Branch data Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * multixact.c
4 : : * PostgreSQL multi-transaction-log manager
5 : : *
6 : : * The pg_multixact manager is a pg_xact-like manager that stores an array of
7 : : * MultiXactMember for each MultiXactId. It is a fundamental part of the
8 : : * shared-row-lock implementation. Each MultiXactMember is comprised of a
9 : : * TransactionId and a set of flag bits. The name is a bit historical:
10 : : * originally, a MultiXactId consisted of more than one TransactionId (except
11 : : * in rare corner cases), hence "multi". Nowadays, however, it's perfectly
12 : : * legitimate to have MultiXactIds that only include a single Xid.
13 : : *
14 : : * The meaning of the flag bits is opaque to this module, but they are mostly
15 : : * used in heapam.c to identify lock modes that each of the member transactions
16 : : * is holding on any given tuple. This module just contains support to store
17 : : * and retrieve the arrays.
18 : : *
19 : : * We use two SLRU areas, one for storing the offsets at which the data
20 : : * starts for each MultiXactId in the other one. This trick allows us to
21 : : * store variable length arrays of TransactionIds. (We could alternatively
22 : : * use one area containing counts and TransactionIds, with valid MultiXactId
23 : : * values pointing at slots containing counts; but that way seems less robust
24 : : * since it would get completely confused if someone inquired about a bogus
25 : : * MultiXactId that pointed to an intermediate slot containing an XID.)
26 : : *
27 : : * XLOG interactions: this module generates a record whenever a new OFFSETs or
28 : : * MEMBERs page is initialized to zeroes, as well as an
29 : : * XLOG_MULTIXACT_CREATE_ID record whenever a new MultiXactId is defined.
30 : : * This module ignores the WAL rule "write xlog before data," because it
31 : : * suffices that actions recording a MultiXactId in a heap xmax do follow that
32 : : * rule. The only way for the MXID to be referenced from any data page is for
33 : : * heap_lock_tuple() or heap_update() to have put it there, and each generates
34 : : * an XLOG record that must follow ours. The normal LSN interlock between the
35 : : * data page and that XLOG record will ensure that our XLOG record reaches
36 : : * disk first. If the SLRU members/offsets data reaches disk sooner than the
37 : : * XLOG records, we do not care; after recovery, no xmax will refer to it. On
38 : : * the flip side, to ensure that all referenced entries _do_ reach disk, this
39 : : * module's XLOG records completely rebuild the data entered since the last
40 : : * checkpoint. We flush and sync all dirty OFFSETs and MEMBERs pages to disk
41 : : * before each checkpoint is considered complete.
42 : : *
43 : : * Like clog.c, and unlike subtrans.c, we have to preserve state across
44 : : * crashes and ensure that MXID and offset numbering increases monotonically
45 : : * across a crash. We do this in the same way as it's done for transaction
46 : : * IDs: the WAL record is guaranteed to contain evidence of every MXID we
47 : : * could need to worry about, and we just make sure that at the end of
48 : : * replay, the next-MXID and next-offset counters are at least as large as
49 : : * anything we saw during replay.
50 : : *
51 : : * We are able to remove segments no longer necessary by carefully tracking
52 : : * each table's used values: during vacuum, any multixact older than a certain
53 : : * value is removed; the cutoff value is stored in pg_class. The minimum value
54 : : * across all tables in each database is stored in pg_database, and the global
55 : : * minimum across all databases is part of pg_control and is kept in shared
56 : : * memory. Whenever that minimum is advanced, the SLRUs are truncated.
57 : : *
58 : : * When new multixactid values are to be created, care is taken that the
59 : : * counter does not fall within the wraparound horizon considering the global
60 : : * minimum value.
61 : : *
62 : : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
63 : : * Portions Copyright (c) 1994, Regents of the University of California
64 : : *
65 : : * src/backend/access/transam/multixact.c
66 : : *
67 : : *-------------------------------------------------------------------------
68 : : */
69 : : #include "postgres.h"
70 : :
71 : : #include "access/multixact.h"
72 : : #include "access/multixact_internal.h"
73 : : #include "access/slru.h"
74 : : #include "access/twophase.h"
75 : : #include "access/twophase_rmgr.h"
76 : : #include "access/xlog.h"
77 : : #include "access/xloginsert.h"
78 : : #include "access/xlogutils.h"
79 : : #include "miscadmin.h"
80 : : #include "pg_trace.h"
81 : : #include "pgstat.h"
82 : : #include "postmaster/autovacuum.h"
83 : : #include "storage/pmsignal.h"
84 : : #include "storage/proc.h"
85 : : #include "storage/procarray.h"
86 : : #include "utils/guc_hooks.h"
87 : : #include "utils/injection_point.h"
88 : : #include "utils/lsyscache.h"
89 : : #include "utils/memutils.h"
90 : :
91 : :
92 : : /*
93 : : * Thresholds used to keep members disk usage in check when multixids have a
94 : : * lot of members. When MULTIXACT_MEMBER_LOW_THRESHOLD is reached, vacuum
95 : : * starts freezing multixids more aggressively, even if the normal multixid
96 : : * age limits haven't been reached yet.
97 : : */
98 : : #define MULTIXACT_MEMBER_LOW_THRESHOLD UINT64CONST(2000000000)
99 : : #define MULTIXACT_MEMBER_HIGH_THRESHOLD UINT64CONST(4000000000)
100 : :
101 : : static inline MultiXactId
102 : 6 : NextMultiXactId(MultiXactId multi)
103 : : {
104 [ + - ]: 6 : return multi == MaxMultiXactId ? FirstMultiXactId : multi + 1;
105 : : }
106 : :
107 : : static inline MultiXactId
108 : 0 : PreviousMultiXactId(MultiXactId multi)
109 : : {
110 [ # # ]: 0 : return multi == FirstMultiXactId ? MaxMultiXactId : multi - 1;
111 : : }
112 : :
113 : : /*
114 : : * Links to shared-memory data structures for MultiXact control
115 : : */
116 : : static SlruCtlData MultiXactOffsetCtlData;
117 : : static SlruCtlData MultiXactMemberCtlData;
118 : :
119 : : #define MultiXactOffsetCtl (&MultiXactOffsetCtlData)
120 : : #define MultiXactMemberCtl (&MultiXactMemberCtlData)
121 : :
122 : : /*
123 : : * MultiXact state shared across all backends. All this state is protected
124 : : * by MultiXactGenLock. (We also use SLRU bank's lock of MultiXactOffset and
125 : : * MultiXactMember to guard accesses to the two sets of SLRU buffers. For
126 : : * concurrency's sake, we avoid holding more than one of these locks at a
127 : : * time.)
128 : : */
129 : : typedef struct MultiXactStateData
130 : : {
131 : : /* next-to-be-assigned MultiXactId */
132 : : MultiXactId nextMXact;
133 : :
134 : : /* next-to-be-assigned offset */
135 : : MultiXactOffset nextOffset;
136 : :
137 : : /* Have we completed multixact startup? */
138 : : bool finishedStartup;
139 : :
140 : : /*
141 : : * Oldest multixact that is still potentially referenced by a relation.
142 : : * Anything older than this should not be consulted. These values are
143 : : * updated by vacuum.
144 : : */
145 : : MultiXactId oldestMultiXactId;
146 : : Oid oldestMultiXactDB;
147 : :
148 : : /*
149 : : * Oldest multixact offset that is potentially referenced by a multixact
150 : : * referenced by a relation.
151 : : */
152 : : MultiXactOffset oldestOffset;
153 : :
154 : : /* support for anti-wraparound measures */
155 : : MultiXactId multiVacLimit;
156 : : MultiXactId multiWarnLimit;
157 : : MultiXactId multiStopLimit;
158 : : MultiXactId multiWrapLimit;
159 : :
160 : : /*
161 : : * Per-backend data starts here. We have two arrays stored in the area
162 : : * immediately following the MultiXactStateData struct. Each is indexed by
163 : : * ProcNumber.
164 : : *
165 : : * In both arrays, there's a slot for all normal backends
166 : : * (0..MaxBackends-1) followed by a slot for max_prepared_xacts prepared
167 : : * transactions.
168 : : *
169 : : * OldestMemberMXactId[k] is the oldest MultiXactId each backend's current
170 : : * transaction(s) could possibly be a member of, or InvalidMultiXactId
171 : : * when the backend has no live transaction that could possibly be a
172 : : * member of a MultiXact. Each backend sets its entry to the current
173 : : * nextMXact counter just before first acquiring a shared lock in a given
174 : : * transaction, and clears it at transaction end. (This works because only
175 : : * during or after acquiring a shared lock could an XID possibly become a
176 : : * member of a MultiXact, and that MultiXact would have to be created
177 : : * during or after the lock acquisition.)
178 : : *
179 : : * OldestVisibleMXactId[k] is the oldest MultiXactId each backend's
180 : : * current transaction(s) think is potentially live, or InvalidMultiXactId
181 : : * when not in a transaction or not in a transaction that's paid any
182 : : * attention to MultiXacts yet. This is computed when first needed in a
183 : : * given transaction, and cleared at transaction end. We can compute it
184 : : * as the minimum of the valid OldestMemberMXactId[] entries at the time
185 : : * we compute it (using nextMXact if none are valid). Each backend is
186 : : * required not to attempt to access any SLRU data for MultiXactIds older
187 : : * than its own OldestVisibleMXactId[] setting; this is necessary because
188 : : * the relevant SLRU data can be concurrently truncated away.
189 : : *
190 : : * The oldest valid value among all of the OldestMemberMXactId[] and
191 : : * OldestVisibleMXactId[] entries is considered by vacuum as the earliest
192 : : * possible value still having any live member transaction -- OldestMxact.
193 : : * Any value older than that is typically removed from tuple headers, or
194 : : * "frozen" via being replaced with a new xmax. VACUUM can sometimes even
195 : : * remove an individual MultiXact xmax whose value is >= its OldestMxact
196 : : * cutoff, though typically only when no individual member XID is still
197 : : * running. See FreezeMultiXactId for full details.
198 : : *
199 : : * Whenever VACUUM advances relminmxid, then either its OldestMxact cutoff
200 : : * or the oldest extant Multi remaining in the table is used as the new
201 : : * pg_class.relminmxid value (whichever is earlier). The minimum of all
202 : : * relminmxid values in each database is stored in pg_database.datminmxid.
203 : : * In turn, the minimum of all of those values is stored in pg_control.
204 : : * This is used as the truncation point for pg_multixact when unneeded
205 : : * segments get removed by vac_truncate_clog() during vacuuming.
206 : : */
207 : : MultiXactId perBackendXactIds[FLEXIBLE_ARRAY_MEMBER];
208 : : } MultiXactStateData;
209 : :
210 : : /*
211 : : * Size of OldestMemberMXactId and OldestVisibleMXactId arrays.
212 : : */
213 : : #define MaxOldestSlot (MaxBackends + max_prepared_xacts)
214 : :
215 : : /* Pointers to the state data in shared memory */
216 : : static MultiXactStateData *MultiXactState;
217 : : static MultiXactId *OldestMemberMXactId;
218 : : static MultiXactId *OldestVisibleMXactId;
219 : :
220 : :
221 : : /*
222 : : * Definitions for the backend-local MultiXactId cache.
223 : : *
224 : : * We use this cache to store known MultiXacts, so we don't need to go to
225 : : * SLRU areas every time.
226 : : *
227 : : * The cache lasts for the duration of a single transaction, the rationale
228 : : * for this being that most entries will contain our own TransactionId and
229 : : * so they will be uninteresting by the time our next transaction starts.
230 : : * (XXX not clear that this is correct --- other members of the MultiXact
231 : : * could hang around longer than we did. However, it's not clear what a
232 : : * better policy for flushing old cache entries would be.) FIXME actually
233 : : * this is plain wrong now that multixact's may contain update Xids.
234 : : *
235 : : * We allocate the cache entries in a memory context that is deleted at
236 : : * transaction end, so we don't need to do retail freeing of entries.
