : 1150 : max_table_size = (MaxBackends + max_prepared_xacts);
1229 : :
1230 : : /*
1231 : : * Allocate a list to hold information on transactions participating in
1232 : : * predicate locking.
1233 : : *
1234 : : * Assume an average of 10 predicate locking transactions per backend.
1235 : : * This allows aggressive cleanup while detail is present before data must
1236 : : * be summarized for storage in SLRU and the "dummy" transaction.
1237 : : */
5515 heikki.linnakangas@i 1238 : 1150 : max_table_size *= 10;
1239 : :
347 tomas.vondra@postgre 1240 : 1150 : requestSize = add_size(PredXactListDataSize,
1241 : : (mul_size((Size) max_table_size,
1242 : : sizeof(SERIALIZABLEXACT))));
1243 : :
5515 heikki.linnakangas@i 1244 : 1150 : PredXact = ShmemInitStruct("PredXactList",
1245 : : requestSize,
1246 : : &found);
3156 tgl@sss.pgh.pa.us 1247 [ - + ]: 1150 : Assert(found == IsUnderPostmaster);
5515 heikki.linnakangas@i 1248 [ + - ]: 1150 : if (!found)
1249 : : {
1250 : : int i;
1251 : :
1252 : : /* clean everything, both the header and the element */
347 tomas.vondra@postgre 1253 : 1150 : memset(PredXact, 0, requestSize);
1254 : :
1151 andres@anarazel.de 1255 : 1150 : dlist_init(&PredXact->availableList);
1256 : 1150 : dlist_init(&PredXact->activeList);
5515 heikki.linnakangas@i 1257 : 1150 : PredXact->SxactGlobalXmin = InvalidTransactionId;
1258 : 1150 : PredXact->SxactGlobalXminCount = 0;
1259 : 1150 : PredXact->WritableSxactCount = 0;
1260 : 1150 : PredXact->LastSxactCommitSeqNo = FirstNormalSerCommitSeqNo - 1;
1261 : 1150 : PredXact->CanPartialClearThrough = 0;
1262 : 1150 : PredXact->HavePartialClearedThrough = 0;
347 tomas.vondra@postgre 1263 : 1150 : PredXact->element
1264 : 1150 : = (SERIALIZABLEXACT *) ((char *) PredXact + PredXactListDataSize);
1265 : : /* Add all elements to available list, clean. */
5515 heikki.linnakangas@i 1266 [ + + ]: 1066670 : for (i = 0; i < max_table_size; i++)
1267 : : {
1151 andres@anarazel.de 1268 : 1065520 : LWLockInitialize(&PredXact->element[i].perXactPredicateListLock,
1269 : : LWTRANCHE_PER_XACT_PREDICATE_LIST);
1270 : 1065520 : dlist_push_tail(&PredXact->availableList, &PredXact->element[i].xactLink);
1271 : : }
5515 heikki.linnakangas@i 1272 : 1150 : PredXact->OldCommittedSxact = CreatePredXact();
1273 : 1150 : SetInvalidVirtualTransactionId(PredXact->OldCommittedSxact->vxid);
5365 1274 : 1150 : PredXact->OldCommittedSxact->prepareSeqNo = 0;
5515 1275 : 1150 : PredXact->OldCommittedSxact->commitSeqNo = 0;
1276 : 1150 : PredXact->OldCommittedSxact->SeqNo.lastCommitBeforeSnapshot = 0;
1151 andres@anarazel.de 1277 : 1150 : dlist_init(&PredXact->OldCommittedSxact->outConflicts);
1278 : 1150 : dlist_init(&PredXact->OldCommittedSxact->inConflicts);
1279 : 1150 : dlist_init(&PredXact->OldCommittedSxact->predicateLocks);
1280 : 1150 : dlist_node_init(&PredXact->OldCommittedSxact->finishedLink);
1281 : 1150 : dlist_init(&PredXact->OldCommittedSxact->possibleUnsafeConflicts);
5515 heikki.linnakangas@i 1282 : 1150 : PredXact->OldCommittedSxact->topXid = InvalidTransactionId;
1283 : 1150 : PredXact->OldCommittedSxact->finishedBefore = InvalidTransactionId;
1284 : 1150 : PredXact->OldCommittedSxact->xmin = InvalidTransactionId;
1285 : 1150 : PredXact->OldCommittedSxact->flags = SXACT_FLAG_COMMITTED;
1286 : 1150 : PredXact->OldCommittedSxact->pid = 0;
742 1287 : 1150 : PredXact->OldCommittedSxact->pgprocno = INVALID_PROC_NUMBER;
1288 : : }
1289 : : /* This never changes, so let's keep a local copy. */
5515 1290 : 1150 : OldCommittedSxact = PredXact->OldCommittedSxact;
1291 : :
1292 : : /*
1293 : : * Allocate hash table for SERIALIZABLEXID structs. This stores per-xid
1294 : : * information for serializable transactions which have accessed data.
1295 : : */
1296 : 1150 : info.keysize = sizeof(SERIALIZABLEXIDTAG);
1297 : 1150 : info.entrysize = sizeof(SERIALIZABLEXID);
1298 : :
1299 : 1150 : SerializableXidHash = ShmemInitHash("SERIALIZABLEXID hash",
1300 : : max_table_size,
1301 : : max_table_size,
1302 : : &info,
1303 : : HASH_ELEM | HASH_BLOBS |
1304 : : HASH_FIXED_SIZE);
1305 : :
1306 : : /*
1307 : : * Allocate space for tracking rw-conflicts in lists attached to the
1308 : : * transactions.
1309 : : *
1310 : : * Assume an average of 5 conflicts per transaction. Calculations suggest
1311 : : * that this will prevent resource exhaustion in even the most pessimal
1312 : : * loads up to max_connections = 200 with all 200 connections pounding the
1313 : : * database with serializable transactions. Beyond that, there may be
1314 : : * occasional transactions canceled when trying to flag conflicts. That's
1315 : : * probably OK.
1316 : : */
1317 : 1150 : max_table_size *= 5;
1318 : :
347 tomas.vondra@postgre 1319 : 1150 : requestSize = RWConflictPoolHeaderDataSize +
1320 : 1150 : mul_size((Size) max_table_size,
1321 : : RWConflictDataSize);
1322 : :
5515 heikki.linnakangas@i 1323 : 1150 : RWConflictPool = ShmemInitStruct("RWConflictPool",
1324 : : requestSize,
1325 : : &found);
3156 tgl@sss.pgh.pa.us 1326 [ - + ]: 1150 : Assert(found == IsUnderPostmaster);
5515 heikki.linnakangas@i 1327 [ + - ]: 1150 : if (!found)
1328 : : {
1329 : : int i;
1330 : :
1331 : : /* clean everything, including the elements */
347 tomas.vondra@postgre 1332 : 1150 : memset(RWConflictPool, 0, requestSize);
1333 : :
1151 andres@anarazel.de 1334 : 1150 : dlist_init(&RWConflictPool->availableList);
347 tomas.vondra@postgre 1335 : 1150 : RWConflictPool->element = (RWConflict) ((char *) RWConflictPool +
1336 : : RWConflictPoolHeaderDataSize);
1337 : : /* Add all elements to available list, clean. */
5515 heikki.linnakangas@i 1338 [ + + ]: 5328750 : for (i = 0; i < max_table_size; i++)
1339 : : {
1151 andres@anarazel.de 1340 : 5327600 : dlist_push_tail(&RWConflictPool->availableList,
1341 : 5327600 : &RWConflictPool->element[i].outLink);
1342 : : }
1343 : : }
1344 : :
1345 : : /*
1346 : : * Create or attach to the header for the list of finished serializable
1347 : : * transactions.
1348 : : */
1349 : 1150 : FinishedSerializableTransactions = (dlist_head *)
5515 heikki.linnakangas@i 1350 : 1150 : ShmemInitStruct("FinishedSerializableTransactions",
1351 : : sizeof(dlist_head),
1352 : : &found);
3156 tgl@sss.pgh.pa.us 1353 [ - + ]: 1150 : Assert(found == IsUnderPostmaster);
5515 heikki.linnakangas@i 1354 [ + - ]: 1150 : if (!found)
1151 andres@anarazel.de 1355 : 1150 : dlist_init(FinishedSerializableTransactions);
1356 : :
1357 : : /*
1358 : : * Initialize the SLRU storage for old committed serializable
1359 : : * transactions.
1360 : : */
2130 tgl@sss.pgh.pa.us 1361 : 1150 : SerialInit();
5515 heikki.linnakangas@i 1362 : 1150 : }
1363 : :
1364 : : /*
1365 : : * Estimate shared-memory space used for predicate lock table
1366 : : */
1367 : : Size
1368 : 2147 : PredicateLockShmemSize(void)
1369 : : {
1370 : 2147 : Size size = 0;
1371 : : long max_table_size;
1372 : :
1373 : : /* predicate lock target hash table */
1374 : 2147 : max_table_size = NPREDICATELOCKTARGETENTS();
1375 : 2147 : size = add_size(size, hash_estimate_size(max_table_size,
1376 : : sizeof(PREDICATELOCKTARGET)));
1377 : :
1378 : : /* predicate lock hash table */
1379 : 2147 : max_table_size *= 2;
1380 : 2147 : size = add_size(size, hash_estimate_size(max_table_size,
1381 : : sizeof(PREDICATELOCK)));
1382 : :
1383 : : /*
1384 : : * Since NPREDICATELOCKTARGETENTS is only an estimate, add 10% safety
1385 : : * margin.
1386 : : */
1387 : 2147 : size = add_size(size, size / 10);
1388 : :
1389 : : /* transaction list */
1433 rhaas@postgresql.org 1390 : 2147 : max_table_size = MaxBackends + max_prepared_xacts;
5515 heikki.linnakangas@i 1391 : 2147 : max_table_size *= 10;
1392 : 2147 : size = add_size(size, PredXactListDataSize);
1393 : 2147 : size = add_size(size, mul_size((Size) max_table_size,
1394 : : sizeof(SERIALIZABLEXACT)));
1395 : :
1396 : : /* transaction xid table */
1397 : 2147 : size = add_size(size, hash_estimate_size(max_table_size,
1398 : : sizeof(SERIALIZABLEXID)));
1399 : :
1400 : : /* rw-conflict pool */
5513 1401 : 2147 : max_table_size *= 5;
1402 : 2147 : size = add_size(size, RWConflictPoolHeaderDataSize);
1403 : 2147 : size = add_size(size, mul_size((Size) max_table_size,
1404 : : RWConflictDataSize));
1405 : :
1406 : : /* Head for list of finished serializable transactions. */
1151 andres@anarazel.de 1407 : 2147 : size = add_size(size, sizeof(dlist_head));
1408 : :
1409 : : /* Shared memory structures for SLRU tracking of old committed xids. */
2130 tgl@sss.pgh.pa.us 1410 : 2147 : size = add_size(size, sizeof(SerialControlData));
746 alvherre@alvh.no-ip. 1411 : 2147 : size = add_size(size, SimpleLruShmemSize(serializable_buffers, 0));
1412 : :
5515 heikki.linnakangas@i 1413 : 2147 : return size;
1414 : : }
1415 : :
1416 : :
1417 : : /*
1418 : : * Compute the hash code associated with a PREDICATELOCKTAG.
1419 : : *
1420 : : * Because we want to use just one set of partition locks for both the
1421 : : * PREDICATELOCKTARGET and PREDICATELOCK hash tables, we have to make sure
1422 : : * that PREDICATELOCKs fall into the same partition number as their
1423 : : * associated PREDICATELOCKTARGETs. dynahash.c expects the partition number
1424 : : * to be the low-order bits of the hash code, and therefore a
1425 : : * PREDICATELOCKTAG's hash code must have the same low-order bits as the
1426 : : * associated PREDICATELOCKTARGETTAG's hash code. We achieve this with this
1427 : : * specialized hash function.
1428 : : */
1429 : : static uint32
5515 heikki.linnakangas@i 1430 :UBC 0 : predicatelock_hash(const void *key, Size keysize)
1431 : : {
5380 tgl@sss.pgh.pa.us 1432 : 0 : const PREDICATELOCKTAG *predicatelocktag = (const PREDICATELOCKTAG *) key;
1433 : : uint32 targethash;
1434 : :
5515 heikki.linnakangas@i 1435 [ # # ]: 0 : Assert(keysize == sizeof(PREDICATELOCKTAG));
1436 : :
1437 : : /* Look into the associated target object, and compute its hash code */
1438 : 0 : targethash = PredicateLockTargetTagHashCode(&predicatelocktag->myTarget->tag);
1439 : :
1440 : 0 : return PredicateLockHashCodeFromTargetHashCode(predicatelocktag, targethash);
1441 : : }
1442 : :
1443 : :
1444 : : /*
1445 : : * GetPredicateLockStatusData
1446 : : * Return a table containing the internal state of the predicate
1447 : : * lock manager for use in pg_lock_status.
1448 : : *
1449 : : * Like GetLockStatusData, this function tries to hold the partition LWLocks
1450 : : * for as short a time as possible by returning two arrays that simply
1451 : : * contain the PREDICATELOCKTARGETTAG and SERIALIZABLEXACT for each lock
1452 : : * table entry. Multiple copies of the same PREDICATELOCKTARGETTAG and
1453 : : * SERIALIZABLEXACT will likely appear.
1454 : : */
1455 : : PredicateLockData *
5515 heikki.linnakangas@i 1456 :CBC 234 : GetPredicateLockStatusData(void)
1457 : : {
1458 : : PredicateLockData *data;
1459 : : int i;
1460 : : int els,
1461 : : el;
1462 : : HASH_SEQ_STATUS seqstat;
1463 : : PREDICATELOCK *predlock;
1464 : :
95 michael@paquier.xyz 1465 :GNC 234 : data = palloc_object(PredicateLockData);
1466 : :
1467 : : /*
1468 : : * To ensure consistency, take simultaneous locks on all partition locks
1469 : : * in ascending order, then SerializableXactHashLock.
1470 : : */
5515 heikki.linnakangas@i 1471 [ + + ]:CBC 3978 : for (i = 0; i < NUM_PREDICATELOCK_PARTITIONS; i++)
4430 rhaas@postgresql.org 1472 : 3744 : LWLockAcquire(PredicateLockHashPartitionLockByIndex(i), LW_SHARED);
5515 heikki.linnakangas@i 1473 : 234 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
1474 : :
1475 : : /* Get number of locks and allocate appropriately-sized arrays. */
1476 : 234 : els = hash_get_num_entries(PredicateLockHash);
1477 : 234 : data->nelements = els;
95 michael@paquier.xyz 1478 :GNC 234 : data->locktags = palloc_array(PREDICATELOCKTARGETTAG, els);
1479 : 234 : data->xacts = palloc_array(SERIALIZABLEXACT, els);
1480 : :
1481 : :
1482 : : /* Scan through PredicateLockHash and copy contents */
5515 heikki.linnakangas@i 1483 :CBC 234 : hash_seq_init(&seqstat, PredicateLockHash);
1484 : :
1485 : 234 : el = 0;
1486 : :
1487 [ + + ]: 237 : while ((predlock = (PREDICATELOCK *) hash_seq_search(&seqstat)))
1488 : : {
1489 : 3 : data->locktags[el] = predlock->tag.myTarget->tag;
1490 : 3 : data->xacts[el] = *predlock->tag.myXact;
1491 : 3 : el++;
1492 : : }
1493 : :
1494 [ - + ]: 234 : Assert(el == els);
1495 : :
1496 : : /* Release locks in reverse order */
1497 : 234 : LWLockRelease(SerializableXactHashLock);
1498 [ + + ]: 3978 : for (i = NUM_PREDICATELOCK_PARTITIONS - 1; i >= 0; i--)
4430 rhaas@postgresql.org 1499 : 3744 : LWLockRelease(PredicateLockHashPartitionLockByIndex(i));
1500 : :
5515 heikki.linnakangas@i 1501 : 234 : return data;
1502 : : }
1503 : :
1504 : : /*
1505 : : * Free up shared memory structures by pushing the oldest sxact (the one at
1506 : : * the front of the SummarizeOldestCommittedSxact queue) into summary form.
1507 : : * Each call will free exactly one SERIALIZABLEXACT structure and may also
1508 : : * free one or more of these structures: SERIALIZABLEXID, PREDICATELOCK,
1509 : : * PREDICATELOCKTARGET, RWConflictData.
1510 : : */
1511 : : static void
5515 heikki.linnakangas@i 1512 :UBC 0 : SummarizeOldestCommittedSxact(void)
1513 : : {
1514 : : SERIALIZABLEXACT *sxact;
1515 : :
1516 : 0 : LWLockAcquire(SerializableFinishedListLock, LW_EXCLUSIVE);
1517 : :
1518 : : /*
1519 : : * This function is only called if there are no sxact slots available.
1520 : : * Some of them must belong to old, already-finished transactions, so
1521 : : * there should be something in FinishedSerializableTransactions list that
1522 : : * we can summarize. However, there's a race condition: while we were not
1523 : : * holding any locks, a transaction might have ended and cleaned up all
1524 : : * the finished sxact entries already, freeing up their sxact slots. In
1525 : : * that case, we have nothing to do here. The caller will find one of the
1526 : : * slots released by the other backend when it retries.
1527 : : */
1151 andres@anarazel.de 1528 [ # # ]: 0 : if (dlist_is_empty(FinishedSerializableTransactions))
1529 : : {
5515 heikki.linnakangas@i 1530 : 0 : LWLockRelease(SerializableFinishedListLock);
1531 : 0 : return;
1532 : : }
1533 : :
1534 : : /*
1535 : : * Grab the first sxact off the finished list -- this will be the earliest
1536 : : * commit. Remove it from the list.
1537 : : */
1151 andres@anarazel.de 1538 : 0 : sxact = dlist_head_element(SERIALIZABLEXACT, finishedLink,
1539 : : FinishedSerializableTransactions);
1540 : 0 : dlist_delete_thoroughly(&sxact->finishedLink);
1541 : :
1542 : : /* Add to SLRU summary information. */
5515 heikki.linnakangas@i 1543 [ # # # # ]: 0 : if (TransactionIdIsValid(sxact->topXid) && !SxactIsReadOnly(sxact))
2130 tgl@sss.pgh.pa.us 1544 [ # # ]: 0 : SerialAdd(sxact->topXid, SxactHasConflictOut(sxact)
1545 : : ? sxact->SeqNo.earliestOutConflictCommit : InvalidSerCommitSeqNo);
1546 : :
1547 : : /* Summarize and release the detail. */
5515 heikki.linnakangas@i 1548 : 0 : ReleaseOneSerializableXact(sxact, false, true);
1549 : :
1550 : 0 : LWLockRelease(SerializableFinishedListLock);
1551 : : }
1552 : :
1553 : : /*
1554 : : * GetSafeSnapshot
1555 : : * Obtain and register a snapshot for a READ ONLY DEFERRABLE
1556 : : * transaction. Ensures that the snapshot is "safe", i.e. a
1557 : : * read-only transaction running on it can execute serializably
1558 : : * without further checks. This requires waiting for concurrent
1559 : : * transactions to complete, and retrying with a new snapshot if
1560 : : * one of them could possibly create a conflict.
1561 : : *
1562 : : * As with GetSerializableTransactionSnapshot (which this is a subroutine
1563 : : * for), the passed-in Snapshot pointer should reference a static data
1564 : : * area that can safely be passed to GetSnapshotData.
1565 : : */
1566 : : static Snapshot
5515 heikki.linnakangas@i 1567 :CBC 4 : GetSafeSnapshot(Snapshot origSnapshot)
1568 : : {
1569 : : Snapshot snapshot;
1570 : :
1571 [ + - + - ]: 4 : Assert(XactReadOnly && XactDeferrable);
1572 : :
1573 : : while (true)
1574 : : {
1575 : : /*
1576 : : * GetSerializableTransactionSnapshotInt is going to call
1577 : : * GetSnapshotData, so we need to provide it the static snapshot area
1578 : : * our caller passed to us. The pointer returned is actually the same
1579 : : * one passed to it, but we avoid assuming that here.
1580 : : */
5258 tgl@sss.pgh.pa.us 1581 : 5 : snapshot = GetSerializableTransactionSnapshotInt(origSnapshot,
1582 : : NULL, InvalidPid);
1583 : :
5515 heikki.linnakangas@i 1584 [ + + ]: 5 : if (MySerializableXact == InvalidSerializableXact)
1585 : 2 : return snapshot; /* no concurrent r/w xacts; it's safe */
1586 : :
5392 1587 : 3 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
1588 : :
1589 : : /*
1590 : : * Wait for concurrent transactions to finish. Stop early if one of
1591 : : * them marked us as conflicted.
1592 : : */
1593 : 3 : MySerializableXact->flags |= SXACT_FLAG_DEFERRABLE_WAITING;
1151 andres@anarazel.de 1594 [ + + ]: 7 : while (!(dlist_is_empty(&MySerializableXact->possibleUnsafeConflicts) ||
5515 heikki.linnakangas@i 1595 [ + - ]: 4 : SxactIsROUnsafe(MySerializableXact)))
1596 : : {
5392 1597 : 4 : LWLockRelease(SerializableXactHashLock);
3449 rhaas@postgresql.org 1598 : 4 : ProcWaitForSignal(WAIT_EVENT_SAFE_SNAPSHOT);
5392 heikki.linnakangas@i 1599 : 4 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
1600 : : }
5515 1601 : 3 : MySerializableXact->flags &= ~SXACT_FLAG_DEFERRABLE_WAITING;
1602 : :
1603 [ + + ]: 3 : if (!SxactIsROUnsafe(MySerializableXact))
1604 : : {
5392 1605 : 2 : LWLockRelease(SerializableXactHashLock);
5515 1606 : 2 : break; /* success */
1607 : : }
1608 : :
5392 1609 : 1 : LWLockRelease(SerializableXactHashLock);
1610 : :
1611 : : /* else, need to retry... */
5515 1612 [ - + ]: 1 : ereport(DEBUG2,
1613 : : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1614 : : errmsg_internal("deferrable snapshot was unsafe; trying a new one")));
2557 tmunro@postgresql.or 1615 : 1 : ReleasePredicateLocks(false, false);
1616 : : }
1617 : :
1618 : : /*
1619 : : * Now we have a safe snapshot, so we don't need to do any further checks.
1620 : : */
5515 heikki.linnakangas@i 1621 [ - + ]: 2 : Assert(SxactIsROSafe(MySerializableXact));
2557 tmunro@postgresql.or 1622 : 2 : ReleasePredicateLocks(false, true);
1623 : :
5515 heikki.linnakangas@i 1624 : 2 : return snapshot;
1625 : : }
1626 : :
1627 : : /*
1628 : : * GetSafeSnapshotBlockingPids
1629 : : * If the specified process is currently blocked in GetSafeSnapshot,
1630 : : * write the process IDs of all processes that it is blocked by
1631 : : * into the caller-supplied buffer output[]. The list is truncated at
1632 : : * output_size, and the number of PIDs written into the buffer is
1633 : : * returned. Returns zero if the given PID is not currently blocked
1634 : : * in GetSafeSnapshot.
1635 : : */
1636 : : int
3261 tgl@sss.pgh.pa.us 1637 : 789 : GetSafeSnapshotBlockingPids(int blocked_pid, int *output, int output_size)
1638 : : {
1639 : 789 : int num_written = 0;
1640 : : dlist_iter iter;
1131 andres@anarazel.de 1641 : 789 : SERIALIZABLEXACT *blocking_sxact = NULL;
1642 : :
3261 tgl@sss.pgh.pa.us 1643 : 789 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
1644 : :
1645 : : /* Find blocked_pid's SERIALIZABLEXACT by linear search. */
1151 andres@anarazel.de 1646 [ + - + + ]: 1737 : dlist_foreach(iter, &PredXact->activeList)
1647 : : {
1131 1648 : 1054 : SERIALIZABLEXACT *sxact =
1649 : 1054 : dlist_container(SERIALIZABLEXACT, xactLink, iter.cur);
1650 : :
3261 tgl@sss.pgh.pa.us 1651 [ + + ]: 1054 : if (sxact->pid == blocked_pid)
1652 : : {
1131 andres@anarazel.de 1653 : 106 : blocking_sxact = sxact;
3261 tgl@sss.pgh.pa.us 1654 : 106 : break;
1655 : : }
1656 : : }
1657 : :
1658 : : /* Did we find it, and is it currently waiting in GetSafeSnapshot? */
1131 andres@anarazel.de 1659 [ + + + + ]: 789 : if (blocking_sxact != NULL && SxactIsDeferrableWaiting(blocking_sxact))
1660 : : {
1661 : : /* Traverse the list of possible unsafe conflicts collecting PIDs. */
1662 [ + - + - ]: 2 : dlist_foreach(iter, &blocking_sxact->possibleUnsafeConflicts)
1663 : : {
1151 1664 : 2 : RWConflict possibleUnsafeConflict =
1031 tgl@sss.pgh.pa.us 1665 : 2 : dlist_container(RWConflictData, inLink, iter.cur);
1666 : :
3261 1667 : 2 : output[num_written++] = possibleUnsafeConflict->sxactOut->pid;
1668 : :
1131 andres@anarazel.de 1669 [ + - ]: 2 : if (num_written >= output_size)
1670 : 2 : break;
1671 : : }
1672 : : }
1673 : :
3261 tgl@sss.pgh.pa.us 1674 : 789 : LWLockRelease(SerializableXactHashLock);
1675 : :
1676 : 789 : return num_written;
1677 : : }
1678 : :
1679 : : /*
1680 : : * Acquire a snapshot that can be used for the current transaction.