237 : : */
238 : : typedef struct mXactCacheEnt
239 : : {
240 : : MultiXactId multi;
241 : : int nmembers;
242 : : dlist_node node;
243 : : MultiXactMember members[FLEXIBLE_ARRAY_MEMBER];
244 : : } mXactCacheEnt;
245 : :
246 : : #define MAX_CACHE_ENTRIES 256
247 : : static dclist_head MXactCache = DCLIST_STATIC_INIT(MXactCache);
248 : : static MemoryContext MXactContext = NULL;
249 : :
250 : : #ifdef MULTIXACT_DEBUG
251 : : #define debug_elog2(a,b) elog(a,b)
252 : : #define debug_elog3(a,b,c) elog(a,b,c)
253 : : #define debug_elog4(a,b,c,d) elog(a,b,c,d)
254 : : #define debug_elog5(a,b,c,d,e) elog(a,b,c,d,e)
255 : : #define debug_elog6(a,b,c,d,e,f) elog(a,b,c,d,e,f)
256 : : #else
257 : : #define debug_elog2(a,b)
258 : : #define debug_elog3(a,b,c)
259 : : #define debug_elog4(a,b,c,d)
260 : : #define debug_elog5(a,b,c,d,e)
261 : : #define debug_elog6(a,b,c,d,e,f)
262 : : #endif
263 : :
264 : : /* internal MultiXactId management */
265 : : static void MultiXactIdSetOldestVisible(void);
266 : : static void RecordNewMultiXact(MultiXactId multi, MultiXactOffset offset,
267 : : int nmembers, MultiXactMember *members);
268 : : static MultiXactId GetNewMultiXactId(int nmembers, MultiXactOffset *offset);
269 : :
270 : : /* MultiXact cache management */
271 : : static int mxactMemberComparator(const void *arg1, const void *arg2);
272 : : static MultiXactId mXactCacheGetBySet(int nmembers, MultiXactMember *members);
273 : : static int mXactCacheGetById(MultiXactId multi, MultiXactMember **members);
274 : : static void mXactCachePut(MultiXactId multi, int nmembers,
275 : : MultiXactMember *members);
276 : :
277 : : /* management of SLRU infrastructure */
278 : : static bool MultiXactOffsetPagePrecedes(int64 page1, int64 page2);
279 : : static bool MultiXactMemberPagePrecedes(int64 page1, int64 page2);
280 : : static void ExtendMultiXactOffset(MultiXactId multi);
281 : : static void ExtendMultiXactMember(MultiXactOffset offset, int nmembers);
282 : : static void SetOldestOffset(void);
283 : : static bool find_multixact_start(MultiXactId multi, MultiXactOffset *result);
284 : : static void WriteMTruncateXlogRec(Oid oldestMultiDB,
285 : : MultiXactId endTruncOff,
286 : : MultiXactOffset endTruncMemb);
287 : :
288 : :
289 : : /*
290 : : * MultiXactIdCreate
291 : : * Construct a MultiXactId representing two TransactionIds.
292 : : *
293 : : * The two XIDs must be different, or be requesting different statuses.
294 : : *
295 : : * NB - we don't worry about our local MultiXactId cache here, because that
296 : : * is handled by the lower-level routines.
297 : : */
298 : : MultiXactId
299 : 1 : MultiXactIdCreate(TransactionId xid1, MultiXactStatus status1,
300 : : TransactionId xid2, MultiXactStatus status2)
301 : : {
302 : 1 : MultiXactId newMulti;
303 : 1 : MultiXactMember members[2];
304 : :
305 [ + - ]: 1 : Assert(TransactionIdIsValid(xid1));
306 [ + - ]: 1 : Assert(TransactionIdIsValid(xid2));
307 : :
308 [ - + # # ]: 1 : Assert(!TransactionIdEquals(xid1, xid2) || (status1 != status2));
309 : :
310 : : /* MultiXactIdSetOldestMember() must have been called already. */
311 [ + - ]: 1 : Assert(MultiXactIdIsValid(OldestMemberMXactId[MyProcNumber]));
312 : :
313 : : /*
314 : : * Note: unlike MultiXactIdExpand, we don't bother to check that both XIDs
315 : : * are still running. In typical usage, xid2 will be our own XID and the
316 : : * caller just did a check on xid1, so it'd be wasted effort.
317 : : */
318 : :
319 : 1 : members[0].xid = xid1;
320 : 1 : members[0].status = status1;
321 : 1 : members[1].xid = xid2;
322 : 1 : members[1].status = status2;
323 : :
324 : 1 : newMulti = MultiXactIdCreateFromMembers(2, members);
325 : :
326 : : debug_elog3(DEBUG2, "Create: %s",
327 : : mxid_to_string(newMulti, 2, members));
328 : :
329 : 2 : return newMulti;
330 : 1 : }
331 : :
332 : : /*
333 : : * MultiXactIdExpand
334 : : * Add a TransactionId to a pre-existing MultiXactId.
335 : : *
336 : : * If the TransactionId is already a member of the passed MultiXactId with the
337 : : * same status, just return it as-is.
338 : : *
339 : : * Note that we do NOT actually modify the membership of a pre-existing
340 : : * MultiXactId; instead we create a new one. This is necessary to avoid
341 : : * a race condition against code trying to wait for one MultiXactId to finish;
342 : : * see notes in heapam.c.
343 : : *
344 : : * NB - we don't worry about our local MultiXactId cache here, because that
345 : : * is handled by the lower-level routines.
346 : : *
347 : : * Note: It is critical that MultiXactIds that come from an old cluster (i.e.
348 : : * one upgraded by pg_upgrade from a cluster older than this feature) are not
349 : : * passed in.
350 : : */
351 : : MultiXactId
352 : 1 : MultiXactIdExpand(MultiXactId multi, TransactionId xid, MultiXactStatus status)
353 : : {
354 : 1 : MultiXactId newMulti;
355 : 1 : MultiXactMember *members;
356 : 1 : MultiXactMember *newMembers;
357 : 1 : int nmembers;
358 : 1 : int i;
359 : 1 : int j;
360 : :
361 [ + - ]: 1 : Assert(MultiXactIdIsValid(multi));
362 [ + - ]: 1 : Assert(TransactionIdIsValid(xid));
363 : :
364 : : /* MultiXactIdSetOldestMember() must have been called already. */
365 [ + - ]: 1 : Assert(MultiXactIdIsValid(OldestMemberMXactId[MyProcNumber]));
366 : :
367 : : debug_elog5(DEBUG2, "Expand: received multi %u, xid %u status %s",
368 : : multi, xid, mxstatus_to_string(status));
369 : :
370 : : /*
371 : : * Note: we don't allow for old multis here. The reason is that the only
372 : : * caller of this function does a check that the multixact is no longer
373 : : * running.
374 : : */
375 : 1 : nmembers = GetMultiXactIdMembers(multi, &members, false, false);
376 : :
377 [ + - ]: 1 : if (nmembers < 0)
378 : : {
379 : 0 : MultiXactMember member;
380 : :
381 : : /*
382 : : * The MultiXactId is obsolete. This can only happen if all the
383 : : * MultiXactId members stop running between the caller checking and
384 : : * passing it to us. It would be better to return that fact to the
385 : : * caller, but it would complicate the API and it's unlikely to happen
386 : : * too often, so just deal with it by creating a singleton MultiXact.
387 : : */
388 : 0 : member.xid = xid;
389 : 0 : member.status = status;
390 : 0 : newMulti = MultiXactIdCreateFromMembers(1, &member);
391 : :
392 : : debug_elog4(DEBUG2, "Expand: %u has no members, create singleton %u",
393 : : multi, newMulti);
394 : 0 : return newMulti;
395 : 0 : }
396 : :
397 : : /*
398 : : * If the TransactionId is already a member of the MultiXactId with the
399 : : * same status, just return the existing MultiXactId.
400 : : */
401 [ + + ]: 3 : for (i = 0; i < nmembers; i++)
402 : : {
403 [ - + # # ]: 2 : if (TransactionIdEquals(members[i].xid, xid) &&
404 : 0 : (members[i].status == status))
405 : : {
406 : : debug_elog4(DEBUG2, "Expand: %u is already a member of %u",
407 : : xid, multi);
408 : 0 : pfree(members);
409 : 0 : return multi;
410 : : }
411 : 2 : }
412 : :
413 : : /*
414 : : * Determine which of the members of the MultiXactId are still of
415 : : * interest. This is any running transaction, and also any transaction
416 : : * that grabbed something stronger than just a lock and was committed. (An
417 : : * update that aborted is of no interest here; and having more than one
418 : : * update Xid in a multixact would cause errors elsewhere.)
419 : : *
420 : : * Removing dead members is not just an optimization: freezing of tuples
421 : : * whose Xmax are multis depends on this behavior.
422 : : *
423 : : * Note we have the same race condition here as above: j could be 0 at the
424 : : * end of the loop.
425 : : */
426 : 1 : newMembers = palloc_array(MultiXactMember, nmembers + 1);
427 : :
428 [ + + ]: 3 : for (i = 0, j = 0; i < nmembers; i++)
429 : : {
430 [ + + - + ]: 3 : if (TransactionIdIsInProgress(members[i].xid) ||
431 [ + - ]: 1 : (ISUPDATE_from_mxstatus(members[i].status) &&
432 : 1 : TransactionIdDidCommit(members[i].xid)))
433 : : {
434 : 1 : newMembers[j].xid = members[i].xid;
435 : 1 : newMembers[j++].status = members[i].status;
436 : 1 : }
437 : 2 : }
438 : :
439 : 1 : newMembers[j].xid = xid;
440 : 1 : newMembers[j++].status = status;
441 : 1 : newMulti = MultiXactIdCreateFromMembers(j, newMembers);
442 : :
443 : 1 : pfree(members);
444 : 1 : pfree(newMembers);
445 : :
446 : : debug_elog3(DEBUG2, "Expand: returning new multi %u", newMulti);
447 : :
448 : 1 : return newMulti;
449 : 1 : }
450 : :
451 : : /*
452 : : * MultiXactIdIsRunning
453 : : * Returns whether a MultiXactId is "running".
454 : : *
455 : : * We return true if at least one member of the given MultiXactId is still
456 : : * running. Note that a "false" result is certain not to change,
457 : : * because it is not legal to add members to an existing MultiXactId.
458 : : *
459 : : * Caller is expected to have verified that the multixact does not come from
460 : : * a pg_upgraded share-locked tuple.
461 : : */
462 : : bool
463 : 2 : MultiXactIdIsRunning(MultiXactId multi, bool isLockOnly)
464 : : {
465 : 2 : MultiXactMember *members;
466 : 2 : int nmembers;
467 : 2 : int i;
468 : :
469 : : debug_elog3(DEBUG2, "IsRunning %u?", multi);
470 : :
471 : : /*
472 : : * "false" here means we assume our callers have checked that the given
473 : : * multi cannot possibly come from a pg_upgraded database.
474 : : */
475 : 2 : nmembers = GetMultiXactIdMembers(multi, &members, false, isLockOnly);
476 : :
477 [ - + ]: 2 : if (nmembers <= 0)
478 : : {
479 : : debug_elog2(DEBUG2, "IsRunning: no members");
480 : 0 : return false;
481 : : }
482 : :
483 : : /*
484 : : * Checking for myself is cheap compared to looking in shared memory;
485 : : * return true if any live subtransaction of the current top-level
486 : : * transaction is a member.
487 : : *
488 : : * This is not needed for correctness, it's just a fast path.
489 : : */
490 [ + - ]: 2 : for (i = 0; i < nmembers; i++)
491 : : {
492 [ + - ]: 2 : if (TransactionIdIsCurrentTransactionId(members[i].xid))
493 : : {
494 : : debug_elog3(DEBUG2, "IsRunning: I (%d) am running!", i);
495 : 2 : pfree(members);
496 : 2 : return true;
497 : : }
498 : 0 : }
499 : :
500 : : /*
501 : : * This could be made faster by having another entry point in procarray.c,
502 : : * walking the PGPROC array only once for all the members. But in most
503 : : * cases nmembers should be small enough that it doesn't much matter.
504 : : */
505 [ # # ]: 0 : for (i = 0; i < nmembers; i++)
506 : : {
507 [ # # ]: 0 : if (TransactionIdIsInProgress(members[i].xid))
508 : : {
509 : : debug_elog4(DEBUG2, "IsRunning: member %d (%u) is running",
510 : : i, members[i].xid);
511 : 0 : pfree(members);
512 : 0 : return true;
513 : : }
514 : 0 : }
515 : :
516 : 0 : pfree(members);
517 : :
518 : : debug_elog3(DEBUG2, "IsRunning: %u is not running", multi);
519 : :
520 : 0 : return false;
521 : 2 : }
522 : :
523 : : /*
524 : : * MultiXactIdSetOldestMember
525 : : * Save the oldest MultiXactId this transaction could be a member of.
526 : : *
527 : : * We set the OldestMemberMXactId for a given transaction the first time it's
528 : : * going to do some operation that might require a MultiXactId (tuple lock,
529 : : * update or delete). We need to do this even if we end up using a
530 : : * TransactionId instead of a MultiXactId, because there is a chance that
531 : : * another transaction would add our XID to a MultiXactId.
532 : : *
533 : : * The value to set is the next-to-be-assigned MultiXactId, so this is meant to
534 : : * be called just before doing any such possibly-MultiXactId-able operation.
535 : : */
536 : : void
537 : 327894 : MultiXactIdSetOldestMember(void)
538 : : {
539 [ + + ]: 327894 : if (!MultiXactIdIsValid(OldestMemberMXactId[MyProcNumber]))
540 : : {
541 : 11753 : MultiXactId nextMXact;
542 : :
543 : : /*
544 : : * You might think we don't need to acquire a lock here, since
545 : : * fetching and storing of TransactionIds is probably atomic, but in
546 : : * fact we do: suppose we pick up nextMXact and then lose the CPU for
547 : : * a long time. Someone else could advance nextMXact, and then
548 : : * another someone else could compute an OldestVisibleMXactId that
549 : : * would be after the value we are going to store when we get control
550 : : * back. Which would be wrong.
551 : : *
552 : : * Note that a shared lock is sufficient, because it's enough to stop
553 : : * someone from advancing nextMXact; and nobody else could be trying
554 : : * to write to our OldestMember entry, only reading (and we assume
555 : : * storing it is atomic.)
556 : : */
557 : 11753 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
558 : :
559 : 11753 : nextMXact = MultiXactState->nextMXact;
560 : :
561 : 11753 : OldestMemberMXactId[MyProcNumber] = nextMXact;
562 : :
563 : 11753 : LWLockRelease(MultiXactGenLock);
564 : :
565 : : debug_elog4(DEBUG2, "MultiXact: setting OldestMember[%d] = %u",
566 : : MyProcNumber, nextMXact);
567 : 11753 : }
568 : 327894 : }
569 : :
570 : : /*
571 : : * MultiXactIdSetOldestVisible
572 : : * Save the oldest MultiXactId this transaction considers possibly live.
573 : : *
574 : : * We set the OldestVisibleMXactId for a given transaction the first time
575 : : * it's going to inspect any MultiXactId. Once we have set this, we are
576 : : * guaranteed that SLRU data for MultiXactIds >= our own OldestVisibleMXactId
577 : : * won't be truncated away.
578 : : *
579 : : * The value to set is the oldest of nextMXact and all the valid per-backend
580 : : * OldestMemberMXactId[] entries. Because of the locking we do, we can be
581 : : * certain that no subsequent call to MultiXactIdSetOldestMember can set
582 : : * an OldestMemberMXactId[] entry older than what we compute here. Therefore
583 : : * there is no live transaction, now or later, that can be a member of any
584 : : * MultiXactId older than the OldestVisibleMXactId we compute here.
585 : : */
586 : : static void
587 : 0 : MultiXactIdSetOldestVisible(void)
588 : : {
589 [ # # ]: 0 : if (!MultiXactIdIsValid(OldestVisibleMXactId[MyProcNumber]))
590 : : {
591 : 0 : MultiXactId oldestMXact;
592 : 0 : int i;
593 : :
594 : 0 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
595 : :
596 : 0 : oldestMXact = MultiXactState->nextMXact;
597 [ # # ]: 0 : for (i = 0; i < MaxOldestSlot; i++)
598 : : {
599 : 0 : MultiXactId thisoldest = OldestMemberMXactId[i];
600 : :
601 [ # # # # ]: 0 : if (MultiXactIdIsValid(thisoldest) &&
602 : 0 : MultiXactIdPrecedes(thisoldest, oldestMXact))
603 : 0 : oldestMXact = thisoldest;
604 : 0 : }
605 : :
606 : 0 : OldestVisibleMXactId[MyProcNumber] = oldestMXact;
607 : :
608 : 0 : LWLockRelease(MultiXactGenLock);
609 : :
610 : : debug_elog4(DEBUG2, "MultiXact: setting OldestVisible[%d] = %u",
611 : : MyProcNumber, oldestMXact);
612 : 0 : }
613 : 0 : }
614 : :
615 : : /*
616 : : * ReadNextMultiXactId
617 : : * Return the next MultiXactId to be assigned, but don't allocate it
618 : : */
619 : : MultiXactId
620 : 1932 : ReadNextMultiXactId(void)
621 : : {
622 : 1932 : MultiXactId mxid;
623 : :
624 : : /* XXX we could presumably do this without a lock. */
625 : 1932 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
626 : 1932 : mxid = MultiXactState->nextMXact;
627 : 1932 : LWLockRelease(MultiXactGenLock);
628 : :
629 : 3864 : return mxid;
630 : 1932 : }
631 : :
632 : : /*
633 : : * ReadMultiXactIdRange
634 : : * Get the range of IDs that may still be referenced by a relation.
635 : : */
636 : : void
637 : 0 : ReadMultiXactIdRange(MultiXactId *oldest, MultiXactId *next)
638 : : {
639 : 0 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
640 : 0 : *oldest = MultiXactState->oldestMultiXactId;
641 : 0 : *next = MultiXactState->nextMXact;
642 : 0 : LWLockRelease(MultiXactGenLock);
643 : 0 : }
644 : :
645 : :
646 : : /*
647 : : * MultiXactIdCreateFromMembers
648 : : * Make a new MultiXactId from the specified set of members
649 : : *
650 : : * Make XLOG, SLRU and cache entries for a new MultiXactId, recording the
651 : : * given TransactionIds as members. Returns the newly created MultiXactId.
652 : : *
653 : : * NB: the passed members[] array will be sorted in-place.
654 : : */
655 : : MultiXactId
656 : 2 : MultiXactIdCreateFromMembers(int nmembers, MultiXactMember *members)
657 : : {
658 : 2 : MultiXactId multi;
659 : 2 : MultiXactOffset offset;
660 : 2 : xl_multixact_create xlrec;
661 : :
662 : : debug_elog3(DEBUG2, "Create: %s",
663 : : mxid_to_string(InvalidMultiXactId, nmembers, members));
664 : :
665 : : /*
666 : : * See if the same set of members already exists in our cache; if so, just
667 : : * re-use that MultiXactId. (Note: it might seem that looking in our
668 : : * cache is insufficient, and we ought to search disk to see if a
669 : : * duplicate definition already exists. But since we only ever create
670 : : * MultiXacts containing our own XID, in most cases any such MultiXacts
671 : : * were in fact created by us, and so will be in our cache. There are
672 : : * corner cases where someone else added us to a MultiXact without our
673 : : * knowledge, but it's not worth checking for.)
674 : : */
675 : 2 : multi = mXactCacheGetBySet(nmembers, members);
676 [ - + ]: 2 : if (MultiXactIdIsValid(multi))
677 : : {
678 : : debug_elog2(DEBUG2, "Create: in cache!");
679 : 0 : return multi;
680 : : }
681 : :
682 : : /* Verify that there is a single update Xid among the given members. */
683 : : {
684 : 2 : int i;
685 : 2 : bool has_update = false;
686 : :
687 [ + + ]: 6 : for (i = 0; i < nmembers; i++)
688 : : {
689 [ + + ]: 4 : if (ISUPDATE_from_mxstatus(members[i].status))
690 : : {
691 [ + - ]: 1 : if (has_update)
692 [ # # # # ]: 0 : elog(ERROR, "new multixact has more than one updating member: %s",
693 : : mxid_to_string(InvalidMultiXactId, nmembers, members));
694 : 1 : has_update = true;
695 : 1 : }
696 : 4 : }
697 : 2 : }
698 : :
699 : : /* Load the injection point before entering the critical section */
700 : : INJECTION_POINT_LOAD("multixact-create-from-members");
701 : :
702 : : /*
703 : : * Assign the MXID and offsets range to use, and make sure there is space
704 : : * in the OFFSETs and MEMBERs files. NB: this routine does
705 : : * START_CRIT_SECTION().
706 : : *
707 : : * Note: unlike MultiXactIdCreate and MultiXactIdExpand, we do not check
708 : : * that we've called MultiXactIdSetOldestMember here. This is because
709 : : * this routine is used in some places to create new MultiXactIds of which
710 : : * the current backend is not a member, notably during freezing of multis
711 : : * in vacuum. During vacuum, in particular, it would be unacceptable to
712 : : * keep OldestMulti set, in case it runs for long.
713 : : */
714 : 2 : multi = GetNewMultiXactId(nmembers, &offset);
715 : :
716 : : INJECTION_POINT_CACHED("multixact-create-from-members", NULL);
717 : :
718 : : /* Make an XLOG entry describing the new MXID. */
719 : 2 : xlrec.mid = multi;
720 : 2 : xlrec.moff = offset;
721 : 2 : xlrec.nmembers = nmembers;
722 : :
723 : : /*
724 : : * XXX Note: there's a lot of padding space in MultiXactMember. We could
725 : : * find a more compact representation of this Xlog record -- perhaps all
726 : : * the status flags in one XLogRecData, then all the xids in another one?
727 : : * Not clear that it's worth the trouble though.
728 : : */
729 : 2 : XLogBeginInsert();
730 : 2 : XLogRegisterData(&xlrec, SizeOfMultiXactCreate);
731 : 2 : XLogRegisterData(members, nmembers * sizeof(MultiXactMember));
732 : :
733 : 2 : (void) XLogInsert(RM_MULTIXACT_ID, XLOG_MULTIXACT_CREATE_ID);
734 : :
735 : : /* Now enter the information into the OFFSETs and MEMBERs logs */
736 : 2 : RecordNewMultiXact(multi, offset, nmembers, members);
737 : :
738 : : /* Done with critical section */
739 [ + - ]: 2 : END_CRIT_SECTION();
740 : :
741 : : /* Store the new MultiXactId in the local cache, too */
742 : 2 : mXactCachePut(multi, nmembers, members);
743 : :
744 : : debug_elog2(DEBUG2, "Create: all done");
745 : :
746 : 2 : return multi;
747 : 2 : }
748 : :
749 : : /*
750 : : * RecordNewMultiXact
751 : : * Write info about a new multixact into the offsets and members files
752 : : *
753 : : * This is broken out of MultiXactIdCreateFromMembers so that xlog replay can
754 : : * use it.
755 : : */
756 : : static void
757 : 2 : RecordNewMultiXact(MultiXactId multi, MultiXactOffset offset,
758 : : int nmembers, MultiXactMember *members)
759 : : {
760 : 2 : int64 pageno;
761 : 2 : int64 prev_pageno;
762 : 2 : int entryno;
763 : 2 : int slotno;
764 : 2 : MultiXactOffset *offptr;
765 : 2 : MultiXactId next;
766 : 2 : int64 next_pageno;
767 : 2 : int next_entryno;
768 : 2 : MultiXactOffset *next_offptr;
769 : 2 : MultiXactOffset next_offset;
770 : 2 : LWLock *lock;
771 : 2 : LWLock *prevlock = NULL;
772 : :
773 : : /* position of this multixid in the offsets SLRU area */
774 : 2 : pageno = MultiXactIdToOffsetPage(multi);
775 : 2 : entryno = MultiXactIdToOffsetEntry(multi);
776 : :
777 : : /* position of the next multixid */
778 : 2 : next = NextMultiXactId(multi);
779 : 2 : next_pageno = MultiXactIdToOffsetPage(next);
780 : 2 : next_entryno = MultiXactIdToOffsetEntry(next);
781 : :
782 : : /*
783 : : * Set the starting offset of this multixid's members.
784 : : *
785 : : * In the common case, it was already be set by the previous
786 : : * RecordNewMultiXact call, as this was the next multixid of the previous
787 : : * multixid. But if multiple backends are generating multixids
788 : : * concurrently, we might race ahead and get called before the previous
789 : : * multixid.
790 : : */
791 : 2 : lock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
792 : 2 : LWLockAcquire(lock, LW_EXCLUSIVE);
793 : :
794 : : /*
795 : : * Note: we pass the MultiXactId to SimpleLruReadPage as the "transaction"
796 : : * to complain about if there's any I/O error. This is kinda bogus, but
797 : : * since the errors will always give the full pathname, it should be clear
798 : : * enough that a MultiXactId is really involved. Perhaps someday we'll
799 : : * take the trouble to generalize the slru.c error reporting code.