1681 : : *
1682 : : * Make sure we have a SERIALIZABLEXACT reference in MySerializableXact.
1683 : : * It should be current for this process and be contained in PredXact.
1684 : : *
1685 : : * The passed-in Snapshot pointer should reference a static data area that
1686 : : * can safely be passed to GetSnapshotData. The return value is actually
1687 : : * always this same pointer; no new snapshot data structure is allocated
1688 : : * within this function.
1689 : : */
1690 : : Snapshot
5284 1691 : 1666 : GetSerializableTransactionSnapshot(Snapshot snapshot)
1692 : : {
5515 heikki.linnakangas@i 1693 [ - + ]: 1666 : Assert(IsolationIsSerializable());
1694 : :
1695 : : /*
1696 : : * Can't use serializable mode while recovery is still active, as it is,
1697 : : * for example, on a hot standby. We could get here despite the check in
1698 : : * check_transaction_isolation() if default_transaction_isolation is set
1699 : : * to serializable, so phrase the hint accordingly.
1700 : : */
4951 tgl@sss.pgh.pa.us 1701 [ - + ]: 1666 : if (RecoveryInProgress())
4951 tgl@sss.pgh.pa.us 1702 [ # # ]:UBC 0 : ereport(ERROR,
1703 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1704 : : errmsg("cannot use serializable mode in a hot standby"),
1705 : : errdetail("\"default_transaction_isolation\" is set to \"serializable\"."),
1706 : : errhint("You can use \"SET default_transaction_isolation = 'repeatable read'\" to change the default.")));
1707 : :
1708 : : /*
1709 : : * A special optimization is available for SERIALIZABLE READ ONLY
1710 : : * DEFERRABLE transactions -- we can wait for a suitable snapshot and
1711 : : * thereby avoid all SSI overhead once it's running.
1712 : : */
5515 heikki.linnakangas@i 1713 [ + + + + ]:CBC 1666 : if (XactReadOnly && XactDeferrable)
1714 : 4 : return GetSafeSnapshot(snapshot);
1715 : :
5258 tgl@sss.pgh.pa.us 1716 : 1662 : return GetSerializableTransactionSnapshotInt(snapshot,
1717 : : NULL, InvalidPid);
1718 : : }
1719 : :
1720 : : /*
1721 : : * Import a snapshot to be used for the current transaction.
1722 : : *
1723 : : * This is nearly the same as GetSerializableTransactionSnapshot, except that
1724 : : * we don't take a new snapshot, but rather use the data we're handed.
1725 : : *
1726 : : * The caller must have verified that the snapshot came from a serializable
1727 : : * transaction; and if we're read-write, the source transaction must not be
1728 : : * read-only.
1729 : : */
1730 : : void
1731 : 13 : SetSerializableTransactionSnapshot(Snapshot snapshot,
1732 : : VirtualTransactionId *sourcevxid,
1733 : : int sourcepid)
1734 : : {
1735 [ - + ]: 13 : Assert(IsolationIsSerializable());
1736 : :
1737 : : /*
1738 : : * If this is called by parallel.c in a parallel worker, we don't want to
1739 : : * create a SERIALIZABLEXACT just yet because the leader's
1740 : : * SERIALIZABLEXACT will be installed with AttachSerializableXact(). We
1741 : : * also don't want to reject SERIALIZABLE READ ONLY DEFERRABLE in this
1742 : : * case, because the leader has already determined that the snapshot it
1743 : : * has passed us is safe. So there is nothing for us to do.
1744 : : */
2557 tmunro@postgresql.or 1745 [ + - ]: 13 : if (IsParallelWorker())
1746 : 13 : return;
1747 : :
1748 : : /*
1749 : : * We do not allow SERIALIZABLE READ ONLY DEFERRABLE transactions to
1750 : : * import snapshots, since there's no way to wait for a safe snapshot when
1751 : : * we're using the snap we're told to. (XXX instead of throwing an error,
1752 : : * we could just ignore the XactDeferrable flag?)
1753 : : */
5258 tgl@sss.pgh.pa.us 1754 [ # # # # ]:UBC 0 : if (XactReadOnly && XactDeferrable)
1755 [ # # ]: 0 : ereport(ERROR,
1756 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1757 : : errmsg("a snapshot-importing transaction must not be READ ONLY DEFERRABLE")));
1758 : :
3196 andres@anarazel.de 1759 : 0 : (void) GetSerializableTransactionSnapshotInt(snapshot, sourcevxid,
1760 : : sourcepid);
1761 : : }
1762 : :
1763 : : /*
1764 : : * Guts of GetSerializableTransactionSnapshot
1765 : : *
1766 : : * If sourcevxid is valid, this is actually an import operation and we should
1767 : : * skip calling GetSnapshotData, because the snapshot contents are already
1768 : : * loaded up. HOWEVER: to avoid race conditions, we must check that the
1769 : : * source xact is still running after we acquire SerializableXactHashLock.
1770 : : * We do that by calling ProcArrayInstallImportedXmin.
1771 : : */
1772 : : static Snapshot
5258 tgl@sss.pgh.pa.us 1773 :CBC 1667 : GetSerializableTransactionSnapshotInt(Snapshot snapshot,
1774 : : VirtualTransactionId *sourcevxid,
1775 : : int sourcepid)
1776 : : {
1777 : : PGPROC *proc;
1778 : : VirtualTransactionId vxid;
1779 : : SERIALIZABLEXACT *sxact,
1780 : : *othersxact;
1781 : :
1782 : : /* We only do this for serializable transactions. Once. */
5515 heikki.linnakangas@i 1783 [ - + ]: 1667 : Assert(MySerializableXact == InvalidSerializableXact);
1784 : :
1785 [ - + ]: 1667 : Assert(!RecoveryInProgress());
1786 : :
1787 : : /*
1788 : : * Since all parts of a serializable transaction must use the same
1789 : : * snapshot, it is too late to establish one after a parallel operation
1790 : : * has begun.
1791 : : */
3972 rhaas@postgresql.org 1792 [ - + ]: 1667 : if (IsInParallelMode())
3972 rhaas@postgresql.org 1793 [ # # ]:UBC 0 : elog(ERROR, "cannot establish serializable snapshot during a parallel operation");
1794 : :
5515 heikki.linnakangas@i 1795 :CBC 1667 : proc = MyProc;
1796 [ - + ]: 1667 : Assert(proc != NULL);
1797 : 1667 : GET_VXID_FROM_PGPROC(vxid, *proc);
1798 : :
1799 : : /*
1800 : : * First we get the sxact structure, which may involve looping and access
1801 : : * to the "finished" list to free a structure for use.
1802 : : *
1803 : : * We must hold SerializableXactHashLock when taking/checking the snapshot
1804 : : * to avoid race conditions, for much the same reasons that
1805 : : * GetSnapshotData takes the ProcArrayLock. Since we might have to
1806 : : * release SerializableXactHashLock to call SummarizeOldestCommittedSxact,
1807 : : * this means we have to create the sxact first, which is a bit annoying
1808 : : * (in particular, an elog(ERROR) in procarray.c would cause us to leak
1809 : : * the sxact). Consider refactoring to avoid this.
1810 : : */
1811 : : #ifdef TEST_SUMMARIZE_SERIAL
1812 : : SummarizeOldestCommittedSxact();
1813 : : #endif
1814 : 1667 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
1815 : : do
1816 : : {
1817 : 1667 : sxact = CreatePredXact();
1818 : : /* If null, push out committed sxact to SLRU summary & retry. */
1819 [ - + ]: 1667 : if (!sxact)
1820 : : {
5515 heikki.linnakangas@i 1821 :UBC 0 : LWLockRelease(SerializableXactHashLock);
1822 : 0 : SummarizeOldestCommittedSxact();
1823 : 0 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
1824 : : }
5515 heikki.linnakangas@i 1825 [ - + ]:CBC 1667 : } while (!sxact);
1826 : :
1827 : : /* Get the snapshot, or check that it's safe to use */
3196 andres@anarazel.de 1828 [ + - ]: 1667 : if (!sourcevxid)
5258 tgl@sss.pgh.pa.us 1829 : 1667 : snapshot = GetSnapshotData(snapshot);
3196 andres@anarazel.de 1830 [ # # ]:UBC 0 : else if (!ProcArrayInstallImportedXmin(snapshot->xmin, sourcevxid))
1831 : : {
5258 tgl@sss.pgh.pa.us 1832 : 0 : ReleasePredXact(sxact);
1833 : 0 : LWLockRelease(SerializableXactHashLock);
1834 [ # # ]: 0 : ereport(ERROR,
1835 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1836 : : errmsg("could not import the requested snapshot"),
1837 : : errdetail("The source process with PID %d is not running anymore.",
1838 : : sourcepid)));
1839 : : }
1840 : :
1841 : : /*
1842 : : * If there are no serializable transactions which are not read-only, we
1843 : : * can "opt out" of predicate locking and conflict checking for a
1844 : : * read-only transaction.
1845 : : *
1846 : : * The reason this is safe is that a read-only transaction can only become
1847 : : * part of a dangerous structure if it overlaps a writable transaction
1848 : : * which in turn overlaps a writable transaction which committed before
1849 : : * the read-only transaction started. A new writable transaction can
1850 : : * overlap this one, but it can't meet the other condition of overlapping
1851 : : * a transaction which committed before this one started.
1852 : : */
5515 heikki.linnakangas@i 1853 [ + + + + ]:CBC 1667 : if (XactReadOnly && PredXact->WritableSxactCount == 0)
1854 : : {
1855 : 110 : ReleasePredXact(sxact);
1856 : 110 : LWLockRelease(SerializableXactHashLock);
1857 : 110 : return snapshot;
1858 : : }
1859 : :
1860 : : /* Initialize the structure. */
1861 : 1557 : sxact->vxid = vxid;
1862 : 1557 : sxact->SeqNo.lastCommitBeforeSnapshot = PredXact->LastSxactCommitSeqNo;
5365 1863 : 1557 : sxact->prepareSeqNo = InvalidSerCommitSeqNo;
5515 1864 : 1557 : sxact->commitSeqNo = InvalidSerCommitSeqNo;
1151 andres@anarazel.de 1865 : 1557 : dlist_init(&(sxact->outConflicts));
1866 : 1557 : dlist_init(&(sxact->inConflicts));
1867 : 1557 : dlist_init(&(sxact->possibleUnsafeConflicts));
5515 heikki.linnakangas@i 1868 : 1557 : sxact->topXid = GetTopTransactionIdIfAny();
1869 : 1557 : sxact->finishedBefore = InvalidTransactionId;
1870 : 1557 : sxact->xmin = snapshot->xmin;
1871 : 1557 : sxact->pid = MyProcPid;
752 1872 : 1557 : sxact->pgprocno = MyProcNumber;
1151 andres@anarazel.de 1873 : 1557 : dlist_init(&sxact->predicateLocks);
1874 : 1557 : dlist_node_init(&sxact->finishedLink);
5515 heikki.linnakangas@i 1875 : 1557 : sxact->flags = 0;
1876 [ + + ]: 1557 : if (XactReadOnly)
1877 : : {
1878 : : dlist_iter iter;
1879 : :
1880 : 108 : sxact->flags |= SXACT_FLAG_READ_ONLY;
1881 : :
1882 : : /*
1883 : : * Register all concurrent r/w transactions as possible conflicts; if
1884 : : * all of them commit without any outgoing conflicts to earlier
1885 : : * transactions then this snapshot can be deemed safe (and we can run
1886 : : * without tracking predicate locks).
1887 : : */
1151 andres@anarazel.de 1888 [ + - + + ]: 473 : dlist_foreach(iter, &PredXact->activeList)
1889 : : {
1890 : 365 : othersxact = dlist_container(SERIALIZABLEXACT, xactLink, iter.cur);
1891 : :
5364 heikki.linnakangas@i 1892 [ + + ]: 365 : if (!SxactIsCommitted(othersxact)
1893 [ + - ]: 244 : && !SxactIsDoomed(othersxact)
1894 [ + + ]: 244 : && !SxactIsReadOnly(othersxact))
1895 : : {
5515 1896 : 135 : SetPossibleUnsafeConflict(sxact, othersxact);
1897 : : }
1898 : : }
1899 : :
1900 : : /*
1901 : : * If we didn't find any possibly unsafe conflicts because every
1902 : : * uncommitted writable transaction turned out to be doomed, then we
1903 : : * can "opt out" immediately. See comments above the earlier check
1904 : : * for PredXact->WritableSxactCount == 0.
1905 : : */
1102 tmunro@postgresql.or 1906 [ - + ]: 108 : if (dlist_is_empty(&sxact->possibleUnsafeConflicts))
1907 : : {
1102 tmunro@postgresql.or 1908 :UBC 0 : ReleasePredXact(sxact);
1909 : 0 : LWLockRelease(SerializableXactHashLock);
1910 : 0 : return snapshot;
1911 : : }
1912 : : }
1913 : : else
1914 : : {
5515 heikki.linnakangas@i 1915 :CBC 1449 : ++(PredXact->WritableSxactCount);
1916 [ - + ]: 1449 : Assert(PredXact->WritableSxactCount <=
1917 : : (MaxBackends + max_prepared_xacts));
1918 : : }
1919 : :
1920 : : /* Maintain serializable global xmin info. */
1102 tmunro@postgresql.or 1921 [ + + ]: 1557 : if (!TransactionIdIsValid(PredXact->SxactGlobalXmin))
1922 : : {
1923 [ - + ]: 847 : Assert(PredXact->SxactGlobalXminCount == 0);
1924 : 847 : PredXact->SxactGlobalXmin = snapshot->xmin;
1925 : 847 : PredXact->SxactGlobalXminCount = 1;
1926 : 847 : SerialSetActiveSerXmin(snapshot->xmin);
1927 : : }
1928 [ + + ]: 710 : else if (TransactionIdEquals(snapshot->xmin, PredXact->SxactGlobalXmin))
1929 : : {
1930 [ - + ]: 672 : Assert(PredXact->SxactGlobalXminCount > 0);
1931 : 672 : PredXact->SxactGlobalXminCount++;
1932 : : }
1933 : : else
1934 : : {
1935 [ - + ]: 38 : Assert(TransactionIdFollows(snapshot->xmin, PredXact->SxactGlobalXmin));
1936 : : }
1937 : :
5515 heikki.linnakangas@i 1938 : 1557 : MySerializableXact = sxact;
5392 1939 : 1557 : MyXactDidWrite = false; /* haven't written anything yet */
1940 : :
5515 1941 : 1557 : LWLockRelease(SerializableXactHashLock);
1942 : :
2557 tmunro@postgresql.or 1943 : 1557 : CreateLocalPredicateLockHash();
1944 : :
1945 : 1557 : return snapshot;
1946 : : }
1947 : :
1948 : : static void
1949 : 1570 : CreateLocalPredicateLockHash(void)
1950 : : {
1951 : : HASHCTL hash_ctl;
1952 : :
1953 : : /* Initialize the backend-local hash table of parent locks */
5515 heikki.linnakangas@i 1954 [ - + ]: 1570 : Assert(LocalPredicateLockHash == NULL);
1955 : 1570 : hash_ctl.keysize = sizeof(PREDICATELOCKTARGETTAG);
1956 : 1570 : hash_ctl.entrysize = sizeof(LOCALPREDICATELOCK);
1957 : 1570 : LocalPredicateLockHash = hash_create("Local predicate lock",
1958 : : max_predicate_locks_per_xact,
1959 : : &hash_ctl,
1960 : : HASH_ELEM | HASH_BLOBS);
1961 : 1570 : }
1962 : :
1963 : : /*
1964 : : * Register the top level XID in SerializableXidHash.
1965 : : * Also store it for easy reference in MySerializableXact.
1966 : : */
1967 : : void
5380 1968 : 132936 : RegisterPredicateLockingXid(TransactionId xid)
1969 : : {
1970 : : SERIALIZABLEXIDTAG sxidtag;
1971 : : SERIALIZABLEXID *sxid;
1972 : : bool found;
1973 : :
1974 : : /*
1975 : : * If we're not tracking predicate lock data for this transaction, we
1976 : : * should ignore the request and return quickly.
1977 : : */
5515 1978 [ + + ]: 132936 : if (MySerializableXact == InvalidSerializableXact)
1979 : 131647 : return;
1980 : :
1981 : : /* We should have a valid XID and be at the top level. */
1982 [ - + ]: 1289 : Assert(TransactionIdIsValid(xid));
1983 : :
5392 1984 : 1289 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
1985 : :
1986 : : /* This should only be done once per transaction. */
1987 [ - + ]: 1289 : Assert(MySerializableXact->topXid == InvalidTransactionId);
1988 : :
5515 1989 : 1289 : MySerializableXact->topXid = xid;
1990 : :
1991 : 1289 : sxidtag.xid = xid;
1992 : 1289 : sxid = (SERIALIZABLEXID *) hash_search(SerializableXidHash,
1993 : : &sxidtag,
1994 : : HASH_ENTER, &found);
1995 [ - + ]: 1289 : Assert(!found);
1996 : :
1997 : : /* Initialize the structure. */
5392 1998 : 1289 : sxid->myXact = MySerializableXact;
5515 1999 : 1289 : LWLockRelease(SerializableXactHashLock);
2000 : : }
2001 : :
2002 : :
2003 : : /*
2004 : : * Check whether there are any predicate locks held by any transaction
2005 : : * for the page at the given block number.
2006 : : *
2007 : : * Note that the transaction may be completed but not yet subject to
2008 : : * cleanup due to overlapping serializable transactions. This must
2009 : : * return valid information regardless of transaction isolation level.
2010 : : *
2011 : : * Also note that this doesn't check for a conflicting relation lock,
2012 : : * just a lock specifically on the given page.
2013 : : *
2014 : : * One use is to support proper behavior during GiST index vacuum.
2015 : : */
2016 : : bool
5380 heikki.linnakangas@i 2017 :UBC 0 : PageIsPredicateLocked(Relation relation, BlockNumber blkno)
2018 : : {
2019 : : PREDICATELOCKTARGETTAG targettag;
2020 : : uint32 targettaghash;
2021 : : LWLock *partitionLock;
2022 : : PREDICATELOCKTARGET *target;
2023 : :
5515 2024 : 0 : SET_PREDICATELOCKTARGETTAG_PAGE(targettag,
2025 : : relation->rd_locator.dbOid,
2026 : : relation->rd_id,
2027 : : blkno);
2028 : :
2029 : 0 : targettaghash = PredicateLockTargetTagHashCode(&targettag);
2030 : 0 : partitionLock = PredicateLockHashPartitionLock(targettaghash);
2031 : 0 : LWLockAcquire(partitionLock, LW_SHARED);
2032 : : target = (PREDICATELOCKTARGET *)
2033 : 0 : hash_search_with_hash_value(PredicateLockTargetHash,
2034 : : &targettag, targettaghash,
2035 : : HASH_FIND, NULL);
2036 : 0 : LWLockRelease(partitionLock);
2037 : :
2038 : 0 : return (target != NULL);
2039 : : }
2040 : :
2041 : :
2042 : : /*
2043 : : * Check whether a particular lock is held by this transaction.
2044 : : *
2045 : : * Important note: this function may return false even if the lock is
2046 : : * being held, because it uses the local lock table which is not
2047 : : * updated if another transaction modifies our lock list (e.g. to
2048 : : * split an index page). It can also return true when a coarser
2049 : : * granularity lock that covers this target is being held. Be careful
2050 : : * to only use this function in circumstances where such errors are
2051 : : * acceptable!
2052 : : */
2053 : : static bool
5380 tgl@sss.pgh.pa.us 2054 :CBC 77299 : PredicateLockExists(const PREDICATELOCKTARGETTAG *targettag)
2055 : : {
2056 : : LOCALPREDICATELOCK *lock;
2057 : :
2058 : : /* check local hash table */
5515 heikki.linnakangas@i 2059 : 77299 : lock = (LOCALPREDICATELOCK *) hash_search(LocalPredicateLockHash,
2060 : : targettag,
2061 : : HASH_FIND, NULL);
2062 : :
2063 [ + + ]: 77299 : if (!lock)
2064 : 30196 : return false;
2065 : :
2066 : : /*
2067 : : * Found entry in the table, but still need to check whether it's actually
2068 : : * held -- it could just be a parent of some held lock.
2069 : : */
2070 : 47103 : return lock->held;
2071 : : }
2072 : :
2073 : : /*
2074 : : * Return the parent lock tag in the lock hierarchy: the next coarser
2075 : : * lock that covers the provided tag.
2076 : : *
2077 : : * Returns true and sets *parent to the parent tag if one exists,
2078 : : * returns false if none exists.
2079 : : */
2080 : : static bool
5380 tgl@sss.pgh.pa.us 2081 : 45355 : GetParentPredicateLockTag(const PREDICATELOCKTARGETTAG *tag,
2082 : : PREDICATELOCKTARGETTAG *parent)
2083 : : {
5515 heikki.linnakangas@i 2084 [ + + + + : 45355 : switch (GET_PREDICATELOCKTARGETTAG_TYPE(*tag))
+ - ]
2085 : : {
2086 : 9894 : case PREDLOCKTAG_RELATION:
2087 : : /* relation locks have no parent lock */
2088 : 9894 : return false;
2089 : :
2090 : 8488 : case PREDLOCKTAG_PAGE:
2091 : : /* parent lock is relation lock */
2092 : 8488 : SET_PREDICATELOCKTARGETTAG_RELATION(*parent,
2093 : : GET_PREDICATELOCKTARGETTAG_DB(*tag),
2094 : : GET_PREDICATELOCKTARGETTAG_RELATION(*tag));
2095 : :
2096 : 8488 : return true;
2097 : :
2098 : 26973 : case PREDLOCKTAG_TUPLE:
2099 : : /* parent lock is page lock */
2100 : 26973 : SET_PREDICATELOCKTARGETTAG_PAGE(*parent,
2101 : : GET_PREDICATELOCKTARGETTAG_DB(*tag),
2102 : : GET_PREDICATELOCKTARGETTAG_RELATION(*tag),
2103 : : GET_PREDICATELOCKTARGETTAG_PAGE(*tag));
2104 : 26973 : return true;
2105 : : }
2106 : :
2107 : : /* not reachable */
5515 heikki.linnakangas@i 2108 :UBC 0 : Assert(false);
2109 : : return false;
2110 : : }
2111 : :
2112 : : /*
2113 : : * Check whether the lock we are considering is already covered by a
2114 : : * coarser lock for our transaction.
2115 : : *
2116 : : * Like PredicateLockExists, this function might return a false
2117 : : * negative, but it will never return a false positive.
2118 : : */
2119 : : static bool
5380 tgl@sss.pgh.pa.us 2120 :CBC 26074 : CoarserLockCovers(const PREDICATELOCKTARGETTAG *newtargettag)
2121 : : {
2122 : : PREDICATELOCKTARGETTAG targettag,
2123 : : parenttag;
2124 : :
5515 heikki.linnakangas@i 2125 : 26074 : targettag = *newtargettag;
2126 : :
2127 : : /* check parents iteratively until no more */
2128 [ + + ]: 31506 : while (GetParentPredicateLockTag(&targettag, &parenttag))
2129 : : {
2130 : 27253 : targettag = parenttag;
2131 [ + + ]: 27253 : if (PredicateLockExists(&targettag))
2132 : 21821 : return true;
2133 : : }
2134 : :
2135 : : /* no more parents to check; lock is not covered */
2136 : 4253 : return false;
2137 : : }
2138 : :
2139 : : /*
2140 : : * Remove the dummy entry from the predicate lock target hash, to free up some
2141 : : * scratch space. The caller must be holding SerializablePredicateListLock,
2142 : : * and must restore the entry with RestoreScratchTarget() before releasing the
2143 : : * lock.
2144 : : *
2145 : : * If lockheld is true, the caller is already holding the partition lock
2146 : : * of the partition containing the scratch entry.
2147 : : */
2148 : : static void
5394 2149 : 49 : RemoveScratchTarget(bool lockheld)
2150 : : {
2151 : : bool found;
2152 : :
2130 tgl@sss.pgh.pa.us 2153 [ - + ]: 49 : Assert(LWLockHeldByMe(SerializablePredicateListLock));
2154 : :
5394 heikki.linnakangas@i 2155 [ - + ]: 49 : if (!lockheld)
5394 heikki.linnakangas@i 2156 :UBC 0 : LWLockAcquire(ScratchPartitionLock, LW_EXCLUSIVE);
5394 heikki.linnakangas@i 2157 :CBC 49 : hash_search_with_hash_value(PredicateLockTargetHash,
2158 : : &ScratchTargetTag,
2159 : : ScratchTargetTagHash,
2160 : : HASH_REMOVE, &found);
2161 [ - + ]: 49 : Assert(found);
2162 [ - + ]: 49 : if (!lockheld)
5394 heikki.linnakangas@i 2163 :UBC 0 : LWLockRelease(ScratchPartitionLock);
5394 heikki.linnakangas@i 2164 :CBC 49 : }
2165 : :
2166 : : /*
2167 : : * Re-insert the dummy entry in predicate lock target hash.
2168 : : */
2169 : : static void
2170 : 49 : RestoreScratchTarget(bool lockheld)
2171 : : {
2172 : : bool found;
2173 : :
2130 tgl@sss.pgh.pa.us 2174 [ - + ]: 49 : Assert(LWLockHeldByMe(SerializablePredicateListLock));
2175 : :
5394 heikki.linnakangas@i 2176 [ - + ]: 49 : if (!lockheld)
5394 heikki.linnakangas@i 2177 :UBC 0 : LWLockAcquire(ScratchPartitionLock, LW_EXCLUSIVE);
5394 heikki.linnakangas@i 2178 :CBC 49 : hash_search_with_hash_value(PredicateLockTargetHash,
2179 : : &ScratchTargetTag,
2180 : : ScratchTargetTagHash,
2181 : : HASH_ENTER, &found);
2182 [ - + ]: 49 : Assert(!found);
2183 [ - + ]: 49 : if (!lockheld)
5394 heikki.linnakangas@i 2184 :UBC 0 : LWLockRelease(ScratchPartitionLock);
5394 heikki.linnakangas@i 2185 :CBC 49 : }
2186 : :
2187 : : /*
2188 : : * Check whether the list of related predicate locks is empty for a
2189 : : * predicate lock target, and remove the target if it is.
2190 : : */
2191 : : static void
5515 2192 : 4247 : RemoveTargetIfNoLongerUsed(PREDICATELOCKTARGET *target, uint32 targettaghash)
2193 : : {
2194 : : PREDICATELOCKTARGET *rmtarget PG_USED_FOR_ASSERTS_ONLY;
2195 : :
2130 tgl@sss.pgh.pa.us 2196 [ - + ]: 4247 : Assert(LWLockHeldByMe(SerializablePredicateListLock));
2197 : :
2198 : : /* Can't remove it until no locks at this target. */
1151 andres@anarazel.de 2199 [ + + ]: 4247 : if (!dlist_is_empty(&target->predicateLocks))
5515 heikki.linnakangas@i 2200 : 973 : return;
2201 : :
2202 : : /* Actually remove the target. */
2203 : 3274 : rmtarget = hash_search_with_hash_value(PredicateLockTargetHash,
2204 : 3274 : &target->tag,
2205 : : targettaghash,
2206 : : HASH_REMOVE, NULL);
2207 [ - + ]: 3274 : Assert(rmtarget == target);
2208 : : }
2209 : :
2210 : : /*
2211 : : * Delete child target locks owned by this process.
2212 : : * This implementation is assuming that the usage of each target tag field
2213 : : * is uniform. No need to make this hard if we don't have to.
2214 : : *
2215 : : * We acquire an LWLock in the case of parallel mode, because worker
2216 : : * backends have access to the leader's SERIALIZABLEXACT. Otherwise,
2217 : : * we aren't acquiring LWLocks for the predicate lock or lock
2218 : : * target structures associated with this transaction unless we're going
2219 : : * to modify them, because no other process is permitted to modify our
2220 : : * locks.
2221 : : */
2222 : : static void
5380 tgl@sss.pgh.pa.us 2223 : 2371 : DeleteChildTargetLocks(const PREDICATELOCKTARGETTAG *newtargettag)
2224 : : {
2225 : : SERIALIZABLEXACT *sxact;
2226 : : PREDICATELOCK *predlock;
2227 : : dlist_mutable_iter iter;
2228 : :
2130 2229 : 2371 : LWLockAcquire(SerializablePredicateListLock, LW_SHARED);
5392 heikki.linnakangas@i 2230 : 2371 : sxact = MySerializableXact;
2557 tmunro@postgresql.or 2231 [ + + ]: 2371 : if (IsInParallelMode())
2130 tgl@sss.pgh.pa.us 2232 : 11 : LWLockAcquire(&sxact->perXactPredicateListLock, LW_EXCLUSIVE);
2233 : :
1151 andres@anarazel.de 2234 [ + - + + ]: 7831 : dlist_foreach_modify(iter, &sxact->predicateLocks)
2235 : : {
2236 : : PREDICATELOCKTAG oldlocktag;
2237 : : PREDICATELOCKTARGET *oldtarget;
2238 : : PREDICATELOCKTARGETTAG oldtargettag;
2239 : :
2240 : 5460 : predlock = dlist_container(PREDICATELOCK, xactLink, iter.cur);
2241 : :
5515 heikki.linnakangas@i 2242 : 5460 : oldlocktag = predlock->tag;
2243 [ - + ]: 5460 : Assert(oldlocktag.myXact == sxact);
2244 : 5460 : oldtarget = oldlocktag.myTarget;
2245 : 5460 : oldtargettag = oldtarget->tag;
2246 : :
2247 [ + + + - : 5460 : if (TargetTagIsCoveredBy(oldtargettag, *newtargettag))
+ + + + +
+ - + +
- ]
2248 : : {
2249 : : uint32 oldtargettaghash;
2250 : : LWLock *partitionLock;
2251 : : PREDICATELOCK *rmpredlock PG_USED_FOR_ASSERTS_ONLY;
2252 : :
2253 : 999 : oldtargettaghash = PredicateLockTargetTagHashCode(&oldtargettag);
2254 : 999 : partitionLock = PredicateLockHashPartitionLock(oldtargettaghash);
2255 : :
2256 : 999 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
2257 : :
1151 andres@anarazel.de 2258 : 999 : dlist_delete(&predlock->xactLink);
2259 : 999 : dlist_delete(&predlock->targetLink);
5515 heikki.linnakangas@i 2260 : 999 : rmpredlock = hash_search_with_hash_value
2261 : : (PredicateLockHash,
2262 : : &oldlocktag,
2263 : 999 : PredicateLockHashCodeFromTargetHashCode(&oldlocktag,
2264 : : oldtargettaghash),
2265 : : HASH_REMOVE, NULL);
2266 [ - + ]: 999 : Assert(rmpredlock == predlock);
2267 : :
2268 : 999 : RemoveTargetIfNoLongerUsed(oldtarget, oldtargettaghash);
2269 : :
2270 : 999 : LWLockRelease(partitionLock);
2271 : :
2272 : 999 : DecrementParentLocks(&oldtargettag);
2273 : : }
2274 : : }
2557 tmunro@postgresql.or 2275 [ + + ]: 2371 : if (IsInParallelMode())
2130 tgl@sss.pgh.pa.us 2276 : 11 : LWLockRelease(&sxact->perXactPredicateListLock);
2277 : 2371 : LWLockRelease(SerializablePredicateListLock);
5515 heikki.linnakangas@i 2278 : 2371 : }
2279 : :
2280 : : /*
2281 : : * Returns the promotion limit for a given predicate lock target. This is the
2282 : : * max number of descendant locks allowed before promoting to the specified
2283 : : * tag. Note that the limit includes non-direct descendants (e.g., both tuples
2284 : : * and pages for a relation lock).
2285 : : *
2286 : : * Currently the default limit is 2 for a page lock, and half of the value of
2287 : : * max_pred_locks_per_transaction - 1 for a relation lock, to match behavior
2288 : : * of earlier releases when upgrading.
2289 : : *
2290 : : * TODO SSI: We should probably add additional GUCs to allow a maximum ratio
2291 : : * of page and tuple locks based on the pages in a relation, and the maximum
2292 : : * ratio of tuple locks to tuples in a page. This would provide more
2293 : : * generally "balanced" allocation of locks to where they are most useful,
2294 : : * while still allowing the absolute numbers to prevent one relation from
2295 : : * tying up all predicate lock resources.
2296 : : */
2297 : : static int
3264 kgrittn@postgresql.o 2298 : 5432 : MaxPredicateChildLocks(const PREDICATELOCKTARGETTAG *tag)
2299 : : {
5515 heikki.linnakangas@i 2300 [ + - + + : 5432 : switch (GET_PREDICATELOCKTARGETTAG_TYPE(*tag))
- - ]
2301 : : {
2302 : 3550 : case PREDLOCKTAG_RELATION:
3264 kgrittn@postgresql.o 2303 : 3550 : return max_predicate_locks_per_relation < 0
2304 : : ? (max_predicate_locks_per_xact
2305 : 3550 : / (-max_predicate_locks_per_relation)) - 1
2306 [ + - ]: 3550 : : max_predicate_locks_per_relation;
2307 : :
5515 heikki.linnakangas@i 2308 : 1882 : case PREDLOCKTAG_PAGE:
3264 kgrittn@postgresql.o 2309 : 1882 : return max_predicate_locks_per_page;
2310 : :
5515 heikki.linnakangas@i 2311 :UBC 0 : case PREDLOCKTAG_TUPLE:
2312 : :
2313 : : /*
2314 : : * not reachable: nothing is finer-granularity than a tuple, so we
2315 : : * should never try to promote to it.
2316 : : */
2317 : 0 : Assert(false);
2318 : : return 0;
2319 : : }
2320 : :
2321 : : /* not reachable */
2322 : 0 : Assert(false);
2323 : : return 0;
2324 : : }
2325 : :
2326 : : /*
2327 : : * For all ancestors of a newly-acquired predicate lock, increment
2328 : : * their child count in the parent hash table. If any of them have
2329 : : * more descendants than their promotion threshold, acquire the
2330 : : * coarsest such lock.
2331 : : *
2332 : : * Returns true if a parent lock was acquired and false otherwise.
2333 : : */
2334 : : static bool
5380 tgl@sss.pgh.pa.us 2335 :CBC 4253 : CheckAndPromotePredicateLockRequest(const PREDICATELOCKTARGETTAG *reqtag)
2336 : : {
2337 : : PREDICATELOCKTARGETTAG targettag,
2338 : : nexttag,
2339 : : promotiontag;
2340 : : LOCALPREDICATELOCK *parentlock;
2341 : : bool found,
2342 : : promote;
2343 : :
5515 heikki.linnakangas@i 2344 : 4253 : promote = false;
2345 : :
2346 : 4253 : targettag = *reqtag;
2347 : :
2348 : : /* check parents iteratively */
2349 [ + + ]: 13938 : while (GetParentPredicateLockTag(&targettag, &nexttag))
2350 : : {
2351 : 5432 : targettag = nexttag;
2352 : 5432 : parentlock = (LOCALPREDICATELOCK *) hash_search(LocalPredicateLockHash,
2353 : : &targettag,
2354 : : HASH_ENTER,
2355 : : &found);
2356 [ + + ]: 5432 : if (!found)
2357 : : {
2358 : 3371 : parentlock->held = false;
2359 : 3371 : parentlock->childLocks = 1;
2360 : : }
2361 : : else
2362 : 2061 : parentlock->childLocks++;
2363 : :
3264 kgrittn@postgresql.o 2364 [ + + ]: 5432 : if (parentlock->childLocks >
2365 : 5432 : MaxPredicateChildLocks(&targettag))
2366 : : {
2367 : : /*
2368 : : * We should promote to this parent lock. Continue to check its
2369 : : * ancestors, however, both to get their child counts right and to
2370 : : * check whether we should just go ahead and promote to one of
2371 : : * them.
2372 : : */
5515 heikki.linnakangas@i 2373 : 333 : promotiontag = targettag;
2374 : 333 : promote = true;
2375 : : }
2376 : : }
2377 : :
2378 [ + + ]: 4253 : if (promote)
2379 : : {
2380 : : /* acquire coarsest ancestor eligible for promotion */
2381 : 333 : PredicateLockAcquire(&promotiontag);
2382 : 333 : return true;
2383 : : }
2384 : : else
2385 : 3920 : return false;
2386 : : }
2387 : :
2388 : : /*
2389 : : * When releasing a lock, decrement the child count on all ancestor
2390 : : * locks.
2391 : : *
2392 : : * This is called only when releasing a lock via
2393 : : * DeleteChildTargetLocks (i.e. when a lock becomes redundant because
2394 : : * we've acquired its parent, possibly due to promotion) or when a new
2395 : : * MVCC write lock makes the predicate lock unnecessary. There's no
2396 : : * point in calling it when locks are released at transaction end, as
2397 : : * this information is no longer needed.
2398 : : */
2399 : : static void
5380 tgl@sss.pgh.pa.us 2400 : 1388 : DecrementParentLocks(const PREDICATELOCKTARGETTAG *targettag)
2401 : : {
2402 : : PREDICATELOCKTARGETTAG parenttag,
2403 : : nexttag;
2404 : :
5515 heikki.linnakangas@i 2405 : 1388 : parenttag = *targettag;
2406 : :
2407 [ + + ]: 4164 : while (GetParentPredicateLockTag(&parenttag, &nexttag))
2408 : : {
2409 : : uint32 targettaghash;
2410 : : LOCALPREDICATELOCK *parentlock,
2411 : : *rmlock PG_USED_FOR_ASSERTS_ONLY;
2412 : :
2413 : 2776 : parenttag = nexttag;
2414 : 2776 : targettaghash = PredicateLockTargetTagHashCode(&parenttag);
2415 : : parentlock = (LOCALPREDICATELOCK *)
2416 : 2776 : hash_search_with_hash_value(LocalPredicateLockHash,
2417 : : &parenttag, targettaghash,
2418 : : HASH_FIND, NULL);
2419 : :
2420 : : /*
2421 : : * There's a small chance the parent lock doesn't exist in the lock
2422 : : * table. This can happen if we prematurely removed it because an
2423 : : * index split caused the child refcount to be off.
2424 : : */
2425 [ - + ]: 2776 : if (parentlock == NULL)
5515 heikki.linnakangas@i 2426 :UBC 0 : continue;
2427 : :
5515 heikki.linnakangas@i 2428 :CBC 2776 : parentlock->childLocks--;
2429 : :
2430 : : /*
2431 : : * Under similar circumstances the parent lock's refcount might be
2432 : : * zero. This only happens if we're holding that lock (otherwise we
2433 : : * would have removed the entry).
2434 : : */
2435 [ - + ]: 2776 : if (parentlock->childLocks < 0)
2436 : : {
5515 heikki.linnakangas@i 2437 [ # # ]:UBC 0 : Assert(parentlock->held);
2438 : 0 : parentlock->childLocks = 0;
2439 : : }
2440 : :
5515 heikki.linnakangas@i 2441 [ + + + + ]:CBC 2776 : if ((parentlock->childLocks == 0) && (!parentlock->held))
2442 : : {
2443 : : rmlock = (LOCALPREDICATELOCK *)
2444 : 766 : hash_search_with_hash_value(LocalPredicateLockHash,
2445 : : &parenttag, targettaghash,
2446 : : HASH_REMOVE, NULL);
2447 [ - + ]: 766 : Assert(rmlock == parentlock);
2448 : : }
2449 : : }
2450 : 1388 : }
2451 : :
2452 : : /*
2453 : : * Indicate that a predicate lock on the given target is held by the
2454 : : * specified transaction. Has no effect if the lock is already held.
2455 : : *
2456 : : * This updates the lock table and the sxact's lock list, and creates
2457 : : * the lock target if necessary, but does *not* do anything related to
2458 : : * granularity promotion or the local lock table. See
2459 : : * PredicateLockAcquire for that.
2460 : : */
2461 : : static void
5380 tgl@sss.pgh.pa.us 2462 : 4253 : CreatePredicateLock(const PREDICATELOCKTARGETTAG *targettag,
2463 : : uint32 targettaghash,
2464 : : SERIALIZABLEXACT *sxact)
2465 : : {
2466 : : PREDICATELOCKTARGET *target;
2467 : : PREDICATELOCKTAG locktag;
2468 : : PREDICATELOCK *lock;
2469 : : LWLock *partitionLock;
2470 : : bool found;
2471 : :
5515 heikki.linnakangas@i 2472 : 4253 : partitionLock = PredicateLockHashPartitionLock(targettaghash);
2473 : :
2130 tgl@sss.pgh.pa.us 2474 : 4253 : LWLockAcquire(SerializablePredicateListLock, LW_SHARED);
2557 tmunro@postgresql.or 2475 [ + + ]: 4253 : if (IsInParallelMode())
2130 tgl@sss.pgh.pa.us 2476 : 16 : LWLockAcquire(&sxact->perXactPredicateListLock, LW_EXCLUSIVE);
5515 heikki.linnakangas@i 2477 : 4253 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
2478 : :
2479 : : /* Make sure that the target is represented. */
2480 : : target = (PREDICATELOCKTARGET *)
2481 : 4253 : hash_search_with_hash_value(PredicateLockTargetHash,
2482 : : targettag, targettaghash,
2483 : : HASH_ENTER_NULL, &found);
2484 [ - + ]: 4253 : if (!target)
5515 heikki.linnakangas@i 2485 [ # # ]:UBC 0 : ereport(ERROR,
2486 : : (errcode(ERRCODE_OUT_OF_MEMORY),
2487 : : errmsg("out of shared memory"),
2488 : : errhint("You might need to increase \"%s\".", "max_pred_locks_per_transaction")));
5515 heikki.linnakangas@i 2489 [ + + ]:CBC 4253 : if (!found)
1151 andres@anarazel.de 2490 : 3274 : dlist_init(&target->predicateLocks);
2491 : :
2492 : : /* We've got the sxact and target, make sure they're joined. */
5515 heikki.linnakangas@i 2493 : 4253 : locktag.myTarget = target;
2494 : 4253 : locktag.myXact = sxact;
2495 : : lock = (PREDICATELOCK *)
2496 : 4253 : hash_search_with_hash_value(PredicateLockHash, &locktag,
3189 tgl@sss.pgh.pa.us 2497 : 4253 : PredicateLockHashCodeFromTargetHashCode(&locktag, targettaghash),
2498 : : HASH_ENTER_NULL, &found);
5515 heikki.linnakangas@i 2499 [ - + ]: 4253 : if (!lock)
5515 heikki.linnakangas@i 2500 [ # # ]:UBC 0 : ereport(ERROR,
2501 : : (errcode(ERRCODE_OUT_OF_MEMORY),
2502 : : errmsg("out of shared memory"),
2503 : : errhint("You might need to increase \"%s\".", "max_pred_locks_per_transaction")));
2504 : :
5515 heikki.linnakangas@i 2505 [ + + ]:CBC 4253 : if (!found)
2506 : : {
1151 andres@anarazel.de 2507 : 4247 : dlist_push_tail(&target->predicateLocks, &lock->targetLink);
2508 : 4247 : dlist_push_tail(&sxact->predicateLocks, &lock->xactLink);
5452 heikki.linnakangas@i 2509 : 4247 : lock->commitSeqNo = InvalidSerCommitSeqNo;
2510 : : }
2511 : :
5515 2512 : 4253 : LWLockRelease(partitionLock);
2557 tmunro@postgresql.or 2513 [ + + ]: 4253 : if (IsInParallelMode())
2130 tgl@sss.pgh.pa.us 2514 : 16 : LWLockRelease(&sxact->perXactPredicateListLock);
2515 : 4253 : LWLockRelease(SerializablePredicateListLock);
5515 heikki.linnakangas@i 2516 : 4253 : }
2517 : :
2518 : : /*
2519 : : * Acquire a predicate lock on the specified target for the current
2520 : : * connection if not already held. This updates the local lock table
2521 : : * and uses it to implement granularity promotion. It will consolidate
2522 : : * multiple locks into a coarser lock if warranted, and will release
2523 : : * any finer-grained locks covered by the new one.
2524 : : */
2525 : : static void
5380 tgl@sss.pgh.pa.us 2526 : 26276 : PredicateLockAcquire(const PREDICATELOCKTARGETTAG *targettag)
2527 : : {
2528 : : uint32 targettaghash;
2529 : : bool found;
2530 : : LOCALPREDICATELOCK *locallock;
2531 : :
2532 : : /* Do we have the lock already, or a covering lock? */
5515 heikki.linnakangas@i 2533 [ + + ]: 26276 : if (PredicateLockExists(targettag))
2534 : 22023 : return;
2535 : :
2536 [ + + ]: 26074 : if (CoarserLockCovers(targettag))
2537 : 21821 : return;
2538 : :
2539 : : /* the same hash and LW lock apply to the lock target and the local lock. */
2540 : 4253 : targettaghash = PredicateLockTargetTagHashCode(targettag);
2541 : :
2542 : : /* Acquire lock in local table */
2543 : : locallock = (LOCALPREDICATELOCK *)
2544 : 4253 : hash_search_with_hash_value(LocalPredicateLockHash,
2545 : : targettag, targettaghash,
2546 : : HASH_ENTER, &found);
2547 : 4253 : locallock->held = true;
2548 [ + + ]: 4253 : if (!found)
2549 : 3920 : locallock->childLocks = 0;
2550 : :
2551 : : /* Actually create the lock */
5392 2552 : 4253 : CreatePredicateLock(targettag, targettaghash, MySerializableXact);
2553 : :
2554 : : /*
2555 : : * Lock has been acquired. Check whether it should be promoted to a
2556 : : * coarser granularity, or whether there are finer-granularity locks to
2557 : : * clean up.
2558 : : */
5515 2559 [ + + ]: 4253 : if (CheckAndPromotePredicateLockRequest(targettag))
2560 : : {
2561 : : /*
2562 : : * Lock request was promoted to a coarser-granularity lock, and that
2563 : : * lock was acquired. It will delete this lock and any of its
2564 : : * children, so we're done.
2565 : : */
2566 : : }
2567 : : else
2568 : : {
2569 : : /* Clean up any finer-granularity locks */
2570 [ + + ]: 3920 : if (GET_PREDICATELOCKTARGETTAG_TYPE(*targettag) != PREDLOCKTAG_TUPLE)
2571 : 2371 : DeleteChildTargetLocks(targettag);
2572 : : }
2573 : : }
2574 : :
2575 : :
2576 : : /*
2577 : : * PredicateLockRelation
2578 : : *
2579 : : * Gets a predicate lock at the relation level.
2580 : : * Skip if not in full serializable transaction isolation level.
2581 : : * Skip if this is a temporary table.
2582 : : * Clear any finer-grained predicate locks this session has on the relation.
2583 : : */
2584 : : void
5380 2585 : 682743 : PredicateLockRelation(Relation relation, Snapshot snapshot)
2586 : : {
2587 : : PREDICATELOCKTARGETTAG tag;
2588 : :
5387 2589 [ + + ]: 682743 : if (!SerializationNeededForRead(relation, snapshot))
5515 2590 : 682022 : return;
2591 : :
2592 : 721 : SET_PREDICATELOCKTARGETTAG_RELATION(tag,
2593 : : relation->rd_locator.dbOid,
2594 : : relation->rd_id);
2595 : 721 : PredicateLockAcquire(&tag);
2596 : : }
2597 : :
2598 : : /*
2599 : : * PredicateLockPage
2600 : : *
2601 : : * Gets a predicate lock at the page level.
2602 : : * Skip if not in full serializable transaction isolation level.
2603 : : * Skip if this is a temporary table.
2604 : : * Skip if a coarser predicate lock already covers this page.
2605 : : * Clear any finer-grained predicate locks this session has on the relation.
2606 : : */
2607 : : void
5380 2608 : 9573820 : PredicateLockPage(Relation relation, BlockNumber blkno, Snapshot snapshot)
2609 : : {
2610 : : PREDICATELOCKTARGETTAG tag;
2611 : :
5387 2612 [ + + ]: 9573820 : if (!SerializationNeededForRead(relation, snapshot))
5515 2613 : 9572368 : return;
2614 : :
2615 : 1452 : SET_PREDICATELOCKTARGETTAG_PAGE(tag,
2616 : : relation->rd_locator.dbOid,
2617 : : relation->rd_id,
2618 : : blkno);
2619 : 1452 : PredicateLockAcquire(&tag);
2620 : : }
2621 : :
2622 : : /*
2623 : : * PredicateLockTID
2624 : : *
2625 : : * Gets a predicate lock at the tuple level.
2626 : : * Skip if not in full serializable transaction isolation level.
2627 : : * Skip if this is a temporary table.
2628 : : */
2629 : : void
136 peter@eisentraut.org 2630 :GNC 16871186 : PredicateLockTID(Relation relation, const ItemPointerData *tid, Snapshot snapshot,
2631 : : TransactionId tuple_xid)
2632 : : {
2633 : : PREDICATELOCKTARGETTAG tag;
2634 : :
5387 heikki.linnakangas@i 2635 [ + + ]:CBC 16871186 : if (!SerializationNeededForRead(relation, snapshot))
5515 2636 : 16847416 : return;
2637 : :
2638 : : /*
2639 : : * Return if this xact wrote it.
2640 : : */
2641 [ + - ]: 23772 : if (relation->rd_index == NULL)
2642 : : {
2643 : : /* If we wrote it; we already have a write lock. */
2238 tmunro@postgresql.or 2644 [ + + ]: 23772 : if (TransactionIdIsCurrentTransactionId(tuple_xid))
2316 tmunro@postgresql.or 2645 :GBC 2 : return;
2646 : : }
2647 : :
2648 : : /*
2649 : : * Do quick-but-not-definitive test for a relation lock first. This will
2650 : : * never cause a return when the relation is *not* locked, but will
2651 : : * occasionally let the check continue when there really *is* a relation
2652 : : * level lock.
2653 : : */
5515 heikki.linnakangas@i 2654 :CBC 23770 : SET_PREDICATELOCKTARGETTAG_RELATION(tag,
2655 : : relation->rd_locator.dbOid,
2656 : : relation->rd_id);
2657 [ - + ]: 23770 : if (PredicateLockExists(&tag))
5515 heikki.linnakangas@i 2658 :UBC 0 : return;
2659 : :
5515 heikki.linnakangas@i 2660 :CBC 23770 : SET_PREDICATELOCKTARGETTAG_TUPLE(tag,
2661 : : relation->rd_locator.dbOid,
2662 : : relation->rd_id,
2663 : : ItemPointerGetBlockNumber(tid),
2664 : : ItemPointerGetOffsetNumber(tid));
2665 : 23770 : PredicateLockAcquire(&tag);
2666 : : }
2667 : :
2668 : :
2669 : : /*
2670 : : * DeleteLockTarget
2671 : : *
2672 : : * Remove a predicate lock target along with any locks held for it.
2673 : : *
2674 : : * Caller must hold SerializablePredicateListLock and the
2675 : : * appropriate hash partition lock for the target.