800 : : */
801 : 2 : slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, multi);
802 : 2 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
803 : 2 : offptr += entryno;
804 : :
805 [ + - ]: 2 : if (*offptr != offset)
806 : : {
807 : : /* should already be set to the correct value, or not at all */
808 [ # # ]: 0 : Assert(*offptr == 0);
809 : 0 : *offptr = offset;
810 : 0 : MultiXactOffsetCtl->shared->page_dirty[slotno] = true;
811 : 0 : }
812 : :
813 : : /*
814 : : * Set the next multixid's offset to the end of this multixid's members.
815 : : */
816 [ + - ]: 2 : if (next_pageno == pageno)
817 : : {
818 : 2 : next_offptr = offptr + 1;
819 : 2 : }
820 : : else
821 : : {
822 : : /* must be the first entry on the page */
823 [ # # # # ]: 0 : Assert(next_entryno == 0 || next == FirstMultiXactId);
824 : :
825 : : /* Swap the lock for a lock on the next page */
826 : 0 : LWLockRelease(lock);
827 : 0 : lock = SimpleLruGetBankLock(MultiXactOffsetCtl, next_pageno);
828 : 0 : LWLockAcquire(lock, LW_EXCLUSIVE);
829 : :
830 : 0 : slotno = SimpleLruReadPage(MultiXactOffsetCtl, next_pageno, true, next);
831 : 0 : next_offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
832 : 0 : next_offptr += next_entryno;
833 : : }
834 : :
835 : : /* Like in GetNewMultiXactId(), skip over offset 0 */
836 : 2 : next_offset = offset + nmembers;
837 [ + - ]: 2 : if (next_offset == 0)
838 : 0 : next_offset = 1;
839 [ - + ]: 2 : if (*next_offptr != next_offset)
840 : : {
841 : : /* should already be set to the correct value, or not at all */
842 [ + - ]: 2 : Assert(*next_offptr == 0);
843 : 2 : *next_offptr = next_offset;
844 : 2 : MultiXactOffsetCtl->shared->page_dirty[slotno] = true;
845 : 2 : }
846 : :
847 : : /* Release MultiXactOffset SLRU lock. */
848 : 2 : LWLockRelease(lock);
849 : :
850 : 2 : prev_pageno = -1;
851 : :
852 [ + + ]: 6 : for (int i = 0; i < nmembers; i++, offset++)
853 : : {
854 : 4 : TransactionId *memberptr;
855 : 4 : uint32 *flagsptr;
856 : 4 : uint32 flagsval;
857 : 4 : int bshift;
858 : 4 : int flagsoff;
859 : 4 : int memberoff;
860 : :
861 [ + - ]: 4 : Assert(members[i].status <= MultiXactStatusUpdate);
862 : :
863 : 4 : pageno = MXOffsetToMemberPage(offset);
864 : 4 : memberoff = MXOffsetToMemberOffset(offset);
865 : 4 : flagsoff = MXOffsetToFlagsOffset(offset);
866 : 4 : bshift = MXOffsetToFlagsBitShift(offset);
867 : :
868 [ + + ]: 4 : if (pageno != prev_pageno)
869 : : {
870 : : /*
871 : : * MultiXactMember SLRU page is changed so check if this new page
872 : : * fall into the different SLRU bank then release the old bank's
873 : : * lock and acquire lock on the new bank.
874 : : */
875 : 2 : lock = SimpleLruGetBankLock(MultiXactMemberCtl, pageno);
876 [ - + ]: 2 : if (lock != prevlock)
877 : : {
878 [ + - ]: 2 : if (prevlock != NULL)
879 : 0 : LWLockRelease(prevlock);
880 : :
881 : 2 : LWLockAcquire(lock, LW_EXCLUSIVE);
882 : 2 : prevlock = lock;
883 : 2 : }
884 : 2 : slotno = SimpleLruReadPage(MultiXactMemberCtl, pageno, true, multi);
885 : 2 : prev_pageno = pageno;
886 : 2 : }
887 : :
888 : 4 : memberptr = (TransactionId *)
889 : 4 : (MultiXactMemberCtl->shared->page_buffer[slotno] + memberoff);
890 : :
891 : 4 : *memberptr = members[i].xid;
892 : :
893 : 4 : flagsptr = (uint32 *)
894 : 4 : (MultiXactMemberCtl->shared->page_buffer[slotno] + flagsoff);
895 : :
896 : 4 : flagsval = *flagsptr;
897 : 4 : flagsval &= ~(((1 << MXACT_MEMBER_BITS_PER_XACT) - 1) << bshift);
898 : 4 : flagsval |= (members[i].status << bshift);
899 : 4 : *flagsptr = flagsval;
900 : :
901 : 4 : MultiXactMemberCtl->shared->page_dirty[slotno] = true;
902 : 4 : }
903 : :
904 [ - + ]: 2 : if (prevlock != NULL)
905 : 2 : LWLockRelease(prevlock);
906 : 2 : }
907 : :
908 : : /*
909 : : * GetNewMultiXactId
910 : : * Get the next MultiXactId.
911 : : *
912 : : * Also, reserve the needed amount of space in the "members" area. The
913 : : * starting offset of the reserved space is returned in *offset.
914 : : *
915 : : * This may generate XLOG records for expansion of the offsets and/or members
916 : : * files. Unfortunately, we have to do that while holding MultiXactGenLock
917 : : * to avoid race conditions --- the XLOG record for zeroing a page must appear
918 : : * before any backend can possibly try to store data in that page!
919 : : *
920 : : * We start a critical section before advancing the shared counters. The
921 : : * caller must end the critical section after writing SLRU data.
922 : : */
923 : : static MultiXactId
924 : 2 : GetNewMultiXactId(int nmembers, MultiXactOffset *offset)
925 : : {
926 : 2 : MultiXactId result;
927 : 2 : MultiXactOffset nextOffset;
928 : :
929 : : debug_elog3(DEBUG2, "GetNew: for %d xids", nmembers);
930 : :
931 : : /* safety check, we should never get this far in a HS standby */
932 [ + - ]: 2 : if (RecoveryInProgress())
933 [ # # # # ]: 0 : elog(ERROR, "cannot assign MultiXactIds during recovery");
934 : :
935 : 2 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
936 : :
937 : : /* Assign the MXID */
938 : 2 : result = MultiXactState->nextMXact;
939 : :
940 : : /*----------
941 : : * Check to see if it's safe to assign another MultiXactId. This protects
942 : : * against catastrophic data loss due to multixact wraparound. The basic
943 : : * rules are:
944 : : *
945 : : * If we're past multiVacLimit or the safe threshold for member storage
946 : : * space, or we don't know what the safe threshold for member storage is,
947 : : * start trying to force autovacuum cycles.
948 : : * If we're past multiWarnLimit, start issuing warnings.
949 : : * If we're past multiStopLimit, refuse to create new MultiXactIds.
950 : : *
951 : : * Note these are pretty much the same protections in GetNewTransactionId.
952 : : *----------
953 : : */
954 [ + - ]: 2 : if (!MultiXactIdPrecedes(result, MultiXactState->multiVacLimit))
955 : : {
956 : : /*
957 : : * For safety's sake, we release MultiXactGenLock while sending
958 : : * signals, warnings, etc. This is not so much because we care about
959 : : * preserving concurrency in this situation, as to avoid any
960 : : * possibility of deadlock while doing get_database_name(). First,
961 : : * copy all the shared values we'll need in this path.
962 : : */
963 : 0 : MultiXactId multiWarnLimit = MultiXactState->multiWarnLimit;
964 : 0 : MultiXactId multiStopLimit = MultiXactState->multiStopLimit;
965 : 0 : MultiXactId multiWrapLimit = MultiXactState->multiWrapLimit;
966 : 0 : Oid oldest_datoid = MultiXactState->oldestMultiXactDB;
967 : :
968 : 0 : LWLockRelease(MultiXactGenLock);
969 : :
970 [ # # # # ]: 0 : if (IsUnderPostmaster &&
971 : 0 : !MultiXactIdPrecedes(result, multiStopLimit))
972 : : {
973 : 0 : char *oldest_datname = get_database_name(oldest_datoid);
974 : :
975 : : /*
976 : : * Immediately kick autovacuum into action as we're already in
977 : : * ERROR territory.
978 : : */
979 : 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
980 : :
981 : : /* complain even if that DB has disappeared */
982 [ # # ]: 0 : if (oldest_datname)
983 [ # # # # ]: 0 : ereport(ERROR,
984 : : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
985 : : errmsg("database is not accepting commands that assign new MultiXactIds to avoid wraparound data loss in database \"%s\"",
986 : : oldest_datname),
987 : : errhint("Execute a database-wide VACUUM in that database.\n"
988 : : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
989 : : else
990 [ # # # # ]: 0 : ereport(ERROR,
991 : : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
992 : : errmsg("database is not accepting commands that assign new MultiXactIds to avoid wraparound data loss in database with OID %u",
993 : : oldest_datoid),
994 : : errhint("Execute a database-wide VACUUM in that database.\n"
995 : : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
996 : 0 : }
997 : :
998 : : /*
999 : : * To avoid swamping the postmaster with signals, we issue the autovac
1000 : : * request only once per 64K multis generated. This still gives
1001 : : * plenty of chances before we get into real trouble.
1002 : : */
1003 [ # # # # : 0 : if (IsUnderPostmaster && ((result % 65536) == 0 || result == FirstMultiXactId))
# # ]
1004 : 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
1005 : :
1006 [ # # ]: 0 : if (!MultiXactIdPrecedes(result, multiWarnLimit))
1007 : : {
1008 : 0 : char *oldest_datname = get_database_name(oldest_datoid);
1009 : :
1010 : : /* complain even if that DB has disappeared */
1011 [ # # ]: 0 : if (oldest_datname)
1012 [ # # # # ]: 0 : ereport(WARNING,
1013 : : (errmsg_plural("database \"%s\" must be vacuumed before %u more MultiXactId is used",
1014 : : "database \"%s\" must be vacuumed before %u more MultiXactIds are used",
1015 : : multiWrapLimit - result,
1016 : : oldest_datname,
1017 : : multiWrapLimit - result),
1018 : : errhint("Execute a database-wide VACUUM in that database.\n"
1019 : : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1020 : : else
1021 [ # # # # ]: 0 : ereport(WARNING,
1022 : : (errmsg_plural("database with OID %u must be vacuumed before %u more MultiXactId is used",
1023 : : "database with OID %u must be vacuumed before %u more MultiXactIds are used",
1024 : : multiWrapLimit - result,
1025 : : oldest_datoid,
1026 : : multiWrapLimit - result),
1027 : : errhint("Execute a database-wide VACUUM in that database.\n"
1028 : : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
1029 : 0 : }
1030 : :
1031 : : /* Re-acquire lock and start over */
1032 : 0 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
1033 : 0 : result = MultiXactState->nextMXact;
1034 : 0 : }
1035 : :
1036 : : /*
1037 : : * Make sure there is room for the next MXID in the file. Assigning this
1038 : : * MXID sets the next MXID's offset already.
1039 : : */
1040 : 2 : ExtendMultiXactOffset(NextMultiXactId(result));
1041 : :
1042 : : /*
1043 : : * Reserve the members space, similarly to above.
1044 : : */
1045 : 2 : nextOffset = MultiXactState->nextOffset;
1046 : :
1047 : : /*
1048 : : * Offsets are 64-bit integers and will never wrap around. Firstly, it
1049 : : * would take an unrealistic amount of time and resources to consume 2^64
1050 : : * offsets. Secondly, multixid creation is WAL-logged, so you would run
1051 : : * out of LSNs before reaching offset wraparound. Nevertheless, check for
1052 : : * wraparound as a sanity check.
1053 : : */
1054 [ + - ]: 2 : if (nextOffset + nmembers < nextOffset)
1055 [ # # # # ]: 0 : ereport(ERROR,
1056 : : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
1057 : : errmsg("MultiXact members would wrap around")));
1058 : 2 : *offset = nextOffset;
1059 : :
1060 : 2 : ExtendMultiXactMember(nextOffset, nmembers);
1061 : :
1062 : : /*
1063 : : * Critical section from here until caller has written the data into the
1064 : : * just-reserved SLRU space; we don't want to error out with a partly
1065 : : * written MultiXact structure. (In particular, failing to write our
1066 : : * start offset after advancing nextMXact would effectively corrupt the
1067 : : * previous MultiXact.)