2676 : : */
2677 : : static void
5515 heikki.linnakangas@i 2678 :UBC 0 : DeleteLockTarget(PREDICATELOCKTARGET *target, uint32 targettaghash)
2679 : : {
2680 : : dlist_mutable_iter iter;
2681 : :
2130 tgl@sss.pgh.pa.us 2682 [ # # ]: 0 : Assert(LWLockHeldByMeInMode(SerializablePredicateListLock,
2683 : : LW_EXCLUSIVE));
5515 heikki.linnakangas@i 2684 [ # # ]: 0 : Assert(LWLockHeldByMe(PredicateLockHashPartitionLock(targettaghash)));
2685 : :
2686 : 0 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
2687 : :
1151 andres@anarazel.de 2688 [ # # # # ]: 0 : dlist_foreach_modify(iter, &target->predicateLocks)
2689 : : {
2690 : 0 : PREDICATELOCK *predlock =
1031 tgl@sss.pgh.pa.us 2691 : 0 : dlist_container(PREDICATELOCK, targetLink, iter.cur);
2692 : : bool found;
2693 : :
1151 andres@anarazel.de 2694 : 0 : dlist_delete(&(predlock->xactLink));
2695 : 0 : dlist_delete(&(predlock->targetLink));
2696 : :
5515 heikki.linnakangas@i 2697 : 0 : hash_search_with_hash_value
2698 : : (PredicateLockHash,
2699 : 0 : &predlock->tag,
2700 : 0 : PredicateLockHashCodeFromTargetHashCode(&predlock->tag,
2701 : : targettaghash),
2702 : : HASH_REMOVE, &found);
2703 [ # # ]: 0 : Assert(found);
2704 : : }
2705 : 0 : LWLockRelease(SerializableXactHashLock);
2706 : :
2707 : : /* Remove the target itself, if possible. */
2708 : 0 : RemoveTargetIfNoLongerUsed(target, targettaghash);
2709 : 0 : }
2710 : :
2711 : :
2712 : : /*
2713 : : * TransferPredicateLocksToNewTarget
2714 : : *
2715 : : * Move or copy all the predicate locks for a lock target, for use by
2716 : : * index page splits/combines and other things that create or replace
2717 : : * lock targets. If 'removeOld' is true, the old locks and the target
2718 : : * will be removed.
2719 : : *
2720 : : * Returns true on success, or false if we ran out of shared memory to
2721 : : * allocate the new target or locks. Guaranteed to always succeed if
2722 : : * removeOld is set (by using the scratch entry in PredicateLockTargetHash
2723 : : * for scratch space).
2724 : : *
2725 : : * Warning: the "removeOld" option should be used only with care,
2726 : : * because this function does not (indeed, can not) update other
2727 : : * backends' LocalPredicateLockHash. If we are only adding new
2728 : : * entries, this is not a problem: the local lock table is used only
2729 : : * as a hint, so missing entries for locks that are held are
2730 : : * OK. Having entries for locks that are no longer held, as can happen
2731 : : * when using "removeOld", is not in general OK. We can only use it
2732 : : * safely when replacing a lock with a coarser-granularity lock that
2733 : : * covers it, or if we are absolutely certain that no one will need to
2734 : : * refer to that lock in the future.
2735 : : *
2736 : : * Caller must hold SerializablePredicateListLock exclusively.
2737 : : */
2738 : : static bool
5380 2739 : 0 : TransferPredicateLocksToNewTarget(PREDICATELOCKTARGETTAG oldtargettag,
2740 : : PREDICATELOCKTARGETTAG newtargettag,
2741 : : bool removeOld)
2742 : : {
2743 : : uint32 oldtargettaghash;
2744 : : LWLock *oldpartitionLock;
2745 : : PREDICATELOCKTARGET *oldtarget;
2746 : : uint32 newtargettaghash;
2747 : : LWLock *newpartitionLock;
2748 : : bool found;
5515 2749 : 0 : bool outOfShmem = false;
2750 : :
2130 tgl@sss.pgh.pa.us 2751 [ # # ]: 0 : Assert(LWLockHeldByMeInMode(SerializablePredicateListLock,
2752 : : LW_EXCLUSIVE));
2753 : :
5515 heikki.linnakangas@i 2754 : 0 : oldtargettaghash = PredicateLockTargetTagHashCode(&oldtargettag);
2755 : 0 : newtargettaghash = PredicateLockTargetTagHashCode(&newtargettag);
2756 : 0 : oldpartitionLock = PredicateLockHashPartitionLock(oldtargettaghash);
2757 : 0 : newpartitionLock = PredicateLockHashPartitionLock(newtargettaghash);
2758 : :
2759 [ # # ]: 0 : if (removeOld)
2760 : : {
2761 : : /*
2762 : : * Remove the dummy entry to give us scratch space, so we know we'll
2763 : : * be able to create the new lock target.
2764 : : */
5394 2765 : 0 : RemoveScratchTarget(false);
2766 : : }
2767 : :
2768 : : /*
2769 : : * We must get the partition locks in ascending sequence to avoid
2770 : : * deadlocks. If old and new partitions are the same, we must request the
2771 : : * lock only once.
2772 : : */
5515 2773 [ # # ]: 0 : if (oldpartitionLock < newpartitionLock)
2774 : : {
2775 : 0 : LWLockAcquire(oldpartitionLock,
2776 : 0 : (removeOld ? LW_EXCLUSIVE : LW_SHARED));
2777 : 0 : LWLockAcquire(newpartitionLock, LW_EXCLUSIVE);
2778 : : }
2779 [ # # ]: 0 : else if (oldpartitionLock > newpartitionLock)
2780 : : {
2781 : 0 : LWLockAcquire(newpartitionLock, LW_EXCLUSIVE);
2782 : 0 : LWLockAcquire(oldpartitionLock,
2783 : 0 : (removeOld ? LW_EXCLUSIVE : LW_SHARED));
2784 : : }
2785 : : else
2786 : 0 : LWLockAcquire(newpartitionLock, LW_EXCLUSIVE);
2787 : :
2788 : : /*
2789 : : * Look for the old target. If not found, that's OK; no predicate locks
2790 : : * are affected, so we can just clean up and return. If it does exist,
2791 : : * walk its list of predicate locks and move or copy them to the new
2792 : : * target.
2793 : : */
2794 : 0 : oldtarget = hash_search_with_hash_value(PredicateLockTargetHash,
2795 : : &oldtargettag,
2796 : : oldtargettaghash,
2797 : : HASH_FIND, NULL);
2798 : :
2799 [ # # ]: 0 : if (oldtarget)
2800 : : {
2801 : : PREDICATELOCKTARGET *newtarget;
2802 : : PREDICATELOCKTAG newpredlocktag;
2803 : : dlist_mutable_iter iter;
2804 : :
2805 : 0 : newtarget = hash_search_with_hash_value(PredicateLockTargetHash,
2806 : : &newtargettag,
2807 : : newtargettaghash,
2808 : : HASH_ENTER_NULL, &found);
2809 : :
2810 [ # # ]: 0 : if (!newtarget)
2811 : : {
2812 : : /* Failed to allocate due to insufficient shmem */
2813 : 0 : outOfShmem = true;
2814 : 0 : goto exit;
2815 : : }
2816 : :
2817 : : /* If we created a new entry, initialize it */
2818 [ # # ]: 0 : if (!found)
1151 andres@anarazel.de 2819 : 0 : dlist_init(&newtarget->predicateLocks);
2820 : :
5504 magnus@hagander.net 2821 : 0 : newpredlocktag.myTarget = newtarget;
2822 : :
2823 : : /*
2824 : : * Loop through all the locks on the old target, replacing them with
2825 : : * locks on the new target.
2826 : : */
5515 heikki.linnakangas@i 2827 : 0 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
2828 : :
1151 andres@anarazel.de 2829 [ # # # # ]: 0 : dlist_foreach_modify(iter, &oldtarget->predicateLocks)
2830 : : {
2831 : 0 : PREDICATELOCK *oldpredlock =
1031 tgl@sss.pgh.pa.us 2832 : 0 : dlist_container(PREDICATELOCK, targetLink, iter.cur);
2833 : : PREDICATELOCK *newpredlock;
5452 heikki.linnakangas@i 2834 : 0 : SerCommitSeqNo oldCommitSeqNo = oldpredlock->commitSeqNo;
2835 : :
5515 2836 : 0 : newpredlocktag.myXact = oldpredlock->tag.myXact;
2837 : :
2838 [ # # ]: 0 : if (removeOld)
2839 : : {
1151 andres@anarazel.de 2840 : 0 : dlist_delete(&(oldpredlock->xactLink));
2841 : 0 : dlist_delete(&(oldpredlock->targetLink));
2842 : :
5515 heikki.linnakangas@i 2843 : 0 : hash_search_with_hash_value
2844 : : (PredicateLockHash,
2845 : 0 : &oldpredlock->tag,
3189 tgl@sss.pgh.pa.us 2846 : 0 : PredicateLockHashCodeFromTargetHashCode(&oldpredlock->tag,
2847 : : oldtargettaghash),
2848 : : HASH_REMOVE, &found);
5515 heikki.linnakangas@i 2849 [ # # ]: 0 : Assert(found);
2850 : : }
2851 : :
2852 : : newpredlock = (PREDICATELOCK *)
5380 tgl@sss.pgh.pa.us 2853 : 0 : hash_search_with_hash_value(PredicateLockHash,
2854 : : &newpredlocktag,
3189 2855 : 0 : PredicateLockHashCodeFromTargetHashCode(&newpredlocktag,
2856 : : newtargettaghash),
2857 : : HASH_ENTER_NULL,
2858 : : &found);
5515 heikki.linnakangas@i 2859 [ # # ]: 0 : if (!newpredlock)
2860 : : {
2861 : : /* Out of shared memory. Undo what we've done so far. */
2862 : 0 : LWLockRelease(SerializableXactHashLock);
2863 : 0 : DeleteLockTarget(newtarget, newtargettaghash);
2864 : 0 : outOfShmem = true;
2865 : 0 : goto exit;
2866 : : }
5504 magnus@hagander.net 2867 [ # # ]: 0 : if (!found)
2868 : : {
1151 andres@anarazel.de 2869 : 0 : dlist_push_tail(&(newtarget->predicateLocks),
2870 : : &(newpredlock->targetLink));
2871 : 0 : dlist_push_tail(&(newpredlocktag.myXact->predicateLocks),
2872 : : &(newpredlock->xactLink));
5452 heikki.linnakangas@i 2873 : 0 : newpredlock->commitSeqNo = oldCommitSeqNo;
2874 : : }
2875 : : else
2876 : : {
2877 [ # # ]: 0 : if (newpredlock->commitSeqNo < oldCommitSeqNo)
2878 : 0 : newpredlock->commitSeqNo = oldCommitSeqNo;
2879 : : }
2880 : :
2881 [ # # ]: 0 : Assert(newpredlock->commitSeqNo != 0);
2882 [ # # # # ]: 0 : Assert((newpredlock->commitSeqNo == InvalidSerCommitSeqNo)
2883 : : || (newpredlock->tag.myXact == OldCommittedSxact));
2884 : : }
5515 2885 : 0 : LWLockRelease(SerializableXactHashLock);
2886 : :
2887 [ # # ]: 0 : if (removeOld)
2888 : : {
1151 andres@anarazel.de 2889 [ # # ]: 0 : Assert(dlist_is_empty(&oldtarget->predicateLocks));
5515 heikki.linnakangas@i 2890 : 0 : RemoveTargetIfNoLongerUsed(oldtarget, oldtargettaghash);
2891 : : }
2892 : : }
2893 : :
2894 : :
2895 : 0 : exit:
2896 : : /* Release partition locks in reverse order of acquisition. */
2897 [ # # ]: 0 : if (oldpartitionLock < newpartitionLock)
2898 : : {
2899 : 0 : LWLockRelease(newpartitionLock);
2900 : 0 : LWLockRelease(oldpartitionLock);
2901 : : }
2902 [ # # ]: 0 : else if (oldpartitionLock > newpartitionLock)
2903 : : {
2904 : 0 : LWLockRelease(oldpartitionLock);
2905 : 0 : LWLockRelease(newpartitionLock);
2906 : : }
2907 : : else
2908 : 0 : LWLockRelease(newpartitionLock);
2909 : :
2910 [ # # ]: 0 : if (removeOld)
2911 : : {
2912 : : /* We shouldn't run out of memory if we're moving locks */
2913 [ # # ]: 0 : Assert(!outOfShmem);
2914 : :
2915 : : /* Put the scratch entry back */
5394 2916 : 0 : RestoreScratchTarget(false);
2917 : : }
2918 : :
5515 2919 : 0 : return !outOfShmem;
2920 : : }
2921 : :
2922 : : /*
2923 : : * Drop all predicate locks of any granularity from the specified relation,
2924 : : * which can be a heap relation or an index relation. If 'transfer' is true,
2925 : : * acquire a relation lock on the heap for any transactions with any lock(s)
2926 : : * on the specified relation.
2927 : : *
2928 : : * This requires grabbing a lot of LW locks and scanning the entire lock
2929 : : * target table for matches. That makes this more expensive than most
2930 : : * predicate lock management functions, but it will only be called for DDL
2931 : : * type commands that are expensive anyway, and there are fast returns when
2932 : : * no serializable transactions are active or the relation is temporary.
2933 : : *
2934 : : * We don't use the TransferPredicateLocksToNewTarget function because it
2935 : : * acquires its own locks on the partitions of the two targets involved,
2936 : : * and we'll already be holding all partition locks.
2937 : : *
2938 : : * We can't throw an error from here, because the call could be from a
2939 : : * transaction which is not serializable.
2940 : : *
2941 : : * NOTE: This is currently only called with transfer set to true, but that may
2942 : : * change. If we decide to clean up the locks from a table on commit of a
2943 : : * transaction which executed DROP TABLE, the false condition will be useful.
2944 : : */
2945 : : static void
5380 heikki.linnakangas@i 2946 :CBC 17889 : DropAllPredicateLocksFromTable(Relation relation, bool transfer)
2947 : : {
2948 : : HASH_SEQ_STATUS seqstat;
2949 : : PREDICATELOCKTARGET *oldtarget;
2950 : : PREDICATELOCKTARGET *heaptarget;
2951 : : Oid dbId;
2952 : : Oid relId;
2953 : : Oid heapId;
2954 : : int i;
2955 : : bool isIndex;
2956 : : bool found;
2957 : : uint32 heaptargettaghash;
2958 : :
2959 : : /*
2960 : : * Bail out quickly if there are no serializable transactions running.
2961 : : * It's safe to check this without taking locks because the caller is
2962 : : * holding an ACCESS EXCLUSIVE lock on the relation. No new locks which
2963 : : * would matter here can be acquired while that is held.
2964 : : */
5394 2965 [ + + ]: 17889 : if (!TransactionIdIsValid(PredXact->SxactGlobalXmin))
2966 : 17840 : return;
2967 : :
5387 2968 [ + + ]: 66 : if (!PredicateLockingNeededForRelation(relation))
5394 2969 : 17 : return;
2970 : :
1348 rhaas@postgresql.org 2971 : 49 : dbId = relation->rd_locator.dbOid;
5394 heikki.linnakangas@i 2972 : 49 : relId = relation->rd_id;
2973 [ + + ]: 49 : if (relation->rd_index == NULL)
2974 : : {
2975 : 4 : isIndex = false;
2976 : 4 : heapId = relId;
2977 : : }
2978 : : else
2979 : : {
2980 : 45 : isIndex = true;
2981 : 45 : heapId = relation->rd_index->indrelid;
2982 : : }
2983 [ - + ]: 49 : Assert(heapId != InvalidOid);
3189 tgl@sss.pgh.pa.us 2984 [ - + - - ]: 49 : Assert(transfer || !isIndex); /* index OID only makes sense with
2985 : : * transfer */
2986 : :
2987 : : /* Retrieve first time needed, then keep. */
5394 heikki.linnakangas@i 2988 : 49 : heaptargettaghash = 0;
2989 : 49 : heaptarget = NULL;
2990 : :
2991 : : /* Acquire locks on all lock partitions */
2130 tgl@sss.pgh.pa.us 2992 : 49 : LWLockAcquire(SerializablePredicateListLock, LW_EXCLUSIVE);
5394 heikki.linnakangas@i 2993 [ + + ]: 833 : for (i = 0; i < NUM_PREDICATELOCK_PARTITIONS; i++)
4430 rhaas@postgresql.org 2994 : 784 : LWLockAcquire(PredicateLockHashPartitionLockByIndex(i), LW_EXCLUSIVE);
5394 heikki.linnakangas@i 2995 : 49 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
2996 : :
2997 : : /*
2998 : : * Remove the dummy entry to give us scratch space, so we know we'll be
2999 : : * able to create the new lock target.
3000 : : */
3001 [ + - ]: 49 : if (transfer)
3002 : 49 : RemoveScratchTarget(true);
3003 : :
3004 : : /* Scan through target map */
3005 : 49 : hash_seq_init(&seqstat, PredicateLockTargetHash);
3006 : :
3007 [ + + ]: 93 : while ((oldtarget = (PREDICATELOCKTARGET *) hash_seq_search(&seqstat)))
3008 : : {
3009 : : dlist_mutable_iter iter;
3010 : :
3011 : : /*
3012 : : * Check whether this is a target which needs attention.
3013 : : */
3014 [ + - ]: 44 : if (GET_PREDICATELOCKTARGETTAG_RELATION(oldtarget->tag) != relId)
3015 : 44 : continue; /* wrong relation id */
5394 heikki.linnakangas@i 3016 [ # # ]:UBC 0 : if (GET_PREDICATELOCKTARGETTAG_DB(oldtarget->tag) != dbId)
3017 : 0 : continue; /* wrong database id */
3018 [ # # # # ]: 0 : if (transfer && !isIndex
3019 [ # # # # ]: 0 : && GET_PREDICATELOCKTARGETTAG_TYPE(oldtarget->tag) == PREDLOCKTAG_RELATION)
3020 : 0 : continue; /* already the right lock */
3021 : :
3022 : : /*
3023 : : * If we made it here, we have work to do. We make sure the heap
3024 : : * relation lock exists, then we walk the list of predicate locks for
3025 : : * the old target we found, moving all locks to the heap relation lock
3026 : : * -- unless they already hold that.
3027 : : */
3028 : :
3029 : : /*
3030 : : * First make sure we have the heap relation target. We only need to
3031 : : * do this once.
3032 : : */
3033 [ # # # # ]: 0 : if (transfer && heaptarget == NULL)
3034 : : {
3035 : : PREDICATELOCKTARGETTAG heaptargettag;
3036 : :
3037 : 0 : SET_PREDICATELOCKTARGETTAG_RELATION(heaptargettag, dbId, heapId);
3038 : 0 : heaptargettaghash = PredicateLockTargetTagHashCode(&heaptargettag);
3039 : 0 : heaptarget = hash_search_with_hash_value(PredicateLockTargetHash,
3040 : : &heaptargettag,
3041 : : heaptargettaghash,
3042 : : HASH_ENTER, &found);
3043 [ # # ]: 0 : if (!found)
1151 andres@anarazel.de 3044 : 0 : dlist_init(&heaptarget->predicateLocks);
3045 : : }
3046 : :
3047 : : /*
3048 : : * Loop through all the locks on the old target, replacing them with
3049 : : * locks on the new target.
3050 : : */
3051 [ # # # # ]: 0 : dlist_foreach_modify(iter, &oldtarget->predicateLocks)
3052 : : {
3053 : 0 : PREDICATELOCK *oldpredlock =
1031 tgl@sss.pgh.pa.us 3054 : 0 : dlist_container(PREDICATELOCK, targetLink, iter.cur);
3055 : : PREDICATELOCK *newpredlock;
3056 : : SerCommitSeqNo oldCommitSeqNo;
3057 : : SERIALIZABLEXACT *oldXact;
3058 : :
3059 : : /*
3060 : : * Remove the old lock first. This avoids the chance of running
3061 : : * out of lock structure entries for the hash table.
3062 : : */
5394 heikki.linnakangas@i 3063 : 0 : oldCommitSeqNo = oldpredlock->commitSeqNo;
3064 : 0 : oldXact = oldpredlock->tag.myXact;
3065 : :
1151 andres@anarazel.de 3066 : 0 : dlist_delete(&(oldpredlock->xactLink));
3067 : :
3068 : : /*
3069 : : * No need for retail delete from oldtarget list, we're removing
3070 : : * the whole target anyway.
3071 : : */
5394 heikki.linnakangas@i 3072 : 0 : hash_search(PredicateLockHash,
3073 : 0 : &oldpredlock->tag,
3074 : : HASH_REMOVE, &found);
3075 [ # # ]: 0 : Assert(found);
3076 : :
3077 [ # # ]: 0 : if (transfer)
3078 : : {
3079 : : PREDICATELOCKTAG newpredlocktag;
3080 : :
3081 : 0 : newpredlocktag.myTarget = heaptarget;
3082 : 0 : newpredlocktag.myXact = oldXact;
3083 : : newpredlock = (PREDICATELOCK *)
5380 tgl@sss.pgh.pa.us 3084 : 0 : hash_search_with_hash_value(PredicateLockHash,
3085 : : &newpredlocktag,
3189 3086 : 0 : PredicateLockHashCodeFromTargetHashCode(&newpredlocktag,
3087 : : heaptargettaghash),
3088 : : HASH_ENTER,
3089 : : &found);
5394 heikki.linnakangas@i 3090 [ # # ]: 0 : if (!found)
3091 : : {
1151 andres@anarazel.de 3092 : 0 : dlist_push_tail(&(heaptarget->predicateLocks),
3093 : : &(newpredlock->targetLink));
3094 : 0 : dlist_push_tail(&(newpredlocktag.myXact->predicateLocks),
3095 : : &(newpredlock->xactLink));
5394 heikki.linnakangas@i 3096 : 0 : newpredlock->commitSeqNo = oldCommitSeqNo;
3097 : : }
3098 : : else
3099 : : {
3100 [ # # ]: 0 : if (newpredlock->commitSeqNo < oldCommitSeqNo)
3101 : 0 : newpredlock->commitSeqNo = oldCommitSeqNo;
3102 : : }
3103 : :
3104 [ # # ]: 0 : Assert(newpredlock->commitSeqNo != 0);
3105 [ # # # # ]: 0 : Assert((newpredlock->commitSeqNo == InvalidSerCommitSeqNo)
3106 : : || (newpredlock->tag.myXact == OldCommittedSxact));
3107 : : }
3108 : : }
3109 : :
3110 : 0 : hash_search(PredicateLockTargetHash, &oldtarget->tag, HASH_REMOVE,
3111 : : &found);
3112 [ # # ]: 0 : Assert(found);
3113 : : }
3114 : :
3115 : : /* Put the scratch entry back */
5394 heikki.linnakangas@i 3116 [ + - ]:CBC 49 : if (transfer)
3117 : 49 : RestoreScratchTarget(true);
3118 : :
3119 : : /* Release locks in reverse order */
3120 : 49 : LWLockRelease(SerializableXactHashLock);
3121 [ + + ]: 833 : for (i = NUM_PREDICATELOCK_PARTITIONS - 1; i >= 0; i--)
4430 rhaas@postgresql.org 3122 : 784 : LWLockRelease(PredicateLockHashPartitionLockByIndex(i));
2130 tgl@sss.pgh.pa.us 3123 : 49 : LWLockRelease(SerializablePredicateListLock);
3124 : : }
3125 : :
3126 : : /*
3127 : : * TransferPredicateLocksToHeapRelation
3128 : : * For all transactions, transfer all predicate locks for the given
3129 : : * relation to a single relation lock on the heap.
3130 : : */
3131 : : void
5380 heikki.linnakangas@i 3132 : 17889 : TransferPredicateLocksToHeapRelation(Relation relation)
3133 : : {
5394 3134 : 17889 : DropAllPredicateLocksFromTable(relation, true);
3135 : 17889 : }
3136 : :
3137 : :
3138 : : /*
3139 : : * PredicateLockPageSplit
3140 : : *
3141 : : * Copies any predicate locks for the old page to the new page.
3142 : : * Skip if this is a temporary table or toast table.
3143 : : *
3144 : : * NOTE: A page split (or overflow) affects all serializable transactions,
3145 : : * even if it occurs in the context of another transaction isolation level.
3146 : : *
3147 : : * NOTE: This currently leaves the local copy of the locks without
3148 : : * information on the new lock which is in shared memory. This could cause
3149 : : * problems if enough page splits occur on locked pages without the processes
3150 : : * which hold the locks getting in and noticing.
3151 : : */
3152 : : void
5380 3153 : 32506 : PredicateLockPageSplit(Relation relation, BlockNumber oldblkno,
3154 : : BlockNumber newblkno)
3155 : : {
3156 : : PREDICATELOCKTARGETTAG oldtargettag;
3157 : : PREDICATELOCKTARGETTAG newtargettag;
3158 : : bool success;
3159 : :
3160 : : /*
3161 : : * Bail out quickly if there are no serializable transactions running.
3162 : : *
3163 : : * It's safe to do this check without taking any additional locks. Even if
3164 : : * a serializable transaction starts concurrently, we know it can't take
3165 : : * any SIREAD locks on the page being split because the caller is holding
3166 : : * the associated buffer page lock. Memory reordering isn't an issue; the
3167 : : * memory barrier in the LWLock acquisition guarantees that this read
3168 : : * occurs while the buffer page lock is held.