1068 : : */
1069 : 2 : START_CRIT_SECTION();
1070 : :
1071 : : /*
1072 : : * Advance counters. As in GetNewTransactionId(), this must not happen
1073 : : * until after file extension has succeeded!
1074 : : */
1075 : 2 : MultiXactState->nextMXact = NextMultiXactId(result);
1076 : 2 : MultiXactState->nextOffset += nmembers;
1077 : :
1078 : 2 : LWLockRelease(MultiXactGenLock);
1079 : :
1080 : : debug_elog4(DEBUG2, "GetNew: returning %u offset %" PRIu64,
1081 : : result, *offset);
1082 : 4 : return result;
1083 : 2 : }
1084 : :
1085 : : /*
1086 : : * GetMultiXactIdMembers
1087 : : * Return the set of MultiXactMembers that make up a MultiXactId
1088 : : *
1089 : : * Return value is the number of members found, or -1 if there are none,
1090 : : * and *members is set to a newly palloc'ed array of members. It's the
1091 : : * caller's responsibility to free it when done with it.
1092 : : *
1093 : : * from_pgupgrade must be passed as true if and only if only the multixact
1094 : : * corresponds to a value from a tuple that was locked in a 9.2-or-older
1095 : : * installation and later pg_upgrade'd (that is, the infomask is
1096 : : * HEAP_LOCKED_UPGRADED). In this case, we know for certain that no members
1097 : : * can still be running, so we return -1 just like for an empty multixact
1098 : : * without any further checking. It would be wrong to try to resolve such a
1099 : : * multixact: either the multixact is within the current valid multixact
1100 : : * range, in which case the returned result would be bogus, or outside that
1101 : : * range, in which case an error would be raised.
1102 : : *
1103 : : * In all other cases, the passed multixact must be within the known valid
1104 : : * range, that is, greater than or equal to oldestMultiXactId, and less than
1105 : : * nextMXact. Otherwise, an error is raised.
1106 : : *
1107 : : * isLockOnly must be set to true if caller is certain that the given multi
1108 : : * is used only to lock tuples; can be false without loss of correctness,
1109 : : * but passing a true means we can return quickly without checking for
1110 : : * old updates.
1111 : : */
1112 : : int
1113 : 11 : GetMultiXactIdMembers(MultiXactId multi, MultiXactMember **members,
1114 : : bool from_pgupgrade, bool isLockOnly)
1115 : : {
1116 : 11 : int64 pageno;
1117 : 11 : int64 prev_pageno;
1118 : 11 : int entryno;
1119 : 11 : int slotno;
1120 : 11 : MultiXactOffset *offptr;
1121 : 11 : MultiXactOffset offset;
1122 : 11 : MultiXactOffset nextMXOffset;
1123 : 11 : int length;
1124 : 11 : MultiXactId oldestMXact;
1125 : 11 : MultiXactId nextMXact;
1126 : 11 : MultiXactMember *ptr;
1127 : 11 : LWLock *lock;
1128 : :
1129 : : debug_elog3(DEBUG2, "GetMembers: asked for %u", multi);
1130 : :
1131 [ + - - + ]: 11 : if (!MultiXactIdIsValid(multi) || from_pgupgrade)
1132 : : {
1133 : 0 : *members = NULL;
1134 : 0 : return -1;
1135 : : }
1136 : :
1137 : : /* See if the MultiXactId is in the local cache */
1138 : 11 : length = mXactCacheGetById(multi, members);
1139 [ + - ]: 11 : if (length >= 0)
1140 : : {
1141 : : debug_elog3(DEBUG2, "GetMembers: found %s in the cache",
1142 : : mxid_to_string(multi, length, *members));
1143 : 11 : return length;
1144 : : }
1145 : :
1146 : : /* Set our OldestVisibleMXactId[] entry if we didn't already */
1147 : 0 : MultiXactIdSetOldestVisible();
1148 : :
1149 : : /*
1150 : : * If we know the multi is used only for locking and not for updates, then
1151 : : * we can skip checking if the value is older than our oldest visible
1152 : : * multi. It cannot possibly still be running.
1153 : : */
1154 [ # # # # ]: 0 : if (isLockOnly &&
1155 : 0 : MultiXactIdPrecedes(multi, OldestVisibleMXactId[MyProcNumber]))
1156 : : {
1157 : : debug_elog2(DEBUG2, "GetMembers: a locker-only multi is too old");
1158 : 0 : *members = NULL;
1159 : 0 : return -1;
1160 : : }
1161 : :
1162 : : /*
1163 : : * We check known limits on MultiXact before resorting to the SLRU area.
1164 : : *
1165 : : * An ID older than MultiXactState->oldestMultiXactId cannot possibly be
1166 : : * useful; it has already been removed, or will be removed shortly, by
1167 : : * truncation. If one is passed, an error is raised.
1168 : : *
1169 : : * Also, an ID >= nextMXact shouldn't ever be seen here; if it is seen, it
1170 : : * implies undetected ID wraparound has occurred. This raises a hard
1171 : : * error.
1172 : : *
1173 : : * Shared lock is enough here since we aren't modifying any global state.
1174 : : * Acquire it just long enough to grab the current counter values.
1175 : : */
1176 : 0 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
1177 : :
1178 : 0 : oldestMXact = MultiXactState->oldestMultiXactId;
1179 : 0 : nextMXact = MultiXactState->nextMXact;
1180 : :
1181 : 0 : LWLockRelease(MultiXactGenLock);
1182 : :
1183 [ # # ]: 0 : if (MultiXactIdPrecedes(multi, oldestMXact))
1184 [ # # # # ]: 0 : ereport(ERROR,
1185 : : (errcode(ERRCODE_INTERNAL_ERROR),
1186 : : errmsg("MultiXactId %u does no longer exist -- apparent wraparound",
1187 : : multi)));
1188 : :
1189 [ # # ]: 0 : if (!MultiXactIdPrecedes(multi, nextMXact))
1190 [ # # # # ]: 0 : ereport(ERROR,
1191 : : (errcode(ERRCODE_INTERNAL_ERROR),
1192 : : errmsg("MultiXactId %u has not been created yet -- apparent wraparound",
1193 : : multi)));
1194 : :
1195 : : /*
1196 : : * Find out the offset at which we need to start reading MultiXactMembers
1197 : : * and the number of members in the multixact. We determine the latter as
1198 : : * the difference between this multixact's starting offset and the next
1199 : : * one's.
1200 : : */
1201 : 0 : pageno = MultiXactIdToOffsetPage(multi);
1202 : 0 : entryno = MultiXactIdToOffsetEntry(multi);
1203 : :
1204 : : /* Acquire the bank lock for the page we need. */
1205 : 0 : lock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
1206 : 0 : LWLockAcquire(lock, LW_EXCLUSIVE);
1207 : :
1208 : : /* read this multi's offset */
1209 : 0 : slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, multi);
1210 : 0 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
1211 : 0 : offptr += entryno;
1212 : 0 : offset = *offptr;
1213 : :
1214 [ # # ]: 0 : if (offset == 0)
1215 [ # # # # ]: 0 : ereport(ERROR,
1216 : : (errcode(ERRCODE_DATA_CORRUPTED),
1217 : : errmsg("MultiXact %u has invalid offset", multi)));
1218 : :
1219 : : /* read next multi's offset */
1220 : : {
1221 : 0 : MultiXactId tmpMXact;
1222 : :
1223 : : /* handle wraparound if needed */
1224 : 0 : tmpMXact = NextMultiXactId(multi);
1225 : :
1226 : 0 : prev_pageno = pageno;
1227 : :
1228 : 0 : pageno = MultiXactIdToOffsetPage(tmpMXact);
1229 : 0 : entryno = MultiXactIdToOffsetEntry(tmpMXact);
1230 : :
1231 [ # # ]: 0 : if (pageno != prev_pageno)
1232 : : {
1233 : 0 : LWLock *newlock;
1234 : :
1235 : : /*
1236 : : * Since we're going to access a different SLRU page, if this page
1237 : : * falls under a different bank, release the old bank's lock and
1238 : : * acquire the lock of the new bank.
1239 : : */
1240 : 0 : newlock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
1241 [ # # ]: 0 : if (newlock != lock)
1242 : : {
1243 : 0 : LWLockRelease(lock);
1244 : 0 : LWLockAcquire(newlock, LW_EXCLUSIVE);
1245 : 0 : lock = newlock;
1246 : 0 : }
1247 : 0 : slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, tmpMXact);
1248 : 0 : }
1249 : :
1250 : 0 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
1251 : 0 : offptr += entryno;
1252 : 0 : nextMXOffset = *offptr;
1253 : 0 : }
1254 : :
1255 : 0 : LWLockRelease(lock);
1256 : 0 : lock = NULL;
1257 : :
1258 : : /* Sanity check the next offset */
1259 [ # # ]: 0 : if (nextMXOffset == 0)
1260 [ # # # # ]: 0 : ereport(ERROR,
1261 : : (errcode(ERRCODE_DATA_CORRUPTED),
1262 : : errmsg("MultiXact %u has invalid next offset", multi)));
1263 [ # # ]: 0 : if (nextMXOffset == offset)
1264 [ # # # # ]: 0 : ereport(ERROR,
1265 : : (errcode(ERRCODE_DATA_CORRUPTED),
1266 : : errmsg("MultiXact %u with offset (%" PRIu64 ") has zero members",
1267 : : multi, offset)));
1268 [ # # ]: 0 : if (nextMXOffset < offset)
1269 [ # # # # ]: 0 : ereport(ERROR,
1270 : : (errcode(ERRCODE_DATA_CORRUPTED),
1271 : : errmsg("MultiXact %u has offset (%" PRIu64 ") greater than its next offset (%" PRIu64 ")",
1272 : : multi, offset, nextMXOffset)));
1273 [ # # ]: 0 : if (nextMXOffset - offset > INT32_MAX)
1274 [ # # # # ]: 0 : ereport(ERROR,
1275 : : (errcode(ERRCODE_DATA_CORRUPTED),
1276 : : errmsg("MultiXact %u has too many members (%" PRIu64 ")",
1277 : : multi, nextMXOffset - offset)));
1278 : 0 : length = nextMXOffset - offset;
1279 : :
1280 : : /* read the members */
1281 : 0 : ptr = (MultiXactMember *) palloc(length * sizeof(MultiXactMember));
1282 : 0 : prev_pageno = -1;
1283 [ # # ]: 0 : for (int i = 0; i < length; i++, offset++)
1284 : : {
1285 : 0 : TransactionId *xactptr;
1286 : 0 : uint32 *flagsptr;
1287 : 0 : int flagsoff;
1288 : 0 : int bshift;
1289 : 0 : int memberoff;
1290 : :
1291 : 0 : pageno = MXOffsetToMemberPage(offset);
1292 : 0 : memberoff = MXOffsetToMemberOffset(offset);
1293 : :
1294 [ # # ]: 0 : if (pageno != prev_pageno)
1295 : : {
1296 : 0 : LWLock *newlock;
1297 : :
1298 : : /*
1299 : : * Since we're going to access a different SLRU page, if this page
1300 : : * falls under a different bank, release the old bank's lock and
1301 : : * acquire the lock of the new bank.
1302 : : */
1303 : 0 : newlock = SimpleLruGetBankLock(MultiXactMemberCtl, pageno);
1304 [ # # ]: 0 : if (newlock != lock)
1305 : : {
1306 [ # # ]: 0 : if (lock)
1307 : 0 : LWLockRelease(lock);
1308 : 0 : LWLockAcquire(newlock, LW_EXCLUSIVE);
1309 : 0 : lock = newlock;
1310 : 0 : }
1311 : :
1312 : 0 : slotno = SimpleLruReadPage(MultiXactMemberCtl, pageno, true, multi);
1313 : 0 : prev_pageno = pageno;
1314 : 0 : }
1315 : :
1316 : 0 : xactptr = (TransactionId *)
1317 : 0 : (MultiXactMemberCtl->shared->page_buffer[slotno] + memberoff);
1318 [ # # ]: 0 : Assert(TransactionIdIsValid(*xactptr));
1319 : :
1320 : 0 : flagsoff = MXOffsetToFlagsOffset(offset);
1321 : 0 : bshift = MXOffsetToFlagsBitShift(offset);
1322 : 0 : flagsptr = (uint32 *) (MultiXactMemberCtl->shared->page_buffer[slotno] + flagsoff);
1323 : :
1324 : 0 : ptr[i].xid = *xactptr;
1325 : 0 : ptr[i].status = (*flagsptr >> bshift) & MXACT_MEMBER_XACT_BITMASK;
1326 : 0 : }
1327 : :
1328 : 0 : LWLockRelease(lock);
1329 : :
1330 : : /*
1331 : : * Copy the result into the local cache.
1332 : : */
1333 : 0 : mXactCachePut(multi, length, ptr);
1334 : :
1335 : : debug_elog3(DEBUG2, "GetMembers: no cache for %s",
1336 : : mxid_to_string(multi, length, ptr));
1337 : 0 : *members = ptr;
1338 : 0 : return length;
1339 : 11 : }
1340 : :
1341 : : /*
1342 : : * mxactMemberComparator
1343 : : * qsort comparison function for MultiXactMember
1344 : : *
1345 : : * We can't use wraparound comparison for XIDs because that does not respect
1346 : : * the triangle inequality! Any old sort order will do.
1347 : : */
1348 : : static int
1349 : 4 : mxactMemberComparator(const void *arg1, const void *arg2)
1350 : : {
1351 : 4 : MultiXactMember member1 = *(const MultiXactMember *) arg1;
1352 : 4 : MultiXactMember member2 = *(const MultiXactMember *) arg2;
1353 : :
1354 [ - + ]: 4 : if (member1.xid > member2.xid)
1355 : 0 : return 1;
1356 [ + - ]: 4 : if (member1.xid < member2.xid)
1357 : 4 : return -1;
1358 [ # # ]: 0 : if (member1.status > member2.status)
1359 : 0 : return 1;
1360 [ # # ]: 0 : if (member1.status < member2.status)
1361 : 0 : return -1;
1362 : 0 : return 0;
1363 : 4 : }
1364 : :
1365 : : /*
1366 : : * mXactCacheGetBySet
1367 : : * returns a MultiXactId from the cache based on the set of
1368 : : * TransactionIds that compose it, or InvalidMultiXactId if
1369 : : * none matches.
1370 : : *
1371 : : * This is helpful, for example, if two transactions want to lock a huge
1372 : : * table. By using the cache, the second will use the same MultiXactId
1373 : : * for the majority of tuples, thus keeping MultiXactId usage low (saving
1374 : : * both I/O and wraparound issues).
1375 : : *
1376 : : * NB: the passed members array will be sorted in-place.
1377 : : */
1378 : : static MultiXactId
1379 : 2 : mXactCacheGetBySet(int nmembers, MultiXactMember *members)
1380 : : {
1381 : 2 : dlist_iter iter;
1382 : :
1383 : : debug_elog3(DEBUG2, "CacheGet: looking for %s",
1384 : : mxid_to_string(InvalidMultiXactId, nmembers, members));
1385 : :
1386 : : /* sort the array so comparison is easy */
1387 : 2 : qsort(members, nmembers, sizeof(MultiXactMember), mxactMemberComparator);
1388 : :
1389 [ + - + + ]: 3 : dclist_foreach(iter, &MXactCache)
1390 : : {
1391 : 1 : mXactCacheEnt *entry = dclist_container(mXactCacheEnt, node,
1392 : : iter.cur);
1393 : :
1394 [ - + ]: 1 : if (entry->nmembers != nmembers)
1395 : 0 : continue;
1396 : :
1397 : : /*
1398 : : * We assume the cache entries are sorted, and that the unused bits in
1399 : : * "status" are zeroed.
1400 : : */
1401 [ + - ]: 1 : if (memcmp(members, entry->members, nmembers * sizeof(MultiXactMember)) == 0)
1402 : : {
1403 : : debug_elog3(DEBUG2, "CacheGet: found %u", entry->multi);
1404 : 0 : dclist_move_head(&MXactCache, iter.cur);
1405 : 0 : return entry->multi;
1406 : : }
1407 [ - - + ]: 1 : }
1408 : :
1409 : : debug_elog2(DEBUG2, "CacheGet: not found :-(");
1410 : 2 : return InvalidMultiXactId;
1411 : 2 : }
1412 : :
1413 : : /*
1414 : : * mXactCacheGetById
1415 : : * returns the composing MultiXactMember set from the cache for a
1416 : : * given MultiXactId, if present.
1417 : : *
1418 : : * If successful, *xids is set to the address of a palloc'd copy of the
1419 : : * MultiXactMember set. Return value is number of members, or -1 on failure.
1420 : : */
1421 : : static int
1422 : 11 : mXactCacheGetById(MultiXactId multi, MultiXactMember **members)
1423 : : {
1424 : 11 : dlist_iter iter;
1425 : :
1426 : : debug_elog3(DEBUG2, "CacheGet: looking for %u", multi);
1427 : :
1428 [ + - + - ]: 11 : dclist_foreach(iter, &MXactCache)
1429 : : {
1430 : 11 : mXactCacheEnt *entry = dclist_container(mXactCacheEnt, node,
1431 : : iter.cur);
1432 : :
1433 [ + - ]: 11 : if (entry->multi == multi)
1434 : : {
1435 : 11 : MultiXactMember *ptr;
1436 : 11 : Size size;
1437 : :
1438 : 11 : size = sizeof(MultiXactMember) * entry->nmembers;
1439 : 11 : ptr = (MultiXactMember *) palloc(size);
1440 : :
1441 : 11 : memcpy(ptr, entry->members, size);
1442 : :
1443 : : debug_elog3(DEBUG2, "CacheGet: found %s",
1444 : : mxid_to_string(multi,
1445 : : entry->nmembers,
1446 : : entry->members));
1447 : :
1448 : : /*
1449 : : * Note we modify the list while not using a modifiable iterator.
1450 : : * This is acceptable only because we exit the iteration
1451 : : * immediately afterwards.
1452 : : */
1453 : 11 : dclist_move_head(&MXactCache, iter.cur);
1454 : :
1455 : 11 : *members = ptr;
1456 : 11 : return entry->nmembers;
1457 : 11 : }
1458 [ + - ]: 11 : }
1459 : :
1460 : : debug_elog2(DEBUG2, "CacheGet: not found");
1461 : 0 : return -1;
1462 : 11 : }
1463 : :
1464 : : /*
1465 : : * mXactCachePut
1466 : : * Add a new MultiXactId and its composing set into the local cache.
1467 : : */
1468 : : static void
1469 : 2 : mXactCachePut(MultiXactId multi, int nmembers, MultiXactMember *members)
1470 : : {
1471 : 2 : mXactCacheEnt *entry;
1472 : :
1473 : : debug_elog3(DEBUG2, "CachePut: storing %s",
1474 : : mxid_to_string(multi, nmembers, members));
1475 : :
1476 [ + + ]: 2 : if (MXactContext == NULL)
1477 : : {
1478 : : /* The cache only lives as long as the current transaction */
1479 : : debug_elog2(DEBUG2, "CachePut: initializing memory context");
1480 : 1 : MXactContext = AllocSetContextCreate(TopTransactionContext,
1481 : : "MultiXact cache context",
1482 : : ALLOCSET_SMALL_SIZES);
1483 : 1 : }
1484 : :
1485 : 2 : entry = (mXactCacheEnt *)
1486 : 4 : MemoryContextAlloc(MXactContext,
1487 : 2 : offsetof(mXactCacheEnt, members) +
1488 : 2 : nmembers * sizeof(MultiXactMember));
1489 : :
1490 : 2 : entry->multi = multi;
1491 : 2 : entry->nmembers = nmembers;
1492 : 2 : memcpy(entry->members, members, nmembers * sizeof(MultiXactMember));
1493 : :
1494 : : /* mXactCacheGetBySet assumes the entries are sorted, so sort them */
1495 : 2 : qsort(entry->members, nmembers, sizeof(MultiXactMember), mxactMemberComparator);
1496 : :
1497 : 2 : dclist_push_head(&MXactCache, &entry->node);
1498 [ + - ]: 2 : if (dclist_count(&MXactCache) > MAX_CACHE_ENTRIES)
1499 : : {
1500 : 0 : dlist_node *node;
1501 : :
1502 : 0 : node = dclist_tail_node(&MXactCache);
1503 : 0 : dclist_delete_from(&MXactCache, node);
1504 : :
1505 : 0 : entry = dclist_container(mXactCacheEnt, node, node);
1506 : : debug_elog3(DEBUG2, "CachePut: pruning cached multi %u",
1507 : : entry->multi);
1508 : :
1509 : 0 : pfree(entry);
1510 : 0 : }
1511 : 2 : }
1512 : :
1513 : : char *
1514 : 0 : mxstatus_to_string(MultiXactStatus status)
1515 : : {
1516 [ # # # # : 0 : switch (status)
# # # ]
1517 : : {
1518 : : case MultiXactStatusForKeyShare:
1519 : 0 : return "keysh";
1520 : : case MultiXactStatusForShare:
1521 : 0 : return "sh";
1522 : : case MultiXactStatusForNoKeyUpdate:
1523 : 0 : return "fornokeyupd";
1524 : : case MultiXactStatusForUpdate:
1525 : 0 : return "forupd";
1526 : : case MultiXactStatusNoKeyUpdate:
1527 : 0 : return "nokeyupd";
1528 : : case MultiXactStatusUpdate:
1529 : 0 : return "upd";
1530 : : default:
1531 [ # # # # ]: 0 : elog(ERROR, "unrecognized multixact status %d", status);
1532 : 0 : return "";
1533 : : }
1534 : 0 : }
1535 : :
1536 : : char *
1537 : 0 : mxid_to_string(MultiXactId multi, int nmembers, MultiXactMember *members)
1538 : : {
1539 : : static char *str = NULL;
1540 : 0 : StringInfoData buf;
1541 : 0 : int i;
1542 : :
1543 [ # # ]: 0 : if (str != NULL)
1544 : 0 : pfree(str);
1545 : :
1546 : 0 : initStringInfo(&buf);
1547 : :
1548 : 0 : appendStringInfo(&buf, "%u %d[%u (%s)", multi, nmembers, members[0].xid,
1549 : 0 : mxstatus_to_string(members[0].status));
1550 : :
1551 [ # # ]: 0 : for (i = 1; i < nmembers; i++)
1552 : 0 : appendStringInfo(&buf, ", %u (%s)", members[i].xid,
1553 : 0 : mxstatus_to_string(members[i].status));
1554 : :
1555 : 0 : appendStringInfoChar(&buf, ']');
1556 : 0 : str = MemoryContextStrdup(TopMemoryContext, buf.data);
1557 : 0 : pfree(buf.data);
1558 : 0 : return str;
1559 : 0 : }
1560 : :
1561 : : /*
1562 : : * AtEOXact_MultiXact
1563 : : * Handle transaction end for MultiXact
1564 : : *
1565 : : * This is called at top transaction commit or abort (we don't care which).
1566 : : */
1567 : : void
1568 : 57917 : AtEOXact_MultiXact(void)
1569 : : {
1570 : : /*
1571 : : * Reset our OldestMemberMXactId and OldestVisibleMXactId values, both of
1572 : : * which should only be valid while within a transaction.
1573 : : *
1574 : : * We assume that storing a MultiXactId is atomic and so we need not take
1575 : : * MultiXactGenLock to do this.
1576 : : */
1577 : 57917 : OldestMemberMXactId[MyProcNumber] = InvalidMultiXactId;
1578 : 57917 : OldestVisibleMXactId[MyProcNumber] = InvalidMultiXactId;
1579 : :
1580 : : /*
1581 : : * Discard the local MultiXactId cache. Since MXactContext was created as
1582 : : * a child of TopTransactionContext, we needn't delete it explicitly.
1583 : : */
1584 : 57917 : MXactContext = NULL;
1585 : 57917 : dclist_init(&MXactCache);
1586 : 57917 : }
1587 : :
1588 : : /*
1589 : : * AtPrepare_MultiXact
1590 : : * Save multixact state at 2PC transaction prepare
1591 : : *
1592 : : * In this phase, we only store our OldestMemberMXactId value in the two-phase
1593 : : * state file.
1594 : : */
1595 : : void
1596 : 0 : AtPrepare_MultiXact(void)
1597 : : {
1598 : 0 : MultiXactId myOldestMember = OldestMemberMXactId[MyProcNumber];
1599 : :
1600 [ # # ]: 0 : if (MultiXactIdIsValid(myOldestMember))
1601 : 0 : RegisterTwoPhaseRecord(TWOPHASE_RM_MULTIXACT_ID, 0,
1602 : : &myOldestMember, sizeof(MultiXactId));
1603 : 0 : }
1604 : :
1605 : : /*
1606 : : * PostPrepare_MultiXact
1607 : : * Clean up after successful PREPARE TRANSACTION
1608 : : */
1609 : : void
1610 : 0 : PostPrepare_MultiXact(FullTransactionId fxid)
1611 : : {
1612 : 0 : MultiXactId myOldestMember;
1613 : :
1614 : : /*
1615 : : * Transfer our OldestMemberMXactId value to the slot reserved for the
1616 : : * prepared transaction.