3169 : : */
5438 rhaas@postgresql.org 3170 [ + + ]: 32506 : if (!TransactionIdIsValid(PredXact->SxactGlobalXmin))
3171 : 32506 : return;
3172 : :
5387 heikki.linnakangas@i 3173 [ + - ]: 19 : if (!PredicateLockingNeededForRelation(relation))
5515 3174 : 19 : return;
3175 : :
5515 heikki.linnakangas@i 3176 [ # # ]:UBC 0 : Assert(oldblkno != newblkno);
3177 [ # # ]: 0 : Assert(BlockNumberIsValid(oldblkno));
3178 [ # # ]: 0 : Assert(BlockNumberIsValid(newblkno));
3179 : :
3180 : 0 : SET_PREDICATELOCKTARGETTAG_PAGE(oldtargettag,
3181 : : relation->rd_locator.dbOid,
3182 : : relation->rd_id,
3183 : : oldblkno);
3184 : 0 : SET_PREDICATELOCKTARGETTAG_PAGE(newtargettag,
3185 : : relation->rd_locator.dbOid,
3186 : : relation->rd_id,
3187 : : newblkno);
3188 : :
2130 tgl@sss.pgh.pa.us 3189 : 0 : LWLockAcquire(SerializablePredicateListLock, LW_EXCLUSIVE);
3190 : :
3191 : : /*
3192 : : * Try copying the locks over to the new page's tag, creating it if
3193 : : * necessary.
3194 : : */
5515 heikki.linnakangas@i 3195 : 0 : success = TransferPredicateLocksToNewTarget(oldtargettag,
3196 : : newtargettag,
3197 : : false);
3198 : :
3199 [ # # ]: 0 : if (!success)
3200 : : {
3201 : : /*
3202 : : * No more predicate lock entries are available. Failure isn't an
3203 : : * option here, so promote the page lock to a relation lock.
3204 : : */
3205 : :
3206 : : /* Get the parent relation lock's lock tag */
3207 : 0 : success = GetParentPredicateLockTag(&oldtargettag,
3208 : : &newtargettag);
3209 [ # # ]: 0 : Assert(success);
3210 : :
3211 : : /*
3212 : : * Move the locks to the parent. This shouldn't fail.
3213 : : *
3214 : : * Note that here we are removing locks held by other backends,
3215 : : * leading to a possible inconsistency in their local lock hash table.
3216 : : * This is OK because we're replacing it with a lock that covers the
3217 : : * old one.
3218 : : */
3219 : 0 : success = TransferPredicateLocksToNewTarget(oldtargettag,
3220 : : newtargettag,
3221 : : true);
3222 [ # # ]: 0 : Assert(success);
3223 : : }
3224 : :
2130 tgl@sss.pgh.pa.us 3225 : 0 : LWLockRelease(SerializablePredicateListLock);
3226 : : }
3227 : :
3228 : : /*
3229 : : * PredicateLockPageCombine
3230 : : *
3231 : : * Combines predicate locks for two existing pages.
3232 : : * Skip if this is a temporary table or toast table.
3233 : : *
3234 : : * NOTE: A page combine affects all serializable transactions, even if it
3235 : : * occurs in the context of another transaction isolation level.
3236 : : */
3237 : : void
5380 heikki.linnakangas@i 3238 :CBC 3076 : PredicateLockPageCombine(Relation relation, BlockNumber oldblkno,
3239 : : BlockNumber newblkno)
3240 : : {
3241 : : /*
3242 : : * Page combines differ from page splits in that we ought to be able to
3243 : : * remove the locks on the old page after transferring them to the new
3244 : : * page, instead of duplicating them. However, because we can't edit other
3245 : : * backends' local lock tables, removing the old lock would leave them
3246 : : * with an entry in their LocalPredicateLockHash for a lock they're not
3247 : : * holding, which isn't acceptable. So we wind up having to do the same
3248 : : * work as a page split, acquiring a lock on the new page and keeping the
3249 : : * old page locked too. That can lead to some false positives, but should
3250 : : * be rare in practice.
3251 : : */
5493 3252 : 3076 : PredicateLockPageSplit(relation, oldblkno, newblkno);
5515 3253 : 3076 : }
3254 : :
3255 : : /*
3256 : : * Walk the list of in-progress serializable transactions and find the new
3257 : : * xmin.
3258 : : */
3259 : : static void
3260 : 868 : SetNewSxactGlobalXmin(void)
3261 : : {
3262 : : dlist_iter iter;
3263 : :
3264 [ - + ]: 868 : Assert(LWLockHeldByMe(SerializableXactHashLock));
3265 : :
3266 : 868 : PredXact->SxactGlobalXmin = InvalidTransactionId;
3267 : 868 : PredXact->SxactGlobalXminCount = 0;
3268 : :
1151 andres@anarazel.de 3269 [ + - + + ]: 3312 : dlist_foreach(iter, &PredXact->activeList)
3270 : : {
3271 : 2444 : SERIALIZABLEXACT *sxact =
1031 tgl@sss.pgh.pa.us 3272 : 2444 : dlist_container(SERIALIZABLEXACT, xactLink, iter.cur);
3273 : :
5381 heikki.linnakangas@i 3274 [ + + ]: 2444 : if (!SxactIsRolledBack(sxact)
5515 3275 [ + + ]: 2146 : && !SxactIsCommitted(sxact)
3276 [ + - ]: 22 : && sxact != OldCommittedSxact)
3277 : : {
3278 [ - + ]: 22 : Assert(sxact->xmin != InvalidTransactionId);
3279 [ + + ]: 22 : if (!TransactionIdIsValid(PredXact->SxactGlobalXmin)
5515 heikki.linnakangas@i 3280 [ - + ]:GBC 1 : || TransactionIdPrecedes(sxact->xmin,
3281 : 1 : PredXact->SxactGlobalXmin))
3282 : : {
5515 heikki.linnakangas@i 3283 :CBC 21 : PredXact->SxactGlobalXmin = sxact->xmin;
3284 : 21 : PredXact->SxactGlobalXminCount = 1;
3285 : : }
5515 heikki.linnakangas@i 3286 [ - + ]:GBC 1 : else if (TransactionIdEquals(sxact->xmin,
3287 : : PredXact->SxactGlobalXmin))
5515 heikki.linnakangas@i 3288 :UBC 0 : PredXact->SxactGlobalXminCount++;
3289 : : }
3290 : : }
3291 : :
2130 tgl@sss.pgh.pa.us 3292 :CBC 868 : SerialSetActiveSerXmin(PredXact->SxactGlobalXmin);
5515 heikki.linnakangas@i 3293 : 868 : }
3294 : :
3295 : : /*
3296 : : * ReleasePredicateLocks
3297 : : *
3298 : : * Releases predicate locks based on completion of the current transaction,
3299 : : * whether committed or rolled back. It can also be called for a read only
3300 : : * transaction when it becomes impossible for the transaction to become
3301 : : * part of a dangerous structure.
3302 : : *
3303 : : * We do nothing unless this is a serializable transaction.
3304 : : *
3305 : : * This method must ensure that shared memory hash tables are cleaned
3306 : : * up in some relatively timely fashion.
3307 : : *
3308 : : * If this transaction is committing and is holding any predicate locks,
3309 : : * it must be added to a list of completed serializable transactions still
3310 : : * holding locks.
3311 : : *
3312 : : * If isReadOnlySafe is true, then predicate locks are being released before
3313 : : * the end of the transaction because MySerializableXact has been determined
3314 : : * to be RO_SAFE. In non-parallel mode we can release it completely, but it
3315 : : * in parallel mode we partially release the SERIALIZABLEXACT and keep it
3316 : : * around until the end of the transaction, allowing each backend to clear its
3317 : : * MySerializableXact variable and benefit from the optimization in its own
3318 : : * time.
3319 : : */
3320 : : void
2557 tmunro@postgresql.or 3321 : 337872 : ReleasePredicateLocks(bool isCommit, bool isReadOnlySafe)
3322 : : {
1105 3323 : 337872 : bool partiallyReleasing = false;
3324 : : bool needToClear;
3325 : : SERIALIZABLEXACT *roXact;
3326 : : dlist_mutable_iter iter;
3327 : :
3328 : : /*
3329 : : * We can't trust XactReadOnly here, because a transaction which started
3330 : : * as READ WRITE can show as READ ONLY later, e.g., within
3331 : : * subtransactions. We want to flag a transaction as READ ONLY if it
3332 : : * commits without writing so that de facto READ ONLY transactions get the
3333 : : * benefit of some RO optimizations, so we will use this local variable to
3334 : : * get some cleanup logic right which is based on whether the transaction
3335 : : * was declared READ ONLY at the top level.
3336 : : */
3337 : : bool topLevelIsDeclaredReadOnly;
3338 : :
3339 : : /* We can't be both committing and releasing early due to RO_SAFE. */
2557 3340 [ + + - + ]: 337872 : Assert(!(isCommit && isReadOnlySafe));
3341 : :
3342 : : /* Are we at the end of a transaction, that is, a commit or abort? */
3343 [ + + ]: 337872 : if (!isReadOnlySafe)
3344 : : {
3345 : : /*
3346 : : * Parallel workers mustn't release predicate locks at the end of
3347 : : * their transaction. The leader will do that at the end of its
3348 : : * transaction.
3349 : : */
3350 [ + + ]: 337837 : if (IsParallelWorker())
3351 : : {
3352 : 4473 : ReleasePredicateLocksLocal();
3353 : 336313 : return;
3354 : : }
3355 : :
3356 : : /*
3357 : : * By the time the leader in a parallel query reaches end of
3358 : : * transaction, it has waited for all workers to exit.
3359 : : */
3360 [ - + ]: 333364 : Assert(!ParallelContextActive());
3361 : :
3362 : : /*
3363 : : * If the leader in a parallel query earlier stashed a partially
3364 : : * released SERIALIZABLEXACT for final clean-up at end of transaction
3365 : : * (because workers might still have been accessing it), then it's
3366 : : * time to restore it.
3367 : : */
3368 [ + + ]: 333364 : if (SavedSerializableXact != InvalidSerializableXact)
3369 : : {
3370 [ - + ]: 1 : Assert(MySerializableXact == InvalidSerializableXact);
3371 : 1 : MySerializableXact = SavedSerializableXact;
3372 : 1 : SavedSerializableXact = InvalidSerializableXact;
3373 [ - + ]: 1 : Assert(SxactIsPartiallyReleased(MySerializableXact));
3374 : : }
3375 : : }
3376 : :
5515 heikki.linnakangas@i 3377 [ + + ]: 333399 : if (MySerializableXact == InvalidSerializableXact)
3378 : : {
3379 [ - + ]: 331837 : Assert(LocalPredicateLockHash == NULL);
3380 : 331837 : return;
3381 : : }
3382 : :
3788 kgrittn@postgresql.o 3383 : 1562 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
3384 : :
3385 : : /*
3386 : : * If the transaction is committing, but it has been partially released
3387 : : * already, then treat this as a roll back. It was marked as rolled back.
3388 : : */
2557 tmunro@postgresql.or 3389 [ + + + + ]: 1562 : if (isCommit && SxactIsPartiallyReleased(MySerializableXact))
3390 : 2 : isCommit = false;
3391 : :
3392 : : /*
3393 : : * If we're called in the middle of a transaction because we discovered
3394 : : * that the SXACT_FLAG_RO_SAFE flag was set, then we'll partially release
3395 : : * it (that is, release the predicate locks and conflicts, but not the
3396 : : * SERIALIZABLEXACT itself) if we're the first backend to have noticed.
3397 : : */
3398 [ + + + + ]: 1562 : if (isReadOnlySafe && IsInParallelMode())
3399 : : {
3400 : : /*
3401 : : * The leader needs to stash a pointer to it, so that it can
3402 : : * completely release it at end-of-transaction.
3403 : : */
3404 [ + + ]: 5 : if (!IsParallelWorker())
3405 : 1 : SavedSerializableXact = MySerializableXact;
3406 : :
3407 : : /*
3408 : : * The first backend to reach this condition will partially release
3409 : : * the SERIALIZABLEXACT. All others will just clear their
3410 : : * backend-local state so that they stop doing SSI checks for the rest
3411 : : * of the transaction.
3412 : : */
3413 [ + + ]: 5 : if (SxactIsPartiallyReleased(MySerializableXact))
3414 : : {
3415 : 3 : LWLockRelease(SerializableXactHashLock);
3416 : 3 : ReleasePredicateLocksLocal();
3417 : 3 : return;
3418 : : }
3419 : : else
3420 : : {
3421 : 2 : MySerializableXact->flags |= SXACT_FLAG_PARTIALLY_RELEASED;
1105 3422 : 2 : partiallyReleasing = true;
3423 : : /* ... and proceed to perform the partial release below. */
3424 : : }
3425 : : }
5515 heikki.linnakangas@i 3426 [ + + - + ]: 1559 : Assert(!isCommit || SxactIsPrepared(MySerializableXact));
5387 3427 [ + + - + ]: 1559 : Assert(!isCommit || !SxactIsDoomed(MySerializableXact));
5515 3428 [ - + ]: 1559 : Assert(!SxactIsCommitted(MySerializableXact));
2557 tmunro@postgresql.or 3429 [ + + - + ]: 1559 : Assert(SxactIsPartiallyReleased(MySerializableXact)
3430 : : || !SxactIsRolledBack(MySerializableXact));
3431 : :
3432 : : /* may not be serializable during COMMIT/ROLLBACK PREPARED */
3788 kgrittn@postgresql.o 3433 [ + + - + ]: 1559 : Assert(MySerializableXact->pid == 0 || IsolationIsSerializable());
3434 : :
3435 : : /* We'd better not already be on the cleanup list. */
5392 heikki.linnakangas@i 3436 [ - + ]: 1559 : Assert(!SxactIsOnFinishedList(MySerializableXact));
3437 : :
5515 3438 : 1559 : topLevelIsDeclaredReadOnly = SxactIsReadOnly(MySerializableXact);
3439 : :
3440 : : /*
3441 : : * We don't hold XidGenLock lock here, assuming that TransactionId is
3442 : : * atomic!
3443 : : *
3444 : : * If this value is changing, we don't care that much whether we get the
3445 : : * old or new value -- it is just used to determine how far
3446 : : * SxactGlobalXmin must advance before this transaction can be fully
3447 : : * cleaned up. The worst that could happen is we wait for one more
3448 : : * transaction to complete before freeing some RAM; correctness of visible
3449 : : * behavior is not affected.
3450 : : */
828 3451 : 1559 : MySerializableXact->finishedBefore = XidFromFullTransactionId(TransamVariables->nextXid);
3452 : :
3453 : : /*
3454 : : * If it's not a commit it's either a rollback or a read-only transaction
3455 : : * flagged SXACT_FLAG_RO_SAFE, and we can clear our locks immediately.
3456 : : */
5515 3457 [ + + ]: 1559 : if (isCommit)
3458 : : {
3459 : 1234 : MySerializableXact->flags |= SXACT_FLAG_COMMITTED;
3460 : 1234 : MySerializableXact->commitSeqNo = ++(PredXact->LastSxactCommitSeqNo);
3461 : : /* Recognize implicit read-only transaction (commit without write). */
5392 3462 [ + + ]: 1234 : if (!MyXactDidWrite)
5515 3463 : 233 : MySerializableXact->flags |= SXACT_FLAG_READ_ONLY;
3464 : : }
3465 : : else
3466 : : {
3467 : : /*
3468 : : * The DOOMED flag indicates that we intend to roll back this
3469 : : * transaction and so it should not cause serialization failures for
3470 : : * other transactions that conflict with it. Note that this flag might
3471 : : * already be set, if another backend marked this transaction for
3472 : : * abort.
3473 : : *
3474 : : * The ROLLED_BACK flag further indicates that ReleasePredicateLocks
3475 : : * has been called, and so the SerializableXact is eligible for
3476 : : * cleanup. This means it should not be considered when calculating
3477 : : * SxactGlobalXmin.
3478 : : */
2502 tmunro@postgresql.or 3479 : 325 : MySerializableXact->flags |= SXACT_FLAG_DOOMED;
5381 heikki.linnakangas@i 3480 : 325 : MySerializableXact->flags |= SXACT_FLAG_ROLLED_BACK;
3481 : :
3482 : : /*
3483 : : * If the transaction was previously prepared, but is now failing due
3484 : : * to a ROLLBACK PREPARED or (hopefully very rare) error after the
3485 : : * prepare, clear the prepared flag. This simplifies conflict
3486 : : * checking.
3487 : : */
5365 3488 : 325 : MySerializableXact->flags &= ~SXACT_FLAG_PREPARED;
3489 : : }
3490 : :
5515 3491 [ + + ]: 1559 : if (!topLevelIsDeclaredReadOnly)
3492 : : {
3493 [ - + ]: 1449 : Assert(PredXact->WritableSxactCount > 0);
3494 [ + + ]: 1449 : if (--(PredXact->WritableSxactCount) == 0)
3495 : : {
3496 : : /*
3497 : : * Release predicate locks and rw-conflicts in for all committed
3498 : : * transactions. There are no longer any transactions which might
3499 : : * conflict with the locks and no chance for new transactions to
3500 : : * overlap. Similarly, existing conflicts in can't cause pivots,
3501 : : * and any conflicts in which could have completed a dangerous
3502 : : * structure would already have caused a rollback, so any
3503 : : * remaining ones must be benign.
3504 : : */
3505 : 857 : PredXact->CanPartialClearThrough = PredXact->LastSxactCommitSeqNo;
3506 : : }
3507 : : }
3508 : : else
3509 : : {
3510 : : /*
3511 : : * Read-only transactions: clear the list of transactions that might
3512 : : * make us unsafe. Note that we use 'inLink' for the iteration as
3513 : : * opposed to 'outLink' for the r/w xacts.
3514 : : */
1151 andres@anarazel.de 3515 [ + - + + ]: 152 : dlist_foreach_modify(iter, &MySerializableXact->possibleUnsafeConflicts)
3516 : : {
3517 : 42 : RWConflict possibleUnsafeConflict =
1031 tgl@sss.pgh.pa.us 3518 : 42 : dlist_container(RWConflictData, inLink, iter.cur);
3519 : :
5515 heikki.linnakangas@i 3520 [ - + ]: 42 : Assert(!SxactIsReadOnly(possibleUnsafeConflict->sxactOut));
3521 [ - + ]: 42 : Assert(MySerializableXact == possibleUnsafeConflict->sxactIn);
3522 : :
3523 : 42 : ReleaseRWConflict(possibleUnsafeConflict);
3524 : : }
3525 : : }
3526 : :
3527 : : /* Check for conflict out to old committed transactions. */
3528 [ + + ]: 1559 : if (isCommit
3529 [ + + ]: 1234 : && !SxactIsReadOnly(MySerializableXact)
3530 [ - + ]: 1001 : && SxactHasSummaryConflictOut(MySerializableXact))
3531 : : {
3532 : : /*
3533 : : * we don't know which old committed transaction we conflicted with,
3534 : : * so be conservative and use FirstNormalSerCommitSeqNo here
3535 : : */
5515 heikki.linnakangas@i 3536 :UBC 0 : MySerializableXact->SeqNo.earliestOutConflictCommit =
3537 : : FirstNormalSerCommitSeqNo;
3538 : 0 : MySerializableXact->flags |= SXACT_FLAG_CONFLICT_OUT;
3539 : : }
3540 : :
3541 : : /*
3542 : : * Release all outConflicts to committed transactions. If we're rolling
3543 : : * back clear them all. Set SXACT_FLAG_CONFLICT_OUT if any point to
3544 : : * previously committed transactions.
3545 : : */
1151 andres@anarazel.de 3546 [ + - + + ]:CBC 2245 : dlist_foreach_modify(iter, &MySerializableXact->outConflicts)
3547 : : {
3548 : 686 : RWConflict conflict =
1031 tgl@sss.pgh.pa.us 3549 :ECB (679) : dlist_container(RWConflictData, outLink, iter.cur);
3550 : :
5515 heikki.linnakangas@i 3551 [ + + ]:CBC 686 : if (isCommit
3552 [ + + ]: 454 : && !SxactIsReadOnly(MySerializableXact)
3553 [ + + ]: 346 : && SxactIsCommitted(conflict->sxactIn))
3554 : : {
3555 [ - + ]: 96 : if ((MySerializableXact->flags & SXACT_FLAG_CONFLICT_OUT) == 0
5365 heikki.linnakangas@i 3556 [ # # ]:UBC 0 : || conflict->sxactIn->prepareSeqNo < MySerializableXact->SeqNo.earliestOutConflictCommit)
5365 heikki.linnakangas@i 3557 :CBC 96 : MySerializableXact->SeqNo.earliestOutConflictCommit = conflict->sxactIn->prepareSeqNo;
5515 3558 : 96 : MySerializableXact->flags |= SXACT_FLAG_CONFLICT_OUT;
3559 : : }
3560 : :
3561 [ + + ]: 686 : if (!isCommit
3562 [ + + ]: 454 : || SxactIsCommitted(conflict->sxactIn)
3563 [ - + ]: 336 : || (conflict->sxactIn->SeqNo.lastCommitBeforeSnapshot >= PredXact->LastSxactCommitSeqNo))
3564 : 350 : ReleaseRWConflict(conflict);
3565 : : }
3566 : :
3567 : : /*
3568 : : * Release all inConflicts from committed and read-only transactions. If
3569 : : * we're rolling back, clear them all.
3570 : : */
1151 andres@anarazel.de 3571 [ + - + + ]: 2339 : dlist_foreach_modify(iter, &MySerializableXact->inConflicts)
3572 : : {
3573 : 780 : RWConflict conflict =
1031 tgl@sss.pgh.pa.us 3574 : 780 : dlist_container(RWConflictData, inLink, iter.cur);
3575 : :
5515 heikki.linnakangas@i 3576 [ + + ]: 780 : if (!isCommit
3577 [ + + ]: 603 : || SxactIsCommitted(conflict->sxactOut)
3578 [ + + ]: 418 : || SxactIsReadOnly(conflict->sxactOut))
3579 : 442 : ReleaseRWConflict(conflict);
3580 : : }
3581 : :
3582 [ + + ]: 1559 : if (!topLevelIsDeclaredReadOnly)
3583 : : {
3584 : : /*
3585 : : * Remove ourselves from the list of possible conflicts for concurrent
3586 : : * READ ONLY transactions, flagging them as unsafe if we have a
3587 : : * conflict out. If any are waiting DEFERRABLE transactions, wake them
3588 : : * up if they are known safe or known unsafe.
3589 : : */
1151 andres@anarazel.de 3590 [ + - + + ]: 1542 : dlist_foreach_modify(iter, &MySerializableXact->possibleUnsafeConflicts)
3591 : : {
3592 : 93 : RWConflict possibleUnsafeConflict =
1031 tgl@sss.pgh.pa.us 3593 :ECB (93) : dlist_container(RWConflictData, outLink, iter.cur);
3594 : :
5515 heikki.linnakangas@i 3595 :CBC 93 : roXact = possibleUnsafeConflict->sxactIn;
3596 [ - + ]: 93 : Assert(MySerializableXact == possibleUnsafeConflict->sxactOut);
3597 [ - + ]: 93 : Assert(SxactIsReadOnly(roXact));
3598 : :
3599 : : /* Mark conflicted if necessary. */
3600 [ + + ]: 93 : if (isCommit
5392 3601 [ + + ]: 90 : && MyXactDidWrite
5515 3602 [ + + ]: 85 : && SxactHasConflictOut(MySerializableXact)
3603 : 13 : && (MySerializableXact->SeqNo.earliestOutConflictCommit
3604 [ + + ]: 13 : <= roXact->SeqNo.lastCommitBeforeSnapshot))
3605 : : {
3606 : : /*
3607 : : * This releases possibleUnsafeConflict (as well as all other
3608 : : * possible conflicts for roXact)
3609 : : */
3610 : 3 : FlagSxactUnsafe(roXact);
3611 : : }
3612 : : else
3613 : : {
3614 : 90 : ReleaseRWConflict(possibleUnsafeConflict);
3615 : :
3616 : : /*
3617 : : * If we were the last possible conflict, flag it safe. The
3618 : : * transaction can now safely release its predicate locks (but
3619 : : * that transaction's backend has to do that itself).
3620 : : */
1151 andres@anarazel.de 3621 [ + + ]: 90 : if (dlist_is_empty(&roXact->possibleUnsafeConflicts))
5515 heikki.linnakangas@i 3622 : 67 : roXact->flags |= SXACT_FLAG_RO_SAFE;
3623 : : }
3624 : :
3625 : : /*
3626 : : * Wake up the process for a waiting DEFERRABLE transaction if we
3627 : : * now know it's either safe or conflicted.
3628 : : */
3629 [ + + ]: 93 : if (SxactIsDeferrableWaiting(roXact) &&
3630 [ + + + + ]: 4 : (SxactIsROUnsafe(roXact) || SxactIsROSafe(roXact)))
1550 tmunro@postgresql.or 3631 : 3 : ProcSendSignal(roXact->pgprocno);
3632 : : }
3633 : : }
3634 : :
3635 : : /*
3636 : : * Check whether it's time to clean up old transactions. This can only be
3637 : : * done when the last serializable transaction with the oldest xmin among
3638 : : * serializable transactions completes. We then find the "new oldest"
3639 : : * xmin and purge any transactions which finished before this transaction
3640 : : * was launched.
3641 : : *
3642 : : * For parallel queries in read-only transactions, it might run twice. We
3643 : : * only release the reference on the first call.