1617 : : */
1618 : 0 : myOldestMember = OldestMemberMXactId[MyProcNumber];
1619 [ # # ]: 0 : if (MultiXactIdIsValid(myOldestMember))
1620 : : {
1621 : 0 : ProcNumber dummyProcNumber = TwoPhaseGetDummyProcNumber(fxid, false);
1622 : :
1623 : : /*
1624 : : * Even though storing MultiXactId is atomic, acquire lock to make
1625 : : * sure others see both changes, not just the reset of the slot of the
1626 : : * current backend. Using a volatile pointer might suffice, but this
1627 : : * isn't a hot spot.
1628 : : */
1629 : 0 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
1630 : :
1631 : 0 : OldestMemberMXactId[dummyProcNumber] = myOldestMember;
1632 : 0 : OldestMemberMXactId[MyProcNumber] = InvalidMultiXactId;
1633 : :
1634 : 0 : LWLockRelease(MultiXactGenLock);
1635 : 0 : }
1636 : :
1637 : : /*
1638 : : * We don't need to transfer OldestVisibleMXactId value, because the
1639 : : * transaction is not going to be looking at any more multixacts once it's
1640 : : * prepared.
1641 : : *
1642 : : * We assume that storing a MultiXactId is atomic and so we need not take
1643 : : * MultiXactGenLock to do this.
1644 : : */
1645 : 0 : OldestVisibleMXactId[MyProcNumber] = InvalidMultiXactId;
1646 : :
1647 : : /*
1648 : : * Discard the local MultiXactId cache like in AtEOXact_MultiXact.
1649 : : */
1650 : 0 : MXactContext = NULL;
1651 : 0 : dclist_init(&MXactCache);
1652 : 0 : }
1653 : :
1654 : : /*
1655 : : * multixact_twophase_recover
1656 : : * Recover the state of a prepared transaction at startup
1657 : : */
1658 : : void
1659 : 0 : multixact_twophase_recover(FullTransactionId fxid, uint16 info,
1660 : : void *recdata, uint32 len)
1661 : : {
1662 : 0 : ProcNumber dummyProcNumber = TwoPhaseGetDummyProcNumber(fxid, false);
1663 : 0 : MultiXactId oldestMember;
1664 : :
1665 : : /*
1666 : : * Get the oldest member XID from the state file record, and set it in the
1667 : : * OldestMemberMXactId slot reserved for this prepared transaction.
1668 : : */
1669 [ # # ]: 0 : Assert(len == sizeof(MultiXactId));
1670 : 0 : oldestMember = *((MultiXactId *) recdata);
1671 : :
1672 : 0 : OldestMemberMXactId[dummyProcNumber] = oldestMember;
1673 : 0 : }
1674 : :
1675 : : /*
1676 : : * multixact_twophase_postcommit
1677 : : * Similar to AtEOXact_MultiXact but for COMMIT PREPARED
1678 : : */
1679 : : void
1680 : 0 : multixact_twophase_postcommit(FullTransactionId fxid, uint16 info,
1681 : : void *recdata, uint32 len)
1682 : : {
1683 : 0 : ProcNumber dummyProcNumber = TwoPhaseGetDummyProcNumber(fxid, true);
1684 : :
1685 [ # # ]: 0 : Assert(len == sizeof(MultiXactId));
1686 : :
1687 : 0 : OldestMemberMXactId[dummyProcNumber] = InvalidMultiXactId;
1688 : 0 : }
1689 : :
1690 : : /*
1691 : : * multixact_twophase_postabort
1692 : : * This is actually just the same as the COMMIT case.
1693 : : */
1694 : : void
1695 : 0 : multixact_twophase_postabort(FullTransactionId fxid, uint16 info,
1696 : : void *recdata, uint32 len)
1697 : : {
1698 : 0 : multixact_twophase_postcommit(fxid, info, recdata, len);
1699 : 0 : }
1700 : :
1701 : : /*
1702 : : * Initialization of shared memory for MultiXact. We use two SLRU areas,
1703 : : * thus double memory. Also, reserve space for the shared MultiXactState
1704 : : * struct and the per-backend MultiXactId arrays (two of those, too).
1705 : : */
1706 : : Size
1707 : 9 : MultiXactShmemSize(void)
1708 : : {
1709 : 9 : Size size;
1710 : :
1711 : : /* We need 2*MaxOldestSlot perBackendXactIds[] entries */
1712 : : #define SHARED_MULTIXACT_STATE_SIZE \
1713 : : add_size(offsetof(MultiXactStateData, perBackendXactIds), \
1714 : : mul_size(sizeof(MultiXactId) * 2, MaxOldestSlot))
1715 : :
1716 : 9 : size = SHARED_MULTIXACT_STATE_SIZE;
1717 : 9 : size = add_size(size, SimpleLruShmemSize(multixact_offset_buffers, 0));
1718 : 9 : size = add_size(size, SimpleLruShmemSize(multixact_member_buffers, 0));
1719 : :
1720 : 18 : return size;
1721 : 9 : }
1722 : :
1723 : : void
1724 : 6 : MultiXactShmemInit(void)
1725 : : {
1726 : 6 : bool found;
1727 : :
1728 : : debug_elog2(DEBUG2, "Shared Memory Init for MultiXact");
1729 : :
1730 : 6 : MultiXactOffsetCtl->PagePrecedes = MultiXactOffsetPagePrecedes;
1731 : 6 : MultiXactMemberCtl->PagePrecedes = MultiXactMemberPagePrecedes;
1732 : :
1733 : 6 : SimpleLruInit(MultiXactOffsetCtl,
1734 : 6 : "multixact_offset", multixact_offset_buffers, 0,
1735 : : "pg_multixact/offsets", LWTRANCHE_MULTIXACTOFFSET_BUFFER,
1736 : : LWTRANCHE_MULTIXACTOFFSET_SLRU,
1737 : : SYNC_HANDLER_MULTIXACT_OFFSET,
1738 : : false);
1739 : 6 : SlruPagePrecedesUnitTests(MultiXactOffsetCtl, MULTIXACT_OFFSETS_PER_PAGE);
1740 : 6 : SimpleLruInit(MultiXactMemberCtl,
1741 : 6 : "multixact_member", multixact_member_buffers, 0,
1742 : : "pg_multixact/members", LWTRANCHE_MULTIXACTMEMBER_BUFFER,
1743 : : LWTRANCHE_MULTIXACTMEMBER_SLRU,
1744 : : SYNC_HANDLER_MULTIXACT_MEMBER,
1745 : : true);
1746 : : /* doesn't call SimpleLruTruncate() or meet criteria for unit tests */
1747 : :
1748 : : /* Initialize our shared state struct */
1749 : 6 : MultiXactState = ShmemInitStruct("Shared MultiXact State",
1750 : 6 : SHARED_MULTIXACT_STATE_SIZE,
1751 : : &found);
1752 [ + - ]: 6 : if (!IsUnderPostmaster)
1753 : : {
1754 [ + - ]: 6 : Assert(!found);
1755 : :
1756 : : /* Make sure we zero out the per-backend state */
1757 [ + - + - : 6 : MemSet(MultiXactState, 0, SHARED_MULTIXACT_STATE_SIZE);
+ - + - #
# ]
1758 : 6 : }
1759 : : else
1760 [ # # ]: 0 : Assert(found);
1761 : :
1762 : : /*
1763 : : * Set up array pointers.
1764 : : */
1765 : 6 : OldestMemberMXactId = MultiXactState->perBackendXactIds;
1766 : 6 : OldestVisibleMXactId = OldestMemberMXactId + MaxOldestSlot;
1767 : 6 : }
1768 : :
1769 : : /*
1770 : : * GUC check_hook for multixact_offset_buffers
1771 : : */
1772 : : bool
1773 : 6 : check_multixact_offset_buffers(int *newval, void **extra, GucSource source)
1774 : : {
1775 : 6 : return check_slru_buffers("multixact_offset_buffers", newval);
1776 : : }
1777 : :
1778 : : /*
1779 : : * GUC check_hook for multixact_member_buffers
1780 : : */
1781 : : bool
1782 : 6 : check_multixact_member_buffers(int *newval, void **extra, GucSource source)
1783 : : {
1784 : 6 : return check_slru_buffers("multixact_member_buffers", newval);
1785 : : }
1786 : :
1787 : : /*
1788 : : * This func must be called ONCE on system install. It creates the initial
1789 : : * MultiXact segments. (The MultiXacts directories are assumed to have been
1790 : : * created by initdb, and MultiXactShmemInit must have been called already.)
1791 : : */
1792 : : void
1793 : 1 : BootStrapMultiXact(void)
1794 : : {
1795 : : /* Zero the initial pages and flush them to disk */
1796 : 1 : SimpleLruZeroAndWritePage(MultiXactOffsetCtl, 0);
1797 : 1 : SimpleLruZeroAndWritePage(MultiXactMemberCtl, 0);
1798 : 1 : }
1799 : :
1800 : : /*
1801 : : * This must be called ONCE during postmaster or standalone-backend startup.
1802 : : *
1803 : : * StartupXLOG has already established nextMXact/nextOffset by calling
1804 : : * MultiXactSetNextMXact and/or MultiXactAdvanceNextMXact, and the oldestMulti
1805 : : * info from pg_control and/or MultiXactAdvanceOldest, but we haven't yet
1806 : : * replayed WAL.
1807 : : */
1808 : : void
1809 : 4 : StartupMultiXact(void)
1810 : : {
1811 : 4 : MultiXactId multi = MultiXactState->nextMXact;
1812 : 4 : MultiXactOffset offset = MultiXactState->nextOffset;
1813 : 4 : int64 pageno;
1814 : :
1815 : : /*
1816 : : * Initialize offset's idea of the latest page number.
1817 : : */
1818 : 4 : pageno = MultiXactIdToOffsetPage(multi);
1819 : 8 : pg_atomic_write_u64(&MultiXactOffsetCtl->shared->latest_page_number,
1820 : 4 : pageno);
1821 : :
1822 : : /*
1823 : : * Initialize member's idea of the latest page number.
1824 : : */
1825 : 4 : pageno = MXOffsetToMemberPage(offset);
1826 : 8 : pg_atomic_write_u64(&MultiXactMemberCtl->shared->latest_page_number,
1827 : 4 : pageno);
1828 : 4 : }
1829 : :
1830 : : /*
1831 : : * This must be called ONCE at the end of startup/recovery.
1832 : : */
1833 : : void
1834 : 4 : TrimMultiXact(void)
1835 : : {
1836 : 4 : MultiXactId nextMXact;
1837 : 4 : MultiXactOffset offset;
1838 : 4 : MultiXactId oldestMXact;
1839 : 4 : Oid oldestMXactDB;
1840 : 4 : int64 pageno;
1841 : 4 : int entryno;
1842 : 4 : int flagsoff;
1843 : :
1844 : 4 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
1845 : 4 : nextMXact = MultiXactState->nextMXact;
1846 : 4 : offset = MultiXactState->nextOffset;
1847 : 4 : oldestMXact = MultiXactState->oldestMultiXactId;
1848 : 4 : oldestMXactDB = MultiXactState->oldestMultiXactDB;
1849 : 4 : LWLockRelease(MultiXactGenLock);
1850 : :
1851 : : /* Clean up offsets state */
1852 : :
1853 : : /*
1854 : : * (Re-)Initialize our idea of the latest page number for offsets.
1855 : : */
1856 : 4 : pageno = MultiXactIdToOffsetPage(nextMXact);
1857 : 8 : pg_atomic_write_u64(&MultiXactOffsetCtl->shared->latest_page_number,
1858 : 4 : pageno);
1859 : :
1860 : : /*
1861 : : * Set the offset of nextMXact on the offsets page. This is normally done
1862 : : * in RecordNewMultiXact() of the previous multixact, but let's be sure
1863 : : * the next page exists, if the nextMXact was reset with pg_resetwal for
1864 : : * example.