3644 : : */
5515 heikki.linnakangas@i 3645 : 1559 : needToClear = false;
1105 tmunro@postgresql.or 3646 [ + + ]: 1559 : if ((partiallyReleasing ||
3647 [ + + ]: 1557 : !SxactIsPartiallyReleased(MySerializableXact)) &&
3648 [ + + ]: 1557 : TransactionIdEquals(MySerializableXact->xmin,
3649 : : PredXact->SxactGlobalXmin))
3650 : : {
5515 heikki.linnakangas@i 3651 [ - + ]: 1540 : Assert(PredXact->SxactGlobalXminCount > 0);
3652 [ + + ]: 1540 : if (--(PredXact->SxactGlobalXminCount) == 0)
3653 : : {
3654 : 868 : SetNewSxactGlobalXmin();
3655 : 868 : needToClear = true;
3656 : : }
3657 : : }
3658 : :
3659 : 1559 : LWLockRelease(SerializableXactHashLock);
3660 : :
3661 : 1559 : LWLockAcquire(SerializableFinishedListLock, LW_EXCLUSIVE);
3662 : :
3663 : : /* Add this to the list of transactions to check for later cleanup. */
3664 [ + + ]: 1559 : if (isCommit)
1151 andres@anarazel.de 3665 : 1234 : dlist_push_tail(FinishedSerializableTransactions,
3666 : 1234 : &MySerializableXact->finishedLink);
3667 : :
3668 : : /*
3669 : : * If we're releasing a RO_SAFE transaction in parallel mode, we'll only
3670 : : * partially release it. That's necessary because other backends may have
3671 : : * a reference to it. The leader will release the SERIALIZABLEXACT itself
3672 : : * at the end of the transaction after workers have stopped running.
3673 : : */
5515 heikki.linnakangas@i 3674 [ + + ]: 1559 : if (!isCommit)
2557 tmunro@postgresql.or 3675 : 325 : ReleaseOneSerializableXact(MySerializableXact,
3676 [ + + + + ]: 325 : isReadOnlySafe && IsInParallelMode(),
3677 : : false);
3678 : :
5515 heikki.linnakangas@i 3679 : 1559 : LWLockRelease(SerializableFinishedListLock);
3680 : :
3681 [ + + ]: 1559 : if (needToClear)
3682 : 868 : ClearOldPredicateLocks();
3683 : :
2557 tmunro@postgresql.or 3684 : 1559 : ReleasePredicateLocksLocal();
3685 : : }
3686 : :
3687 : : static void
3688 : 6035 : ReleasePredicateLocksLocal(void)
3689 : : {
5515 heikki.linnakangas@i 3690 : 6035 : MySerializableXact = InvalidSerializableXact;
5392 3691 : 6035 : MyXactDidWrite = false;
3692 : :
3693 : : /* Delete per-transaction lock table */
5515 3694 [ + + ]: 6035 : if (LocalPredicateLockHash != NULL)
3695 : : {
3696 : 1558 : hash_destroy(LocalPredicateLockHash);
3697 : 1558 : LocalPredicateLockHash = NULL;
3698 : : }
3699 : 6035 : }
3700 : :
3701 : : /*
3702 : : * Clear old predicate locks, belonging to committed transactions that are no
3703 : : * longer interesting to any in-progress transaction.
3704 : : */
3705 : : static void
3706 : 868 : ClearOldPredicateLocks(void)
3707 : : {
3708 : : dlist_mutable_iter iter;
3709 : :
3710 : : /*
3711 : : * Loop through finished transactions. They are in commit order, so we can
3712 : : * stop as soon as we find one that's still interesting.
3713 : : */
3714 : 868 : LWLockAcquire(SerializableFinishedListLock, LW_EXCLUSIVE);
3715 : 868 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
1151 andres@anarazel.de 3716 [ + - + + ]: 2111 : dlist_foreach_modify(iter, FinishedSerializableTransactions)
3717 : : {
3718 : 1253 : SERIALIZABLEXACT *finishedSxact =
1031 tgl@sss.pgh.pa.us 3719 : 1253 : dlist_container(SERIALIZABLEXACT, finishedLink, iter.cur);
3720 : :
5515 heikki.linnakangas@i 3721 [ + + ]: 1253 : if (!TransactionIdIsValid(PredXact->SxactGlobalXmin)
3722 [ + + ]: 29 : || TransactionIdPrecedesOrEquals(finishedSxact->finishedBefore,
3723 : 29 : PredXact->SxactGlobalXmin))
3724 : : {
3725 : : /*
3726 : : * This transaction committed before any in-progress transaction
3727 : : * took its snapshot. It's no longer interesting.
3728 : : */
3729 : 1234 : LWLockRelease(SerializableXactHashLock);
1151 andres@anarazel.de 3730 : 1234 : dlist_delete_thoroughly(&finishedSxact->finishedLink);
5515 heikki.linnakangas@i 3731 : 1234 : ReleaseOneSerializableXact(finishedSxact, false, false);
3732 : 1234 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
3733 : : }
3734 [ + - ]: 19 : else if (finishedSxact->commitSeqNo > PredXact->HavePartialClearedThrough
3189 tgl@sss.pgh.pa.us 3735 [ + + ]: 19 : && finishedSxact->commitSeqNo <= PredXact->CanPartialClearThrough)
3736 : : {
3737 : : /*
3738 : : * Any active transactions that took their snapshot before this
3739 : : * transaction committed are read-only, so we can clear part of
3740 : : * its state.
3741 : : */
5515 heikki.linnakangas@i 3742 : 9 : LWLockRelease(SerializableXactHashLock);
3743 : :
5170 3744 [ - + ]: 9 : if (SxactIsReadOnly(finishedSxact))
3745 : : {
3746 : : /* A read-only transaction can be removed entirely */
1151 andres@anarazel.de 3747 :UBC 0 : dlist_delete_thoroughly(&(finishedSxact->finishedLink));
5170 heikki.linnakangas@i 3748 : 0 : ReleaseOneSerializableXact(finishedSxact, false, false);
3749 : : }
3750 : : else
3751 : : {
3752 : : /*
3753 : : * A read-write transaction can only be partially cleared. We
3754 : : * need to keep the SERIALIZABLEXACT but can release the
3755 : : * SIREAD locks and conflicts in.
3756 : : */
5170 heikki.linnakangas@i 3757 :CBC 9 : ReleaseOneSerializableXact(finishedSxact, true, false);
3758 : : }
3759 : :
5515 3760 : 9 : PredXact->HavePartialClearedThrough = finishedSxact->commitSeqNo;
3761 : 9 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
3762 : : }
3763 : : else
3764 : : {
3765 : : /* Still interesting. */
3766 : : break;
3767 : : }
3768 : : }
3769 : 868 : LWLockRelease(SerializableXactHashLock);
3770 : :
3771 : : /*
3772 : : * Loop through predicate locks on dummy transaction for summarized data.
3773 : : */
2130 tgl@sss.pgh.pa.us 3774 : 868 : LWLockAcquire(SerializablePredicateListLock, LW_SHARED);
1151 andres@anarazel.de 3775 [ + - - + ]: 868 : dlist_foreach_modify(iter, &OldCommittedSxact->predicateLocks)
3776 : : {
1151 andres@anarazel.de 3777 :UBC 0 : PREDICATELOCK *predlock =
1031 tgl@sss.pgh.pa.us 3778 : 0 : dlist_container(PREDICATELOCK, xactLink, iter.cur);
3779 : : bool canDoPartialCleanup;
3780 : :
5515 heikki.linnakangas@i 3781 : 0 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
5452 3782 [ # # ]: 0 : Assert(predlock->commitSeqNo != 0);
3783 [ # # ]: 0 : Assert(predlock->commitSeqNo != InvalidSerCommitSeqNo);
5515 3784 : 0 : canDoPartialCleanup = (predlock->commitSeqNo <= PredXact->CanPartialClearThrough);
3785 : 0 : LWLockRelease(SerializableXactHashLock);
3786 : :
3787 : : /*
3788 : : * If this lock originally belonged to an old enough transaction, we
3789 : : * can release it.
3790 : : */
3791 [ # # ]: 0 : if (canDoPartialCleanup)
3792 : : {
3793 : : PREDICATELOCKTAG tag;
3794 : : PREDICATELOCKTARGET *target;
3795 : : PREDICATELOCKTARGETTAG targettag;
3796 : : uint32 targettaghash;
3797 : : LWLock *partitionLock;
3798 : :
3799 : 0 : tag = predlock->tag;
3800 : 0 : target = tag.myTarget;
3801 : 0 : targettag = target->tag;
3802 : 0 : targettaghash = PredicateLockTargetTagHashCode(&targettag);
3803 : 0 : partitionLock = PredicateLockHashPartitionLock(targettaghash);
3804 : :
3805 : 0 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
3806 : :
1151 andres@anarazel.de 3807 : 0 : dlist_delete(&(predlock->targetLink));
3808 : 0 : dlist_delete(&(predlock->xactLink));
3809 : :
5515 heikki.linnakangas@i 3810 : 0 : hash_search_with_hash_value(PredicateLockHash, &tag,
3189 tgl@sss.pgh.pa.us 3811 : 0 : PredicateLockHashCodeFromTargetHashCode(&tag,
3812 : : targettaghash),
3813 : : HASH_REMOVE, NULL);
5515 heikki.linnakangas@i 3814 : 0 : RemoveTargetIfNoLongerUsed(target, targettaghash);
3815 : :
3816 : 0 : LWLockRelease(partitionLock);
3817 : : }
3818 : : }
3819 : :
2130 tgl@sss.pgh.pa.us 3820 :CBC 868 : LWLockRelease(SerializablePredicateListLock);
5515 heikki.linnakangas@i 3821 : 868 : LWLockRelease(SerializableFinishedListLock);
3822 : 868 : }
3823 : :
3824 : : /*
3825 : : * This is the normal way to delete anything from any of the predicate
3826 : : * locking hash tables. Given a transaction which we know can be deleted:
3827 : : * delete all predicate locks held by that transaction and any predicate
3828 : : * lock targets which are now unreferenced by a lock; delete all conflicts
3829 : : * for the transaction; delete all xid values for the transaction; then
3830 : : * delete the transaction.
3831 : : *
3832 : : * When the partial flag is set, we can release all predicate locks and
3833 : : * in-conflict information -- we've established that there are no longer
3834 : : * any overlapping read write transactions for which this transaction could
3835 : : * matter -- but keep the transaction entry itself and any outConflicts.
3836 : : *
3837 : : * When the summarize flag is set, we've run short of room for sxact data
3838 : : * and must summarize to the SLRU. Predicate locks are transferred to a
3839 : : * dummy "old" transaction, with duplicate locks on a single target
3840 : : * collapsing to a single lock with the "latest" commitSeqNo from among
3841 : : * the conflicting locks..
3842 : : */
3843 : : static void
3844 : 1568 : ReleaseOneSerializableXact(SERIALIZABLEXACT *sxact, bool partial,
3845 : : bool summarize)
3846 : : {
3847 : : SERIALIZABLEXIDTAG sxidtag;
3848 : : dlist_mutable_iter iter;
3849 : :
3850 [ - + ]: 1568 : Assert(sxact != NULL);
5381 3851 [ + + - + ]: 1568 : Assert(SxactIsRolledBack(sxact) || SxactIsCommitted(sxact));
5170 3852 [ + + - + ]: 1568 : Assert(partial || !SxactIsOnFinishedList(sxact));
5515 3853 [ - + ]: 1568 : Assert(LWLockHeldByMe(SerializableFinishedListLock));
3854 : :
3855 : : /*
3856 : : * First release all the predicate locks held by this xact (or transfer
3857 : : * them to OldCommittedSxact if summarize is true)
3858 : : */
2130 tgl@sss.pgh.pa.us 3859 : 1568 : LWLockAcquire(SerializablePredicateListLock, LW_SHARED);
2557 tmunro@postgresql.or 3860 [ + + ]: 1568 : if (IsInParallelMode())
2130 tgl@sss.pgh.pa.us 3861 : 3 : LWLockAcquire(&sxact->perXactPredicateListLock, LW_EXCLUSIVE);
1151 andres@anarazel.de 3862 [ + - + + ]: 4427 : dlist_foreach_modify(iter, &sxact->predicateLocks)
3863 : : {
3864 : 2859 : PREDICATELOCK *predlock =
1031 tgl@sss.pgh.pa.us 3865 : 2859 : dlist_container(PREDICATELOCK, xactLink, iter.cur);
3866 : : PREDICATELOCKTAG tag;
3867 : : PREDICATELOCKTARGET *target;
3868 : : PREDICATELOCKTARGETTAG targettag;
3869 : : uint32 targettaghash;
3870 : : LWLock *partitionLock;
3871 : :
5515 heikki.linnakangas@i 3872 : 2859 : tag = predlock->tag;
3873 : 2859 : target = tag.myTarget;
3874 : 2859 : targettag = target->tag;
3875 : 2859 : targettaghash = PredicateLockTargetTagHashCode(&targettag);
3876 : 2859 : partitionLock = PredicateLockHashPartitionLock(targettaghash);
3877 : :
3878 : 2859 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
3879 : :
1151 andres@anarazel.de 3880 : 2859 : dlist_delete(&predlock->targetLink);
3881 : :
5515 heikki.linnakangas@i 3882 : 2859 : hash_search_with_hash_value(PredicateLockHash, &tag,
3189 tgl@sss.pgh.pa.us 3883 : 2859 : PredicateLockHashCodeFromTargetHashCode(&tag,
3884 : : targettaghash),
3885 : : HASH_REMOVE, NULL);
5515 heikki.linnakangas@i 3886 [ - + ]: 2859 : if (summarize)
3887 : : {
3888 : : bool found;
3889 : :
3890 : : /* Fold into dummy transaction list. */
5515 heikki.linnakangas@i 3891 :UBC 0 : tag.myXact = OldCommittedSxact;
3892 : 0 : predlock = hash_search_with_hash_value(PredicateLockHash, &tag,
3189 tgl@sss.pgh.pa.us 3893 : 0 : PredicateLockHashCodeFromTargetHashCode(&tag,
3894 : : targettaghash),
3895 : : HASH_ENTER_NULL, &found);
5515 heikki.linnakangas@i 3896 [ # # ]: 0 : if (!predlock)
3897 [ # # ]: 0 : ereport(ERROR,
3898 : : (errcode(ERRCODE_OUT_OF_MEMORY),
3899 : : errmsg("out of shared memory"),
3900 : : errhint("You might need to increase \"%s\".", "max_pred_locks_per_transaction")));
3901 [ # # ]: 0 : if (found)
3902 : : {
5452 3903 [ # # ]: 0 : Assert(predlock->commitSeqNo != 0);
3904 [ # # ]: 0 : Assert(predlock->commitSeqNo != InvalidSerCommitSeqNo);
5515 3905 [ # # ]: 0 : if (predlock->commitSeqNo < sxact->commitSeqNo)
3906 : 0 : predlock->commitSeqNo = sxact->commitSeqNo;
3907 : : }
3908 : : else
3909 : : {
1151 andres@anarazel.de 3910 : 0 : dlist_push_tail(&target->predicateLocks,
3911 : : &predlock->targetLink);
3912 : 0 : dlist_push_tail(&OldCommittedSxact->predicateLocks,
3913 : : &predlock->xactLink);
5515 heikki.linnakangas@i 3914 : 0 : predlock->commitSeqNo = sxact->commitSeqNo;
3915 : : }
3916 : : }
3917 : : else
5515 heikki.linnakangas@i 3918 :CBC 2859 : RemoveTargetIfNoLongerUsed(target, targettaghash);
3919 : :
3920 : 2859 : LWLockRelease(partitionLock);
3921 : : }
3922 : :
3923 : : /*
3924 : : * Rather than retail removal, just re-init the head after we've run
3925 : : * through the list.
3926 : : */
1151 andres@anarazel.de 3927 : 1568 : dlist_init(&sxact->predicateLocks);
3928 : :
2557 tmunro@postgresql.or 3929 [ + + ]: 1568 : if (IsInParallelMode())
2130 tgl@sss.pgh.pa.us 3930 : 3 : LWLockRelease(&sxact->perXactPredicateListLock);
3931 : 1568 : LWLockRelease(SerializablePredicateListLock);
3932 : :
5515 heikki.linnakangas@i 3933 : 1568 : sxidtag.xid = sxact->topXid;
3934 : 1568 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
3935 : :
3936 : : /* Release all outConflicts (unless 'partial' is true) */
3937 [ + + ]: 1568 : if (!partial)
3938 : : {
1151 andres@anarazel.de 3939 [ + - - + ]: 1557 : dlist_foreach_modify(iter, &sxact->outConflicts)
3940 : : {
1151 andres@anarazel.de 3941 :UBC 0 : RWConflict conflict =
1031 tgl@sss.pgh.pa.us 3942 :EUB : dlist_container(RWConflictData, outLink, iter.cur);
3943 : :
5515 heikki.linnakangas@i 3944 [ # # ]:UBC 0 : if (summarize)
3945 : 0 : conflict->sxactIn->flags |= SXACT_FLAG_SUMMARY_CONFLICT_IN;
3946 : 0 : ReleaseRWConflict(conflict);
3947 : : }
3948 : : }
3949 : :
3950 : : /* Release all inConflicts. */
1151 andres@anarazel.de 3951 [ + - - + ]:CBC 1568 : dlist_foreach_modify(iter, &sxact->inConflicts)
3952 : : {
1151 andres@anarazel.de 3953 :UBC 0 : RWConflict conflict =
1031 tgl@sss.pgh.pa.us 3954 : 0 : dlist_container(RWConflictData, inLink, iter.cur);
3955 : :
5515 heikki.linnakangas@i 3956 [ # # ]: 0 : if (summarize)
3957 : 0 : conflict->sxactOut->flags |= SXACT_FLAG_SUMMARY_CONFLICT_OUT;
3958 : 0 : ReleaseRWConflict(conflict);
3959 : : }
3960 : :
3961 : : /* Finally, get rid of the xid and the record of the transaction itself. */
5515 heikki.linnakangas@i 3962 [ + + ]:CBC 1568 : if (!partial)
3963 : : {
3964 [ + + ]: 1557 : if (sxidtag.xid != InvalidTransactionId)
3965 : 1289 : hash_search(SerializableXidHash, &sxidtag, HASH_REMOVE, NULL);
3966 : 1557 : ReleasePredXact(sxact);
3967 : : }
3968 : :
3969 : 1568 : LWLockRelease(SerializableXactHashLock);
3970 : 1568 : }
3971 : :
3972 : : /*
3973 : : * Tests whether the given top level transaction is concurrent with
3974 : : * (overlaps) our current transaction.
3975 : : *
3976 : : * We need to identify the top level transaction for SSI, anyway, so pass
3977 : : * that to this function to save the overhead of checking the snapshot's
3978 : : * subxip array.
3979 : : */
3980 : : static bool
3981 : 536 : XidIsConcurrent(TransactionId xid)
3982 : : {
3983 : : Snapshot snap;
3984 : :
3985 [ - + ]: 536 : Assert(TransactionIdIsValid(xid));
3986 [ - + ]: 536 : Assert(!TransactionIdEquals(xid, GetTopTransactionIdIfAny()));
3987 : :
3988 : 536 : snap = GetTransactionSnapshot();
3989 : :
3990 [ - + ]: 536 : if (TransactionIdPrecedes(xid, snap->xmin))
5515 heikki.linnakangas@i 3991 :UBC 0 : return false;
3992 : :
5515 heikki.linnakangas@i 3993 [ + + ]:CBC 536 : if (TransactionIdFollowsOrEquals(xid, snap->xmax))
3994 : 528 : return true;
3995 : :
1270 michael@paquier.xyz 3996 : 8 : return pg_lfind32(xid, snap->xip, snap->xcnt);
3997 : : }
3998 : :
3999 : : bool
2238 tmunro@postgresql.or 4000 : 35149041 : CheckForSerializableConflictOutNeeded(Relation relation, Snapshot snapshot)
4001 : : {
4002 [ + + ]: 35149041 : if (!SerializationNeededForRead(relation, snapshot))
4003 : 35123070 : return false;
4004 : :
4005 : : /* Check if someone else has already decided that we need to die */
4006 [ - + ]: 25971 : if (SxactIsDoomed(MySerializableXact))
4007 : : {
2238 tmunro@postgresql.or 4008 [ # # ]:UBC 0 : ereport(ERROR,
4009 : : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
4010 : : errmsg("could not serialize access due to read/write dependencies among transactions"),
4011 : : errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict out checking."),
4012 : : errhint("The transaction might succeed if retried.")));
4013 : : }
4014 : :
2238 tmunro@postgresql.or 4015 :CBC 25971 : return true;
4016 : : }
4017 : :
4018 : : /*
4019 : : * CheckForSerializableConflictOut
4020 : : * A table AM is reading a tuple that has been modified. If it determines
4021 : : * that the tuple version it is reading is not visible to us, it should
4022 : : * pass in the top level xid of the transaction that created it.
4023 : : * Otherwise, if it determines that it is visible to us but it has been
4024 : : * deleted or there is a newer version available due to an update, it
4025 : : * should pass in the top level xid of the modifying transaction.
4026 : : *
4027 : : * This function will check for overlap with our own transaction. If the given
4028 : : * xid is also serializable and the transactions overlap (i.e., they cannot see
4029 : : * each other's writes), then we have a conflict out.
4030 : : */
4031 : : void
4032 : 570 : CheckForSerializableConflictOut(Relation relation, TransactionId xid, Snapshot snapshot)
4033 : : {
4034 : : SERIALIZABLEXIDTAG sxidtag;
4035 : : SERIALIZABLEXID *sxid;
4036 : : SERIALIZABLEXACT *sxact;
4037 : :
5387 heikki.linnakangas@i 4038 [ - + ]: 570 : if (!SerializationNeededForRead(relation, snapshot))
5515 4039 : 203 : return;
4040 : :
4041 : : /* Check if someone else has already decided that we need to die */
5387 4042 [ - + ]: 570 : if (SxactIsDoomed(MySerializableXact))
4043 : : {
5515 heikki.linnakangas@i 4044 [ # # ]:UBC 0 : ereport(ERROR,
4045 : : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
4046 : : errmsg("could not serialize access due to read/write dependencies among transactions"),
4047 : : errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict out checking."),
4048 : : errhint("The transaction might succeed if retried.")));
4049 : : }
5515 heikki.linnakangas@i 4050 [ - + ]:CBC 570 : Assert(TransactionIdIsValid(xid));
4051 : :
4052 [ - + ]: 570 : if (TransactionIdEquals(xid, GetTopTransactionIdIfAny()))
5515 heikki.linnakangas@i 4053 :UBC 0 : return;
4054 : :
4055 : : /*
4056 : : * Find sxact or summarized info for the top level xid.
4057 : : */
5515 heikki.linnakangas@i 4058 :CBC 570 : sxidtag.xid = xid;
4059 : 570 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
4060 : : sxid = (SERIALIZABLEXID *)
4061 : 570 : hash_search(SerializableXidHash, &sxidtag, HASH_FIND, NULL);
4062 [ + + ]: 570 : if (!sxid)
4063 : : {
4064 : : /*
4065 : : * Transaction not found in "normal" SSI structures. Check whether it
4066 : : * got pushed out to SLRU storage for "old committed" transactions.
4067 : : */
4068 : : SerCommitSeqNo conflictCommitSeqNo;
4069 : :
2130 tgl@sss.pgh.pa.us 4070 : 24 : conflictCommitSeqNo = SerialGetMinConflictCommitSeqNo(xid);
5515 heikki.linnakangas@i 4071 [ - + ]: 24 : if (conflictCommitSeqNo != 0)
4072 : : {
5515 heikki.linnakangas@i 4073 [ # # ]:UBC 0 : if (conflictCommitSeqNo != InvalidSerCommitSeqNo
4074 [ # # ]: 0 : && (!SxactIsReadOnly(MySerializableXact)
4075 : 0 : || conflictCommitSeqNo
4076 [ # # ]: 0 : <= MySerializableXact->SeqNo.lastCommitBeforeSnapshot))
4077 [ # # ]: 0 : ereport(ERROR,
4078 : : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
4079 : : errmsg("could not serialize access due to read/write dependencies among transactions"),
4080 : : errdetail_internal("Reason code: Canceled on conflict out to old pivot %u.", xid),
4081 : : errhint("The transaction might succeed if retried.")));
4082 : :
4083 [ # # ]: 0 : if (SxactHasSummaryConflictIn(MySerializableXact)
1151 andres@anarazel.de 4084 [ # # ]: 0 : || !dlist_is_empty(&MySerializableXact->inConflicts))
5515 heikki.linnakangas@i 4085 [ # # ]: 0 : ereport(ERROR,
4086 : : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
4087 : : errmsg("could not serialize access due to read/write dependencies among transactions"),
4088 : : errdetail_internal("Reason code: Canceled on identification as a pivot, with conflict out to old committed transaction %u.", xid),
4089 : : errhint("The transaction might succeed if retried.")));
4090 : :
4091 : 0 : MySerializableXact->flags |= SXACT_FLAG_SUMMARY_CONFLICT_OUT;
4092 : : }
4093 : :
4094 : : /* It's not serializable or otherwise not important. */
5515 heikki.linnakangas@i 4095 :CBC 24 : LWLockRelease(SerializableXactHashLock);
4096 : 24 : return;
4097 : : }
4098 : 546 : sxact = sxid->myXact;
4099 [ - + ]: 546 : Assert(TransactionIdEquals(sxact->topXid, xid));
5387 4100 [ + - + + ]: 546 : if (sxact == MySerializableXact || SxactIsDoomed(sxact))
4101 : : {
4102 : : /* Can't conflict with ourself or a transaction that will roll back. */
5515 4103 : 4 : LWLockRelease(SerializableXactHashLock);
4104 : 4 : return;
4105 : : }
4106 : :
4107 : : /*
4108 : : * We have a conflict out to a transaction which has a conflict out to a
4109 : : * summarized transaction. That summarized transaction must have
4110 : : * committed first, and we can't tell when it committed in relation to our
4111 : : * snapshot acquisition, so something needs to be canceled.
4112 : : */
4113 [ - + ]: 542 : if (SxactHasSummaryConflictOut(sxact))
4114 : : {
5515 heikki.linnakangas@i 4115 [ # # ]:UBC 0 : if (!SxactIsPrepared(sxact))
4116 : : {
5387 4117 : 0 : sxact->flags |= SXACT_FLAG_DOOMED;
5515 4118 : 0 : LWLockRelease(SerializableXactHashLock);
4119 : 0 : return;
4120 : : }
4121 : : else
4122 : : {
4123 : 0 : LWLockRelease(SerializableXactHashLock);
4124 [ # # ]: 0 : ereport(ERROR,
4125 : : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
4126 : : errmsg("could not serialize access due to read/write dependencies among transactions"),
4127 : : errdetail_internal("Reason code: Canceled on conflict out to old pivot."),
4128 : : errhint("The transaction might succeed if retried.")));
4129 : : }
4130 : : }
4131 : :
4132 : : /*
4133 : : * If this is a read-only transaction and the writing transaction has
4134 : : * committed, and it doesn't have a rw-conflict to a transaction which
4135 : : * committed before it, no conflict.
4136 : : */
5515 heikki.linnakangas@i 4137 [ + + ]:CBC 542 : if (SxactIsReadOnly(MySerializableXact)
4138 [ + + ]: 119 : && SxactIsCommitted(sxact)
4139 [ + - ]: 8 : && !SxactHasSummaryConflictOut(sxact)
4140 [ + + ]: 8 : && (!SxactHasConflictOut(sxact)
4141 [ - + ]: 2 : || MySerializableXact->SeqNo.lastCommitBeforeSnapshot < sxact->SeqNo.earliestOutConflictCommit))
4142 : : {
4143 : : /* Read-only transaction will appear to run first. No conflict. */
4144 : 6 : LWLockRelease(SerializableXactHashLock);
4145 : 6 : return;
4146 : : }
4147 : :
4148 [ - + ]: 536 : if (!XidIsConcurrent(xid))
4149 : : {
4150 : : /* This write was already in our snapshot; no conflict. */
5515 heikki.linnakangas@i 4151 :UBC 0 : LWLockRelease(SerializableXactHashLock);
4152 : 0 : return;
4153 : : }
4154 : :
5392 heikki.linnakangas@i 4155 [ + + ]:CBC 536 : if (RWConflictExists(MySerializableXact, sxact))
4156 : : {
4157 : : /* We don't want duplicate conflict records in the list. */
5515 4158 : 169 : LWLockRelease(SerializableXactHashLock);
4159 : 169 : return;
4160 : : }
4161 : :
4162 : : /*
4163 : : * Flag the conflict. But first, if this conflict creates a dangerous
4164 : : * structure, ereport an error.
4165 : : */
5392 4166 : 367 : FlagRWConflict(MySerializableXact, sxact);
5515 4167 : 354 : LWLockRelease(SerializableXactHashLock);
4168 : : }
4169 : :
4170 : : /*
4171 : : * Check a particular target for rw-dependency conflict in. A subroutine of
4172 : : * CheckForSerializableConflictIn().
4173 : : */
4174 : : static void
4175 : 7547 : CheckTargetForConflictsIn(PREDICATELOCKTARGETTAG *targettag)
4176 : : {
4177 : : uint32 targettaghash;
4178 : : LWLock *partitionLock;
4179 : : PREDICATELOCKTARGET *target;
5414 rhaas@postgresql.org 4180 : 7547 : PREDICATELOCK *mypredlock = NULL;
4181 : : PREDICATELOCKTAG mypredlocktag;
4182 : : dlist_mutable_iter iter;
4183 : :
5515 heikki.linnakangas@i 4184 [ - + ]: 7547 : Assert(MySerializableXact != InvalidSerializableXact);
4185 : :
4186 : : /*
4187 : : * The same hash and LW lock apply to the lock target and the lock itself.
4188 : : */
4189 : 7547 : targettaghash = PredicateLockTargetTagHashCode(targettag);
4190 : 7547 : partitionLock = PredicateLockHashPartitionLock(targettaghash);
4191 : 7547 : LWLockAcquire(partitionLock, LW_SHARED);
4192 : : target = (PREDICATELOCKTARGET *)
4193 : 7547 : hash_search_with_hash_value(PredicateLockTargetHash,
4194 : : targettag, targettaghash,
4195 : : HASH_FIND, NULL);
4196 [ + + ]: 7547 : if (!target)
4197 : : {
4198 : : /* Nothing has this target locked; we're done here. */
4199 : 5662 : LWLockRelease(partitionLock);
5493 4200 : 5662 : return;
4201 : : }
4202 : :
4203 : : /*
4204 : : * Each lock for an overlapping transaction represents a conflict: a
4205 : : * rw-dependency in to this transaction.
4206 : : */
5515 4207 : 1885 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
4208 : :
1151 andres@anarazel.de 4209 [ + - + + ]: 4250 : dlist_foreach_modify(iter, &target->predicateLocks)
4210 : : {
4211 : 2432 : PREDICATELOCK *predlock =
1031 tgl@sss.pgh.pa.us 4212 : 2432 : dlist_container(PREDICATELOCK, targetLink, iter.cur);
1151 andres@anarazel.de 4213 : 2432 : SERIALIZABLEXACT *sxact = predlock->tag.myXact;
4214 : :
5515 heikki.linnakangas@i 4215 [ + + ]: 2432 : if (sxact == MySerializableXact)
4216 : : {
4217 : : /*
4218 : : * If we're getting a write lock on a tuple, we don't need a
4219 : : * predicate (SIREAD) lock on the same tuple. We can safely remove
4220 : : * our SIREAD lock, but we'll defer doing so until after the loop
4221 : : * because that requires upgrading to an exclusive partition lock.
4222 : : *
4223 : : * We can't use this optimization within a subtransaction because
4224 : : * the subtransaction could roll back, and we would be left
4225 : : * without any lock at the top level.
4226 : : */
5455 rhaas@postgresql.org 4227 [ + - ]: 1574 : if (!IsSubTransaction()
4228 [ + + ]: 1574 : && GET_PREDICATELOCKTARGETTAG_OFFSET(*targettag))
4229 : : {
5414 4230 : 396 : mypredlock = predlock;
4231 : 396 : mypredlocktag = predlock->tag;
4232 : : }
4233 : : }
5387 heikki.linnakangas@i 4234 [ + - ]: 858 : else if (!SxactIsDoomed(sxact)
5515 4235 [ + + ]: 858 : && (!SxactIsCommitted(sxact)
4236 [ + + ]: 89 : || TransactionIdPrecedes(GetTransactionSnapshot()->xmin,
4237 : : sxact->finishedBefore))
5392 4238 [ + + ]: 849 : && !RWConflictExists(sxact, MySerializableXact))
4239 : : {
5515 4240 : 505 : LWLockRelease(SerializableXactHashLock);
4241 : 505 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
4242 : :
4243 : : /*
4244 : : * Re-check after getting exclusive lock because the other
4245 : : * transaction may have flagged a conflict.
4246 : : */
5387 4247 [ + - ]: 505 : if (!SxactIsDoomed(sxact)
5458 rhaas@postgresql.org 4248 [ + + ]: 505 : && (!SxactIsCommitted(sxact)
4249 [ + - ]: 78 : || TransactionIdPrecedes(GetTransactionSnapshot()->xmin,
4250 : : sxact->finishedBefore))
5392 heikki.linnakangas@i 4251 [ + - ]: 505 : && !RWConflictExists(sxact, MySerializableXact))
4252 : : {
4253 : 505 : FlagRWConflict(sxact, MySerializableXact);
4254 : : }
4255 : :
5515 4256 : 438 : LWLockRelease(SerializableXactHashLock);
4257 : 438 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
4258 : : }
4259 : : }
4260 : 1818 : LWLockRelease(SerializableXactHashLock);
4261 : 1818 : LWLockRelease(partitionLock);
4262 : :
4263 : : /*
4264 : : * If we found one of our own SIREAD locks to remove, remove it now.
4265 : : *
4266 : : * At this point our transaction already has a RowExclusiveLock on the
4267 : : * relation, so we are OK to drop the predicate lock on the tuple, if
4268 : : * found, without fearing that another write against the tuple will occur
4269 : : * before the MVCC information makes it to the buffer.
4270 : : */
5414 rhaas@postgresql.org 4271 [ + + ]: 1818 : if (mypredlock != NULL)
4272 : : {
4273 : : uint32 predlockhashcode;
4274 : : PREDICATELOCK *rmpredlock;
4275 : :
2130 tgl@sss.pgh.pa.us 4276 : 389 : LWLockAcquire(SerializablePredicateListLock, LW_SHARED);
2557 tmunro@postgresql.or 4277 [ - + ]: 389 : if (IsInParallelMode())
2130 tgl@sss.pgh.pa.us 4278 :UBC 0 : LWLockAcquire(&MySerializableXact->perXactPredicateListLock, LW_EXCLUSIVE);
5414 rhaas@postgresql.org 4279 :CBC 389 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
4280 : 389 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
4281 : :
4282 : : /*
4283 : : * Remove the predicate lock from shared memory, if it wasn't removed
4284 : : * while the locks were released. One way that could happen is from
4285 : : * autovacuum cleaning up an index.
4286 : : */
4287 : 389 : predlockhashcode = PredicateLockHashCodeFromTargetHashCode
4288 : : (&mypredlocktag, targettaghash);
4289 : : rmpredlock = (PREDICATELOCK *)
4290 : 389 : hash_search_with_hash_value(PredicateLockHash,
4291 : : &mypredlocktag,
4292 : : predlockhashcode,
4293 : : HASH_FIND, NULL);
4294 [ + - ]: 389 : if (rmpredlock != NULL)
4295 : : {
4296 [ - + ]: 389 : Assert(rmpredlock == mypredlock);
4297 : :
1151 andres@anarazel.de 4298 : 389 : dlist_delete(&(mypredlock->targetLink));
4299 : 389 : dlist_delete(&(mypredlock->xactLink));
4300 : :
4301 : : rmpredlock = (PREDICATELOCK *)
5414 rhaas@postgresql.org 4302 : 389 : hash_search_with_hash_value(PredicateLockHash,
4303 : : &mypredlocktag,
4304 : : predlockhashcode,
4305 : : HASH_REMOVE, NULL);
4306 [ - + ]: 389 : Assert(rmpredlock == mypredlock);
4307 : :
4308 : 389 : RemoveTargetIfNoLongerUsed(target, targettaghash);
4309 : : }
4310 : :
4311 : 389 : LWLockRelease(SerializableXactHashLock);
4312 : 389 : LWLockRelease(partitionLock);
2557 tmunro@postgresql.or 4313 [ - + ]: 389 : if (IsInParallelMode())
2130 tgl@sss.pgh.pa.us 4314 :UBC 0 : LWLockRelease(&MySerializableXact->perXactPredicateListLock);
2130 tgl@sss.pgh.pa.us 4315 :CBC 389 : LWLockRelease(SerializablePredicateListLock);
4316 : :
5414 rhaas@postgresql.org 4317 [ + - ]: 389 : if (rmpredlock != NULL)
4318 : : {
4319 : : /*
4320 : : * Remove entry in local lock table if it exists. It's OK if it
4321 : : * doesn't exist; that means the lock was transferred to a new
4322 : : * target by a different backend.
4323 : : */
4324 : 389 : hash_search_with_hash_value(LocalPredicateLockHash,
4325 : : targettag, targettaghash,
4326 : : HASH_REMOVE, NULL);
4327 : :
4328 : 389 : DecrementParentLocks(targettag);
4329 : : }
4330 : : }
4331 : : }
4332 : :
4333 : : /*
4334 : : * CheckForSerializableConflictIn
4335 : : * We are writing the given tuple. If that indicates a rw-conflict
4336 : : * in from another serializable transaction, take appropriate action.
4337 : : *
4338 : : * Skip checking for any granularity for which a parameter is missing.
4339 : : *
4340 : : * A tuple update or delete is in conflict if we have a predicate lock
4341 : : * against the relation or page in which the tuple exists, or against the
4342 : : * tuple itself.
4343 : : */
4344 : : void
136 peter@eisentraut.org 4345 :GNC 18053698 : CheckForSerializableConflictIn(Relation relation, const ItemPointerData *tid, BlockNumber blkno)
4346 : : {
4347 : : PREDICATELOCKTARGETTAG targettag;
4348 : :
5387 heikki.linnakangas@i 4349 [ + + ]:CBC 18053698 : if (!SerializationNeededForWrite(relation))
5515 4350 : 18049230 : return;
4351 : :
4352 : : /* Check if someone else has already decided that we need to die */
5387 4353 [ + + ]: 4468 : if (SxactIsDoomed(MySerializableXact))
5515 4354 [ + - ]: 1 : ereport(ERROR,
4355 : : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
4356 : : errmsg("could not serialize access due to read/write dependencies among transactions"),
4357 : : errdetail_internal("Reason code: Canceled on identification as a pivot, during conflict in checking."),
4358 : : errhint("The transaction might succeed if retried.")));
4359 : :
4360 : : /*
4361 : : * We're doing a write which might cause rw-conflicts now or later.
4362 : : * Memorize that fact.
4363 : : */
5392 4364 : 4467 : MyXactDidWrite = true;
4365 : :
4366 : : /*
4367 : : * It is important that we check for locks from the finest granularity to
4368 : : * the coarsest granularity, so that granularity promotion doesn't cause
4369 : : * us to miss a lock. The new (coarser) lock will be acquired before the
4370 : : * old (finer) locks are released.
4371 : : *
4372 : : * It is not possible to take and hold a lock across the checks for all
4373 : : * granularities because each target could be in a separate partition.
4374 : : */
2238 tmunro@postgresql.or 4375 [ + + ]: 4467 : if (tid != NULL)
4376 : : {
5515 heikki.linnakangas@i 4377 : 650 : SET_PREDICATELOCKTARGETTAG_TUPLE(targettag,
4378 : : relation->rd_locator.dbOid,
4379 : : relation->rd_id,
4380 : : ItemPointerGetBlockNumber(tid),
4381 : : ItemPointerGetOffsetNumber(tid));
4382 : 650 : CheckTargetForConflictsIn(&targettag);
4383 : : }
4384 : :
2238 tmunro@postgresql.or 4385 [ + + ]: 4444 : if (blkno != InvalidBlockNumber)
4386 : : {
5515 heikki.linnakangas@i 4387 : 2483 : SET_PREDICATELOCKTARGETTAG_PAGE(targettag,
4388 : : relation->rd_locator.dbOid,
4389 : : relation->rd_id,
4390 : : blkno);
4391 : 2483 : CheckTargetForConflictsIn(&targettag);
4392 : : }
4393 : :
4394 : 4414 : SET_PREDICATELOCKTARGETTAG_RELATION(targettag,
4395 : : relation->rd_locator.dbOid,
4396 : : relation->rd_id);
4397 : 4414 : CheckTargetForConflictsIn(&targettag);
4398 : : }
4399 : :
4400 : : /*
4401 : : * CheckTableForSerializableConflictIn
4402 : : * The entire table is going through a DDL-style logical mass delete
4403 : : * like TRUNCATE or DROP TABLE. If that causes a rw-conflict in from
4404 : : * another serializable transaction, take appropriate action.
4405 : : *
4406 : : * While these operations do not operate entirely within the bounds of
4407 : : * snapshot isolation, they can occur inside a serializable transaction, and
4408 : : * will logically occur after any reads which saw rows which were destroyed
4409 : : * by these operations, so we do what we can to serialize properly under
4410 : : * SSI.
4411 : : *
4412 : : * The relation passed in must be a heap relation. Any predicate lock of any
4413 : : * granularity on the heap will cause a rw-conflict in to this transaction.
4414 : : * Predicate locks on indexes do not matter because they only exist to guard
4415 : : * against conflicting inserts into the index, and this is a mass *delete*.
4416 : : * When a table is truncated or dropped, the index will also be truncated
4417 : : * or dropped, and we'll deal with locks on the index when that happens.
4418 : : *
4419 : : * Dropping or truncating a table also needs to drop any existing predicate
4420 : : * locks on heap tuples or pages, because they're about to go away. This
4421 : : * should be done before altering the predicate locks because the transaction
4422 : : * could be rolled back because of a conflict, in which case the lock changes
4423 : : * are not needed. (At the moment, we don't actually bother to drop the
4424 : : * existing locks on a dropped or truncated table at the moment. That might
4425 : : * lead to some false positives, but it doesn't seem worth the trouble.)
4426 : : */
4427 : : void
5380 4428 : 27597 : CheckTableForSerializableConflictIn(Relation relation)
4429 : : {
4430 : : HASH_SEQ_STATUS seqstat;
4431 : : PREDICATELOCKTARGET *target;
4432 : : Oid dbId;
4433 : : Oid heapId;
4434 : : int i;
4435 : :
4436 : : /*
4437 : : * Bail out quickly if there are no serializable transactions running.
4438 : : * It's safe to check this without taking locks because the caller is
4439 : : * holding an ACCESS EXCLUSIVE lock on the relation. No new locks which
4440 : : * would matter here can be acquired while that is held.
4441 : : */
5394 4442 [ + + ]: 27597 : if (!TransactionIdIsValid(PredXact->SxactGlobalXmin))
4443 : 27577 : return;
4444 : :
5387 4445 [ + + ]: 154 : if (!SerializationNeededForWrite(relation))
5394 4446 : 134 : return;
4447 : :
4448 : : /*
4449 : : * We're doing a write which might cause rw-conflicts now or later.
4450 : : * Memorize that fact.
4451 : : */
5392 4452 : 20 : MyXactDidWrite = true;
4453 : :
5394 4454 [ - + ]: 20 : Assert(relation->rd_index == NULL); /* not an index relation */
4455 : :
1348 rhaas@postgresql.org 4456 : 20 : dbId = relation->rd_locator.dbOid;
5394 heikki.linnakangas@i 4457 : 20 : heapId = relation->rd_id;
4458 : :
2130 tgl@sss.pgh.pa.us 4459 : 20 : LWLockAcquire(SerializablePredicateListLock, LW_EXCLUSIVE);
5394 heikki.linnakangas@i 4460 [ + + ]: 340 : for (i = 0; i < NUM_PREDICATELOCK_PARTITIONS; i++)
4430 rhaas@postgresql.org 4461 : 320 : LWLockAcquire(PredicateLockHashPartitionLockByIndex(i), LW_SHARED);
3788 kgrittn@postgresql.o 4462 : 20 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
4463 : :
4464 : : /* Scan through target list */
5394 heikki.linnakangas@i 4465 : 20 : hash_seq_init(&seqstat, PredicateLockTargetHash);
4466 : :
4467 [ + + ]: 70 : while ((target = (PREDICATELOCKTARGET *) hash_seq_search(&seqstat)))
4468 : : {
4469 : : dlist_mutable_iter iter;
4470 : :
4471 : : /*
4472 : : * Check whether this is a target which needs attention.
4473 : : */
4474 [ + + ]: 50 : if (GET_PREDICATELOCKTARGETTAG_RELATION(target->tag) != heapId)
4475 : 41 : continue; /* wrong relation id */
5394 heikki.linnakangas@i 4476 [ - + ]:GBC 9 : if (GET_PREDICATELOCKTARGETTAG_DB(target->tag) != dbId)
5394 heikki.linnakangas@i 4477 :UBC 0 : continue; /* wrong database id */
4478 : :
4479 : : /*
4480 : : * Loop through locks for this target and flag conflicts.
4481 : : */
1151 andres@anarazel.de 4482 [ + - + + ]:GBC 18 : dlist_foreach_modify(iter, &target->predicateLocks)
4483 : : {
4484 : 9 : PREDICATELOCK *predlock =
1031 tgl@sss.pgh.pa.us 4485 : 9 : dlist_container(PREDICATELOCK, targetLink, iter.cur);
4486 : :
5394 heikki.linnakangas@i 4487 [ - + ]: 9 : if (predlock->tag.myXact != MySerializableXact
3189 tgl@sss.pgh.pa.us 4488 [ # # ]:UBC 0 : && !RWConflictExists(predlock->tag.myXact, MySerializableXact))
4489 : : {
5392 heikki.linnakangas@i 4490 : 0 : FlagRWConflict(predlock->tag.myXact, MySerializableXact);
4491 : : }
4492 : : }
4493 : : }
4494 : :
4495 : : /* Release locks in reverse order */
5394 heikki.linnakangas@i 4496 :CBC 20 : LWLockRelease(SerializableXactHashLock);
4497 [ + + ]: 340 : for (i = NUM_PREDICATELOCK_PARTITIONS - 1; i >= 0; i--)
4430 rhaas@postgresql.org 4498 : 320 : LWLockRelease(PredicateLockHashPartitionLockByIndex(i));
2130 tgl@sss.pgh.pa.us 4499 : 20 : LWLockRelease(SerializablePredicateListLock);
4500 : : }
4501 : :
4502 : :
4503 : : /*
4504 : : * Flag a rw-dependency between two serializable transactions.
4505 : : *
4506 : : * The caller is responsible for ensuring that we have a LW lock on
4507 : : * the transaction hash table.
4508 : : */
4509 : : static void
5515 heikki.linnakangas@i 4510 : 872 : FlagRWConflict(SERIALIZABLEXACT *reader, SERIALIZABLEXACT *writer)
4511 : : {
4512 [ - + ]: 872 : Assert(reader != writer);
4513 : :
4514 : : /* First, see if this conflict causes failure. */
4515 : 872 : OnConflict_CheckForSerializationFailure(reader, writer);
4516 : :
4517 : : /* Actually do the conflict flagging. */
4518 [ - + ]: 792 : if (reader == OldCommittedSxact)
5515 heikki.linnakangas@i 4519 :UBC 0 : writer->flags |= SXACT_FLAG_SUMMARY_CONFLICT_IN;
5515 heikki.linnakangas@i 4520 [ - + ]:CBC 792 : else if (writer == OldCommittedSxact)
5515 heikki.linnakangas@i 4521 :UBC 0 : reader->flags |= SXACT_FLAG_SUMMARY_CONFLICT_OUT;
4522 : : else
5515 heikki.linnakangas@i 4523 :CBC 792 : SetRWConflict(reader, writer);
4524 : 792 : }
4525 : :
4526 : : /*----------------------------------------------------------------------------
4527 : : * We are about to add a RW-edge to the dependency graph - check that we don't
4528 : : * introduce a dangerous structure by doing so, and abort one of the
4529 : : * transactions if so.
4530 : : *
4531 : : * A serialization failure can only occur if there is a dangerous structure
4532 : : * in the dependency graph:
4533 : : *
4534 : : * Tin ------> Tpivot ------> Tout
4535 : : * rw rw
4536 : : *
4537 : : * Furthermore, Tout must commit first.
4538 : : *
4539 : : * One more optimization is that if Tin is declared READ ONLY (or commits
4540 : : * without writing), we can only have a problem if Tout committed before Tin
4541 : : * acquired its snapshot.
4542 : : *----------------------------------------------------------------------------
4543 : : */
4544 : : static void
5380 tgl@sss.pgh.pa.us 4545 : 872 : OnConflict_CheckForSerializationFailure(const SERIALIZABLEXACT *reader,
4546 : : SERIALIZABLEXACT *writer)
4547 : : {
4548 : : bool failure;
4549 : :
5515 heikki.linnakangas@i 4550 [ - + ]: 872 : Assert(LWLockHeldByMe(SerializableXactHashLock));
4551 : :
4552 : 872 : failure = false;
4553 : :
4554 : : /*------------------------------------------------------------------------
4555 : : * Check for already-committed writer with rw-conflict out flagged
4556 : : * (conflict-flag on W means that T2 committed before W):
4557 : : *
4558 : : * R ------> W ------> T2
4559 : : * rw rw
4560 : : *
4561 : : * That is a dangerous structure, so we must abort. (Since the writer
4562 : : * has already committed, we must be the reader)
4563 : : *------------------------------------------------------------------------
4564 : : */
4565 [ + + ]: 872 : if (SxactIsCommitted(writer)
3189 tgl@sss.pgh.pa.us 4566 [ + + - + ]: 18 : && (SxactHasConflictOut(writer) || SxactHasSummaryConflictOut(writer)))
5515 heikki.linnakangas@i 4567 : 2 : failure = true;
4568 : :
4569 : : /*------------------------------------------------------------------------
4570 : : * Check whether the writer has become a pivot with an out-conflict
4571 : : * committed transaction (T2), and T2 committed first:
4572 : : *
4573 : : * R ------> W ------> T2
4574 : : * rw rw
4575 : : *
4576 : : * Because T2 must've committed first, there is no anomaly if:
4577 : : * - the reader committed before T2
4578 : : * - the writer committed before T2
4579 : : * - the reader is a READ ONLY transaction and the reader was concurrent
4580 : : * with T2 (= reader acquired its snapshot before T2 committed)
4581 : : *
4582 : : * We also handle the case that T2 is prepared but not yet committed
4583 : : * here. In that case T2 has already checked for conflicts, so if it
4584 : : * commits first, making the above conflict real, it's too late for it
4585 : : * to abort.
4586 : : *------------------------------------------------------------------------
4587 : : */
1151 andres@anarazel.de 4588 [ + + - + ]: 872 : if (!failure && SxactHasSummaryConflictOut(writer))
1151 andres@anarazel.de 4589 :UBC 0 : failure = true;
1151 andres@anarazel.de 4590 [ + + ]:CBC 872 : else if (!failure)
4591 : : {
4592 : : dlist_iter iter;
4593 : :
4594 [ + - + + ]: 1087 : dlist_foreach(iter, &writer->outConflicts)
4595 : : {
4596 : 292 : RWConflict conflict =
1031 tgl@sss.pgh.pa.us 4597 :ECB (288) : dlist_container(RWConflictData, outLink, iter.cur);
5403 heikki.linnakangas@i 4598 :CBC 292 : SERIALIZABLEXACT *t2 = conflict->sxactIn;
4599 : :
5365 4600 [ + + ]: 292 : if (SxactIsPrepared(t2)
5403 4601 [ + + ]: 81 : && (!SxactIsCommitted(reader)
5365 4602 [ + - ]: 65 : || t2->prepareSeqNo <= reader->commitSeqNo)
5403 4603 [ - + ]: 81 : && (!SxactIsCommitted(writer)
5365 heikki.linnakangas@i 4604 [ # # ]:UBC 0 : || t2->prepareSeqNo <= writer->commitSeqNo)
5403 heikki.linnakangas@i 4605 [ + + ]:CBC 81 : && (!SxactIsReadOnly(reader)
3189 tgl@sss.pgh.pa.us 4606 [ + + ]: 12 : || t2->prepareSeqNo <= reader->SeqNo.lastCommitBeforeSnapshot))
4607 : : {
5515 heikki.linnakangas@i 4608 : 75 : failure = true;
4609 : 75 : break;
4610 : : }
4611 : : }
4612 : : }
4613 : :
4614 : : /*------------------------------------------------------------------------
4615 : : * Check whether the reader has become a pivot with a writer
4616 : : * that's committed (or prepared):
4617 : : *
4618 : : * T0 ------> R ------> W
4619 : : * rw rw
4620 : : *
4621 : : * Because W must've committed first for an anomaly to occur, there is no
4622 : : * anomaly if:
4623 : : * - T0 committed before the writer
4624 : : * - T0 is READ ONLY, and overlaps the writer
4625 : : *------------------------------------------------------------------------
4626 : : */
5365 4627 [ + + + + : 872 : if (!failure && SxactIsPrepared(writer) && !SxactIsReadOnly(reader))
+ - ]
4628 : : {
5403 4629 [ - + ]: 18 : if (SxactHasSummaryConflictIn(reader))
4630 : : {
5515 heikki.linnakangas@i 4631 :UBC 0 : failure = true;
4632 : : }
4633 : : else
4634 : : {
4635 : : dlist_iter iter;
4636 : :
4637 : : /*
4638 : : * The unconstify is needed as we have no const version of
4639 : : * dlist_foreach().
4640 : : */
1151 andres@anarazel.de 4641 [ + - + + ]:CBC 18 : dlist_foreach(iter, &unconstify(SERIALIZABLEXACT *, reader)->inConflicts)
4642 : : {
4643 : 11 : const RWConflict conflict =
1031 tgl@sss.pgh.pa.us 4644 : 11 : dlist_container(RWConflictData, inLink, iter.cur);
1151 andres@anarazel.de 4645 : 11 : const SERIALIZABLEXACT *t0 = conflict->sxactOut;
4646 : :
4647 [ + - ]: 11 : if (!SxactIsDoomed(t0)
4648 [ + - ]: 11 : && (!SxactIsCommitted(t0)
4649 [ + - ]: 11 : || t0->commitSeqNo >= writer->prepareSeqNo)
4650 [ - + ]: 11 : && (!SxactIsReadOnly(t0)
1151 andres@anarazel.de 4651 [ # # ]:UBC 0 : || t0->SeqNo.lastCommitBeforeSnapshot >= writer->prepareSeqNo))
4652 : : {
1151 andres@anarazel.de 4653 :CBC 11 : failure = true;
4654 : 11 : break;
4655 : : }
4656 : : }
4657 : : }
4658 : : }
4659 : :
5515 heikki.linnakangas@i 4660 [ + + ]: 872 : if (failure)
4661 : : {
4662 : : /*
4663 : : * We have to kill a transaction to avoid a possible anomaly from
4664 : : * occurring. If the writer is us, we can just ereport() to cause a
4665 : : * transaction abort. Otherwise we flag the writer for termination,
4666 : : * causing it to abort when it tries to commit. However, if the writer
4667 : : * is a prepared transaction, already prepared, we can't abort it
4668 : : * anymore, so we have to kill the reader instead.
4669 : : */
4670 [ + + ]: 88 : if (MySerializableXact == writer)
4671 : : {
4672 : 67 : LWLockRelease(SerializableXactHashLock);
4673 [ + - ]: 67 : ereport(ERROR,
4674 : : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
4675 : : errmsg("could not serialize access due to read/write dependencies among transactions"),
4676 : : errdetail_internal("Reason code: Canceled on identification as a pivot, during write."),
4677 : : errhint("The transaction might succeed if retried.")));
4678 : : }
4679 [ + + ]: 21 : else if (SxactIsPrepared(writer))
4680 : : {
4681 : 13 : LWLockRelease(SerializableXactHashLock);
4682 : :
4683 : : /* if we're not the writer, we have to be the reader */
5403 4684 [ - + ]: 13 : Assert(MySerializableXact == reader);
5515 4685 [ + - ]: 13 : ereport(ERROR,
4686 : : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
4687 : : errmsg("could not serialize access due to read/write dependencies among transactions"),
4688 : : errdetail_internal("Reason code: Canceled on conflict out to pivot %u, during read.", writer->topXid),
4689 : : errhint("The transaction might succeed if retried.")));
4690 : : }
5387 4691 : 8 : writer->flags |= SXACT_FLAG_DOOMED;
4692 : : }
5515 4693 : 792 : }
4694 : :
4695 : : /*
4696 : : * PreCommit_CheckForSerializationFailure
4697 : : * Check for dangerous structures in a serializable transaction
4698 : : * at commit.
4699 : : *
4700 : : * We're checking for a dangerous structure as each conflict is recorded.
4701 : : * The only way we could have a problem at commit is if this is the "out"
4702 : : * side of a pivot, and neither the "in" side nor the pivot has yet
4703 : : * committed.
4704 : : *
4705 : : * If a dangerous structure is found, the pivot (the near conflict) is
4706 : : * marked for death, because rolling back another transaction might mean
4707 : : * that we fail without ever making progress. This transaction is
4708 : : * committing writes, so letting it commit ensures progress. If we
4709 : : * canceled the far conflict, it might immediately fail again on retry.
4710 : : */
4711 : : void
4712 : 309758 : PreCommit_CheckForSerializationFailure(void)
4713 : : {
4714 : : dlist_iter near_iter;
4715 : :
4716 [ + + ]: 309758 : if (MySerializableXact == InvalidSerializableXact)
4717 : 308360 : return;
4718 : :
4719 [ - + ]: 1398 : Assert(IsolationIsSerializable());
4720 : :
4721 : 1398 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
4722 : :
4723 : : /*
4724 : : * Check if someone else has already decided that we need to die. Since
4725 : : * we set our own DOOMED flag when partially releasing, ignore in that
4726 : : * case.
4727 : : */
1105 tmunro@postgresql.or 4728 [ + + ]: 1398 : if (SxactIsDoomed(MySerializableXact) &&
4729 [ + + ]: 156 : !SxactIsPartiallyReleased(MySerializableXact))
4730 : : {
5515 heikki.linnakangas@i 4731 : 155 : LWLockRelease(SerializableXactHashLock);
4732 [ + - ]: 155 : ereport(ERROR,
4733 : : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
4734 : : errmsg("could not serialize access due to read/write dependencies among transactions"),
4735 : : errdetail_internal("Reason code: Canceled on identification as a pivot, during commit attempt."),
4736 : : errhint("The transaction might succeed if retried.")));
4737 : : }
4738 : :
1151 andres@anarazel.de 4739 [ + - + + ]: 1849 : dlist_foreach(near_iter, &MySerializableXact->inConflicts)
4740 : : {
4741 : 606 : RWConflict nearConflict =
1031 tgl@sss.pgh.pa.us 4742 : 606 : dlist_container(RWConflictData, inLink, near_iter.cur);
4743 : :
5515 heikki.linnakangas@i 4744 [ + + ]: 606 : if (!SxactIsCommitted(nearConflict->sxactOut)
5387 4745 [ + - ]: 421 : && !SxactIsDoomed(nearConflict->sxactOut))
4746 : : {
4747 : : dlist_iter far_iter;
4748 : :
1151 andres@anarazel.de 4749 [ + - + + ]: 451 : dlist_foreach(far_iter, &nearConflict->sxactOut->inConflicts)
4750 : : {
4751 : 182 : RWConflict farConflict =
1031 tgl@sss.pgh.pa.us 4752 : 182 : dlist_container(RWConflictData, inLink, far_iter.cur);
4753 : :
5515 heikki.linnakangas@i 4754 [ + + ]: 182 : if (farConflict->sxactOut == MySerializableXact
4755 [ + + ]: 42 : || (!SxactIsCommitted(farConflict->sxactOut)
4756 [ + + ]: 24 : && !SxactIsReadOnly(farConflict->sxactOut)
5387 4757 [ + - ]: 12 : && !SxactIsDoomed(farConflict->sxactOut)))
4758 : : {
4759 : : /*
4760 : : * Normally, we kill the pivot transaction to make sure we
4761 : : * make progress if the failing transaction is retried.
4762 : : * However, we can't kill it if it's already prepared, so
4763 : : * in that case we commit suicide instead.
4764 : : */
5381 4765 [ - + ]: 152 : if (SxactIsPrepared(nearConflict->sxactOut))
4766 : : {
5381 heikki.linnakangas@i 4767 :UBC 0 : LWLockRelease(SerializableXactHashLock);
4768 [ # # ]: 0 : ereport(ERROR,
4769 : : (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
4770 : : errmsg("could not serialize access due to read/write dependencies among transactions"),
4771 : : errdetail_internal("Reason code: Canceled on commit attempt with conflict in from prepared pivot."),
4772 : : errhint("The transaction might succeed if retried.")));
4773 : : }
5387 heikki.linnakangas@i 4774 :CBC 152 : nearConflict->sxactOut->flags |= SXACT_FLAG_DOOMED;
5515 4775 : 152 : break;
4776 : : }
4777 : : }
4778 : : }
4779 : : }
4780 : :
5365 4781 : 1243 : MySerializableXact->prepareSeqNo = ++(PredXact->LastSxactCommitSeqNo);
5515 4782 : 1243 : MySerializableXact->flags |= SXACT_FLAG_PREPARED;
4783 : :
4784 : 1243 : LWLockRelease(SerializableXactHashLock);
4785 : : }
4786 : :
4787 : : /*------------------------------------------------------------------------*/
4788 : :
4789 : : /*
4790 : : * Two-phase commit support
4791 : : */
4792 : :
4793 : : /*
4794 : : * AtPrepare_Locks
4795 : : * Do the preparatory work for a PREPARE: make 2PC state file
4796 : : * records for all predicate locks currently held.
4797 : : */
4798 : : void
4799 : 315 : AtPrepare_PredicateLocks(void)
4800 : : {
4801 : : SERIALIZABLEXACT *sxact;
4802 : : TwoPhasePredicateRecord record;
4803 : : TwoPhasePredicateXactRecord *xactRecord;
4804 : : TwoPhasePredicateLockRecord *lockRecord;
4805 : : dlist_iter iter;
4806 : :
5392 4807 : 315 : sxact = MySerializableXact;
5515 4808 : 315 : xactRecord = &(record.data.xactRecord);
4809 : 315 : lockRecord = &(record.data.lockRecord);
4810 : :
4811 [ + + ]: 315 : if (MySerializableXact == InvalidSerializableXact)
4812 : 303 : return;
4813 : :
4814 : : /* Generate an xact record for our SERIALIZABLEXACT */
4815 : 12 : record.type = TWOPHASEPREDICATERECORD_XACT;
4816 : 12 : xactRecord->xmin = MySerializableXact->xmin;
4817 : 12 : xactRecord->flags = MySerializableXact->flags;
4818 : :
4819 : : /*
4820 : : * Note that we don't include the list of conflicts in our out in the
4821 : : * statefile, because new conflicts can be added even after the
4822 : : * transaction prepares. We'll just make a conservative assumption during
4823 : : * recovery instead.
4824 : : */
4825 : :
4826 : 12 : RegisterTwoPhaseRecord(TWOPHASE_RM_PREDICATELOCK_ID, 0,
4827 : : &record, sizeof(record));
4828 : :
4829 : : /*
4830 : : * Generate a lock record for each lock.
4831 : : *
4832 : : * To do this, we need to walk the predicate lock list in our sxact rather
4833 : : * than using the local predicate lock table because the latter is not
4834 : : * guaranteed to be accurate.
4835 : : */
2130 tgl@sss.pgh.pa.us 4836 : 12 : LWLockAcquire(SerializablePredicateListLock, LW_SHARED);
4837 : :
4838 : : /*
4839 : : * No need to take sxact->perXactPredicateListLock in parallel mode
4840 : : * because there cannot be any parallel workers running while we are
4841 : : * preparing a transaction.
4842 : : */
2557 tmunro@postgresql.or 4843 [ + - - + ]: 12 : Assert(!IsParallelWorker() && !ParallelContextActive());
4844 : :
1151 andres@anarazel.de 4845 [ + - + + ]: 22 : dlist_foreach(iter, &sxact->predicateLocks)
4846 : : {
4847 : 10 : PREDICATELOCK *predlock =
1031 tgl@sss.pgh.pa.us 4848 : 10 : dlist_container(PREDICATELOCK, xactLink, iter.cur);
4849 : :
5515 heikki.linnakangas@i 4850 : 10 : record.type = TWOPHASEPREDICATERECORD_LOCK;
4851 : 10 : lockRecord->target = predlock->tag.myTarget->tag;
4852 : :
4853 : 10 : RegisterTwoPhaseRecord(TWOPHASE_RM_PREDICATELOCK_ID, 0,
4854 : : &record, sizeof(record));
4855 : : }
4856 : :
2130 tgl@sss.pgh.pa.us 4857 : 12 : LWLockRelease(SerializablePredicateListLock);
4858 : : }
4859 : :
4860 : : /*
4861 : : * PostPrepare_Locks
4862 : : * Clean up after successful PREPARE. Unlike the non-predicate
4863 : : * lock manager, we do not need to transfer locks to a dummy
4864 : : * PGPROC because our SERIALIZABLEXACT will stay around
4865 : : * anyway. We only need to clean up our local state.
4866 : : */
4867 : : void
251 michael@paquier.xyz 4868 :GNC 315 : PostPrepare_PredicateLocks(FullTransactionId fxid)
4869 : : {
5515 heikki.linnakangas@i 4870 [ + + ]:CBC 315 : if (MySerializableXact == InvalidSerializableXact)
4871 : 303 : return;
4872 : :
4873 [ - + ]: 12 : Assert(SxactIsPrepared(MySerializableXact));
4874 : :
4875 : 12 : MySerializableXact->pid = 0;
742 4876 : 12 : MySerializableXact->pgprocno = INVALID_PROC_NUMBER;
4877 : :
5515 4878 : 12 : hash_destroy(LocalPredicateLockHash);
4879 : 12 : LocalPredicateLockHash = NULL;
4880 : :
4881 : 12 : MySerializableXact = InvalidSerializableXact;
5392 4882 : 12 : MyXactDidWrite = false;
4883 : : }
4884 : :
4885 : : /*
4886 : : * PredicateLockTwoPhaseFinish
4887 : : * Release a prepared transaction's predicate locks once it
4888 : : * commits or aborts.
4889 : : */
4890 : : void
251 michael@paquier.xyz 4891 :GNC 322 : PredicateLockTwoPhaseFinish(FullTransactionId fxid, bool isCommit)
4892 : : {
4893 : : SERIALIZABLEXID *sxid;
4894 : : SERIALIZABLEXIDTAG sxidtag;
4895 : :
4896 : 322 : sxidtag.xid = XidFromFullTransactionId(fxid);
4897 : :
5515 heikki.linnakangas@i 4898 :CBC 322 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
4899 : : sxid = (SERIALIZABLEXID *)
4900 : 322 : hash_search(SerializableXidHash, &sxidtag, HASH_FIND, NULL);
4901 : 322 : LWLockRelease(SerializableXactHashLock);
4902 : :
4903 : : /* xid will not be found if it wasn't a serializable transaction */
4904 [ + + ]: 322 : if (sxid == NULL)
4905 : 310 : return;
4906 : :
4907 : : /* Release its locks */
4908 : 12 : MySerializableXact = sxid->myXact;
5392 4909 : 12 : MyXactDidWrite = true; /* conservatively assume that we wrote
4910 : : * something */
2557 tmunro@postgresql.or 4911 : 12 : ReleasePredicateLocks(isCommit, false);
4912 : : }
4913 : :
4914 : : /*
4915 : : * Re-acquire a predicate lock belonging to a transaction that was prepared.
4916 : : */
4917 : : void
251 michael@paquier.xyz 4918 :UNC 0 : predicatelock_twophase_recover(FullTransactionId fxid, uint16 info,
4919 : : void *recdata, uint32 len)
4920 : : {
4921 : : TwoPhasePredicateRecord *record;
4922 : 0 : TransactionId xid = XidFromFullTransactionId(fxid);
4923 : :
5515 heikki.linnakangas@i 4924 [ # # ]:UBC 0 : Assert(len == sizeof(TwoPhasePredicateRecord));
4925 : :
4926 : 0 : record = (TwoPhasePredicateRecord *) recdata;
4927 : :
4928 [ # # # # ]: 0 : Assert((record->type == TWOPHASEPREDICATERECORD_XACT) ||
4929 : : (record->type == TWOPHASEPREDICATERECORD_LOCK));
4930 : :
4931 [ # # ]: 0 : if (record->type == TWOPHASEPREDICATERECORD_XACT)
4932 : : {
4933 : : /* Per-transaction record. Set up a SERIALIZABLEXACT. */
4934 : : TwoPhasePredicateXactRecord *xactRecord;
4935 : : SERIALIZABLEXACT *sxact;
4936 : : SERIALIZABLEXID *sxid;
4937 : : SERIALIZABLEXIDTAG sxidtag;
4938 : : bool found;
4939 : :
4940 : 0 : xactRecord = (TwoPhasePredicateXactRecord *) &record->data.xactRecord;
4941 : :
4942 : 0 : LWLockAcquire(SerializableXactHashLock, LW_EXCLUSIVE);
4943 : 0 : sxact = CreatePredXact();
4944 [ # # ]: 0 : if (!sxact)
4945 [ # # ]: 0 : ereport(ERROR,
4946 : : (errcode(ERRCODE_OUT_OF_MEMORY),
4947 : : errmsg("out of shared memory")));
4948 : :
4949 : : /* vxid for a prepared xact is INVALID_PROC_NUMBER/xid; no pid */
742 4950 : 0 : sxact->vxid.procNumber = INVALID_PROC_NUMBER;
5515 4951 : 0 : sxact->vxid.localTransactionId = (LocalTransactionId) xid;
4952 : 0 : sxact->pid = 0;
742 4953 : 0 : sxact->pgprocno = INVALID_PROC_NUMBER;
4954 : :
4955 : : /* a prepared xact hasn't committed yet */
5365 4956 : 0 : sxact->prepareSeqNo = RecoverySerCommitSeqNo;
5515 4957 : 0 : sxact->commitSeqNo = InvalidSerCommitSeqNo;
4958 : 0 : sxact->finishedBefore = InvalidTransactionId;
4959 : :
4960 : 0 : sxact->SeqNo.lastCommitBeforeSnapshot = RecoverySerCommitSeqNo;
4961 : :
4962 : : /*
4963 : : * Don't need to track this; no transactions running at the time the
4964 : : * recovered xact started are still active, except possibly other
4965 : : * prepared xacts and we don't care whether those are RO_SAFE or not.
4966 : : */
1151 andres@anarazel.de 4967 : 0 : dlist_init(&(sxact->possibleUnsafeConflicts));
4968 : :
4969 : 0 : dlist_init(&(sxact->predicateLocks));
4970 : 0 : dlist_node_init(&sxact->finishedLink);
4971 : :
5515 heikki.linnakangas@i 4972 : 0 : sxact->topXid = xid;
4973 : 0 : sxact->xmin = xactRecord->xmin;
4974 : 0 : sxact->flags = xactRecord->flags;
4975 [ # # ]: 0 : Assert(SxactIsPrepared(sxact));
4976 [ # # ]: 0 : if (!SxactIsReadOnly(sxact))
4977 : : {
4978 : 0 : ++(PredXact->WritableSxactCount);
4979 [ # # ]: 0 : Assert(PredXact->WritableSxactCount <=
4980 : : (MaxBackends + max_prepared_xacts));
4981 : : }
4982 : :
4983 : : /*
4984 : : * We don't know whether the transaction had any conflicts or not, so
4985 : : * we'll conservatively assume that it had both a conflict in and a
4986 : : * conflict out, and represent that with the summary conflict flags.
4987 : : */
1151 andres@anarazel.de 4988 : 0 : dlist_init(&(sxact->outConflicts));
4989 : 0 : dlist_init(&(sxact->inConflicts));
5128 heikki.linnakangas@i 4990 : 0 : sxact->flags |= SXACT_FLAG_SUMMARY_CONFLICT_IN;
4991 : 0 : sxact->flags |= SXACT_FLAG_SUMMARY_CONFLICT_OUT;
4992 : :
4993 : : /* Register the transaction's xid */
5515 4994 : 0 : sxidtag.xid = xid;
4995 : 0 : sxid = (SERIALIZABLEXID *) hash_search(SerializableXidHash,
4996 : : &sxidtag,
4997 : : HASH_ENTER, &found);
5456 rhaas@postgresql.org 4998 [ # # ]: 0 : Assert(sxid != NULL);
5515 heikki.linnakangas@i 4999 [ # # ]: 0 : Assert(!found);
103 peter@eisentraut.org 5000 :UNC 0 : sxid->myXact = sxact;
5001 : :
5002 : : /*
5003 : : * Update global xmin. Note that this is a special case compared to
5004 : : * registering a normal transaction, because the global xmin might go
5005 : : * backwards. That's OK, because until recovery is over we're not
5006 : : * going to complete any transactions or create any non-prepared
5007 : : * transactions, so there's no danger of throwing away.
5008 : : */
5515 heikki.linnakangas@i 5009 [ # # # # ]:UBC 0 : if ((!TransactionIdIsValid(PredXact->SxactGlobalXmin)) ||
5010 : 0 : (TransactionIdFollows(PredXact->SxactGlobalXmin, sxact->xmin)))
5011 : : {
5012 : 0 : PredXact->SxactGlobalXmin = sxact->xmin;
5013 : 0 : PredXact->SxactGlobalXminCount = 1;
2130 tgl@sss.pgh.pa.us 5014 : 0 : SerialSetActiveSerXmin(sxact->xmin);
5015 : : }
5515 heikki.linnakangas@i 5016 [ # # ]: 0 : else if (TransactionIdEquals(sxact->xmin, PredXact->SxactGlobalXmin))
5017 : : {
5018 [ # # ]: 0 : Assert(PredXact->SxactGlobalXminCount > 0);
5019 : 0 : PredXact->SxactGlobalXminCount++;
5020 : : }
5021 : :
5022 : 0 : LWLockRelease(SerializableXactHashLock);
5023 : : }
5024 [ # # ]: 0 : else if (record->type == TWOPHASEPREDICATERECORD_LOCK)
5025 : : {
5026 : : /* Lock record. Recreate the PREDICATELOCK */
5027 : : TwoPhasePredicateLockRecord *lockRecord;
5028 : : SERIALIZABLEXID *sxid;
5029 : : SERIALIZABLEXACT *sxact;
5030 : : SERIALIZABLEXIDTAG sxidtag;
5031 : : uint32 targettaghash;
5032 : :
5033 : 0 : lockRecord = (TwoPhasePredicateLockRecord *) &record->data.lockRecord;
5034 : 0 : targettaghash = PredicateLockTargetTagHashCode(&lockRecord->target);
5035 : :
5036 : 0 : LWLockAcquire(SerializableXactHashLock, LW_SHARED);
5037 : 0 : sxidtag.xid = xid;
5038 : : sxid = (SERIALIZABLEXID *)
5039 : 0 : hash_search(SerializableXidHash, &sxidtag, HASH_FIND, NULL);
5040 : 0 : LWLockRelease(SerializableXactHashLock);
5041 : :
5042 [ # # ]: 0 : Assert(sxid != NULL);
5043 : 0 : sxact = sxid->myXact;
5044 [ # # ]: 0 : Assert(sxact != InvalidSerializableXact);
5045 : :
5046 : 0 : CreatePredicateLock(&lockRecord->target, targettaghash, sxact);
5047 : : }
5048 : 0 : }
5049 : :
5050 : : /*
5051 : : * Prepare to share the current SERIALIZABLEXACT with parallel workers.
5052 : : * Return a handle object that can be used by AttachSerializableXact() in a
5053 : : * parallel worker.
5054 : : */
5055 : : SerializableXactHandle
2557 tmunro@postgresql.or 5056 :CBC 502 : ShareSerializableXact(void)
5057 : : {
5058 : 502 : return MySerializableXact;
5059 : : }
5060 : :
5061 : : /*
5062 : : * Allow parallel workers to import the leader's SERIALIZABLEXACT.
5063 : : */
5064 : : void
5065 : 1491 : AttachSerializableXact(SerializableXactHandle handle)
5066 : : {
5067 : :
5068 [ - + ]: 1491 : Assert(MySerializableXact == InvalidSerializableXact);
5069 : :
5070 : 1491 : MySerializableXact = (SERIALIZABLEXACT *) handle;
5071 [ + + ]: 1491 : if (MySerializableXact != InvalidSerializableXact)
5072 : 13 : CreateLocalPredicateLockHash();
5073 : 1491 : }
|