1865 : : *
1866 : : * Zero out the remainder of the page. See notes in TrimCLOG() for
1867 : : * background. Unlike CLOG, some WAL record covers every pg_multixact
1868 : : * SLRU mutation. Since, also unlike CLOG, we ignore the WAL rule "write
1869 : : * xlog before data," nextMXact successors may carry obsolete, nonzero
1870 : : * offset values.
1871 : : */
1872 : 4 : entryno = MultiXactIdToOffsetEntry(nextMXact);
1873 : : {
1874 : 4 : int slotno;
1875 : 4 : MultiXactOffset *offptr;
1876 : 4 : LWLock *lock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
1877 : :
1878 : 4 : LWLockAcquire(lock, LW_EXCLUSIVE);
1879 [ + - + - ]: 4 : if (entryno == 0 || nextMXact == FirstMultiXactId)
1880 : 4 : slotno = SimpleLruZeroPage(MultiXactOffsetCtl, pageno);
1881 : : else
1882 : 0 : slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, nextMXact);
1883 : 4 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
1884 : 4 : offptr += entryno;
1885 : :
1886 : 4 : *offptr = offset;
1887 [ + - - + ]: 4 : if (entryno != 0 && (entryno + 1) * sizeof(MultiXactOffset) != BLCKSZ)
1888 [ + - + - : 4 : MemSet(offptr + 1, 0, BLCKSZ - (entryno + 1) * sizeof(MultiXactOffset));
+ - + - #
# ]
1889 : :
1890 : 4 : MultiXactOffsetCtl->shared->page_dirty[slotno] = true;
1891 : 4 : LWLockRelease(lock);
1892 : 4 : }
1893 : :
1894 : : /*
1895 : : * And the same for members.
1896 : : *
1897 : : * (Re-)Initialize our idea of the latest page number for members.
1898 : : */
1899 : 4 : pageno = MXOffsetToMemberPage(offset);
1900 : 8 : pg_atomic_write_u64(&MultiXactMemberCtl->shared->latest_page_number,
1901 : 4 : pageno);
1902 : :
1903 : : /*
1904 : : * Zero out the remainder of the current members page. See notes in
1905 : : * TrimCLOG() for motivation.
1906 : : */
1907 : 4 : flagsoff = MXOffsetToFlagsOffset(offset);
1908 [ + - ]: 4 : if (flagsoff != 0)
1909 : : {
1910 : 0 : int slotno;
1911 : 0 : TransactionId *xidptr;
1912 : 0 : int memberoff;
1913 : 0 : LWLock *lock = SimpleLruGetBankLock(MultiXactMemberCtl, pageno);
1914 : :
1915 : 0 : LWLockAcquire(lock, LW_EXCLUSIVE);
1916 : 0 : memberoff = MXOffsetToMemberOffset(offset);
1917 : 0 : slotno = SimpleLruReadPage(MultiXactMemberCtl, pageno, true, offset);
1918 : 0 : xidptr = (TransactionId *)
1919 : 0 : (MultiXactMemberCtl->shared->page_buffer[slotno] + memberoff);
1920 : :
1921 [ # # # # : 0 : MemSet(xidptr, 0, BLCKSZ - memberoff);
# # # # #
# ]
1922 : :
1923 : : /*
1924 : : * Note: we don't need to zero out the flag bits in the remaining
1925 : : * members of the current group, because they are always reset before
1926 : : * writing.
1927 : : */
1928 : :
1929 : 0 : MultiXactMemberCtl->shared->page_dirty[slotno] = true;
1930 : 0 : LWLockRelease(lock);
1931 : 0 : }
1932 : :
1933 : : /* signal that we're officially up */
1934 : 4 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
1935 : 4 : MultiXactState->finishedStartup = true;
1936 : 4 : LWLockRelease(MultiXactGenLock);
1937 : :
1938 : : /* Now compute how far away the next multixid wraparound is. */
1939 : 4 : SetMultiXactIdLimit(oldestMXact, oldestMXactDB);
1940 : 4 : }
1941 : :
1942 : : /*
1943 : : * Get the MultiXact data to save in a checkpoint record
1944 : : */
1945 : : void
1946 : 7 : MultiXactGetCheckptMulti(bool is_shutdown,
1947 : : MultiXactId *nextMulti,
1948 : : MultiXactOffset *nextMultiOffset,
1949 : : MultiXactId *oldestMulti,
1950 : : Oid *oldestMultiDB)
1951 : : {
1952 : 7 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
1953 : 7 : *nextMulti = MultiXactState->nextMXact;
1954 : 7 : *nextMultiOffset = MultiXactState->nextOffset;
1955 : 7 : *oldestMulti = MultiXactState->oldestMultiXactId;
1956 : 7 : *oldestMultiDB = MultiXactState->oldestMultiXactDB;
1957 : 7 : LWLockRelease(MultiXactGenLock);
1958 : :
1959 : : debug_elog6(DEBUG2,
1960 : : "MultiXact: checkpoint is nextMulti %u, nextOffset %" PRIu64 ", oldestMulti %u in DB %u",
1961 : : *nextMulti, *nextMultiOffset, *oldestMulti, *oldestMultiDB);
1962 : 7 : }
1963 : :
1964 : : /*
1965 : : * Perform a checkpoint --- either during shutdown, or on-the-fly
1966 : : */
1967 : : void
1968 : 7 : CheckPointMultiXact(void)
1969 : : {
1970 : 7 : TRACE_POSTGRESQL_MULTIXACT_CHECKPOINT_START(true);
1971 : :
1972 : : /*
1973 : : * Write dirty MultiXact pages to disk. This may result in sync requests
1974 : : * queued for later handling by ProcessSyncRequests(), as part of the
1975 : : * checkpoint.
1976 : : */
1977 : 7 : SimpleLruWriteAll(MultiXactOffsetCtl, true);
1978 : 7 : SimpleLruWriteAll(MultiXactMemberCtl, true);
1979 : :
1980 : 7 : TRACE_POSTGRESQL_MULTIXACT_CHECKPOINT_DONE(true);
1981 : 7 : }
1982 : :
1983 : : /*
1984 : : * Set the next-to-be-assigned MultiXactId and offset
1985 : : *
1986 : : * This is used when we can determine the correct next ID/offset exactly
1987 : : * from a checkpoint record. Although this is only called during bootstrap
1988 : : * and XLog replay, we take the lock in case any hot-standby backends are
1989 : : * examining the values.
1990 : : */
1991 : : void
1992 : 5 : MultiXactSetNextMXact(MultiXactId nextMulti,
1993 : : MultiXactOffset nextMultiOffset)
1994 : : {
1995 [ + - ]: 5 : Assert(MultiXactIdIsValid(nextMulti));
1996 : : debug_elog4(DEBUG2, "MultiXact: setting next multi to %u offset %" PRIu64,
1997 : : nextMulti, nextMultiOffset);
1998 : :
1999 : 5 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2000 : 5 : MultiXactState->nextMXact = nextMulti;
2001 : 5 : MultiXactState->nextOffset = nextMultiOffset;
2002 : 5 : LWLockRelease(MultiXactGenLock);
2003 : 5 : }
2004 : :
2005 : : /*
2006 : : * Determine the last safe MultiXactId to allocate given the currently oldest
2007 : : * datminmxid (ie, the oldest MultiXactId that might exist in any database
2008 : : * of our cluster), and the OID of the (or a) database with that value.
2009 : : *
2010 : : * This also updates MultiXactState->oldestOffset, by looking up the offset of
2011 : : * MultiXactState->oldestMultiXactId.
2012 : : */
2013 : : void
2014 : 11 : SetMultiXactIdLimit(MultiXactId oldest_datminmxid, Oid oldest_datoid)
2015 : : {
2016 : 11 : MultiXactId multiVacLimit;
2017 : 11 : MultiXactId multiWarnLimit;
2018 : 11 : MultiXactId multiStopLimit;
2019 : 11 : MultiXactId multiWrapLimit;
2020 : 11 : MultiXactId curMulti;
2021 : :
2022 [ + - ]: 11 : Assert(MultiXactIdIsValid(oldest_datminmxid));
2023 : :
2024 : : /*
2025 : : * We pretend that a wrap will happen halfway through the multixact ID
2026 : : * space, but that's not really true, because multixacts wrap differently
2027 : : * from transaction IDs.
2028 : : */
2029 : 11 : multiWrapLimit = oldest_datminmxid + (MaxMultiXactId >> 1);
2030 [ + - ]: 11 : if (multiWrapLimit < FirstMultiXactId)
2031 : 0 : multiWrapLimit += FirstMultiXactId;
2032 : :
2033 : : /*
2034 : : * We'll refuse to continue assigning MultiXactIds once we get within 3M
2035 : : * multi of data loss. See SetTransactionIdLimit.
2036 : : */
2037 : 11 : multiStopLimit = multiWrapLimit - 3000000;
2038 [ + - ]: 11 : if (multiStopLimit < FirstMultiXactId)
2039 : 0 : multiStopLimit -= FirstMultiXactId;
2040 : :
2041 : : /*
2042 : : * We'll start complaining loudly when we get within 40M multis of data
2043 : : * loss. This is kind of arbitrary, but if you let your gas gauge get
2044 : : * down to 2% of full, would you be looking for the next gas station? We
2045 : : * need to be fairly liberal about this number because there are lots of
2046 : : * scenarios where most transactions are done by automatic clients that
2047 : : * won't pay attention to warnings. (No, we're not gonna make this
2048 : : * configurable. If you know enough to configure it, you know enough to
2049 : : * not get in this kind of trouble in the first place.)
2050 : : */
2051 : 11 : multiWarnLimit = multiWrapLimit - 40000000;
2052 [ + - ]: 11 : if (multiWarnLimit < FirstMultiXactId)
2053 : 0 : multiWarnLimit -= FirstMultiXactId;
2054 : :
2055 : : /*
2056 : : * We'll start trying to force autovacuums when oldest_datminmxid gets to
2057 : : * be more than autovacuum_multixact_freeze_max_age mxids old.
2058 : : *
2059 : : * Note: autovacuum_multixact_freeze_max_age is a PGC_POSTMASTER parameter
2060 : : * so that we don't have to worry about dealing with on-the-fly changes in
2061 : : * its value. See SetTransactionIdLimit.
2062 : : */
2063 : 11 : multiVacLimit = oldest_datminmxid + autovacuum_multixact_freeze_max_age;
2064 [ + - ]: 11 : if (multiVacLimit < FirstMultiXactId)
2065 : 0 : multiVacLimit += FirstMultiXactId;
2066 : :
2067 : : /* Grab lock for just long enough to set the new limit values */
2068 : 11 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2069 : 11 : MultiXactState->oldestMultiXactId = oldest_datminmxid;
2070 : 11 : MultiXactState->oldestMultiXactDB = oldest_datoid;
2071 : 11 : MultiXactState->multiVacLimit = multiVacLimit;
2072 : 11 : MultiXactState->multiWarnLimit = multiWarnLimit;
2073 : 11 : MultiXactState->multiStopLimit = multiStopLimit;
2074 : 11 : MultiXactState->multiWrapLimit = multiWrapLimit;
2075 : 11 : curMulti = MultiXactState->nextMXact;
2076 : 11 : LWLockRelease(MultiXactGenLock);
2077 : :
2078 : : /* Log the info */
2079 [ - + - + ]: 11 : ereport(DEBUG1,
2080 : : (errmsg_internal("MultiXactId wrap limit is %u, limited by database with OID %u",
2081 : : multiWrapLimit, oldest_datoid)));
2082 : :
2083 : : /*
2084 : : * Computing the actual limits is only possible once the data directory is
2085 : : * in a consistent state. There's no need to compute the limits while
2086 : : * still replaying WAL - no decisions about new multis are made even
2087 : : * though multixact creations might be replayed. So we'll only do further
2088 : : * checks after TrimMultiXact() has been called.
2089 : : */
2090 [ + + ]: 11 : if (!MultiXactState->finishedStartup)
2091 : 5 : return;
2092 : :
2093 [ + - ]: 6 : Assert(!InRecovery);
2094 : :
2095 : : /*
2096 : : * Offsets are 64-bits wide and never wrap around, so we don't need to
2097 : : * consider them for emergency autovacuum purposes. But now that we're in
2098 : : * a consistent state, determine MultiXactState->oldestOffset. It will be
2099 : : * used to adjust the freezing cutoff, to keep the offsets disk usage in
2100 : : * check.
2101 : : */
2102 : 6 : SetOldestOffset();
2103 : :
2104 : : /*
2105 : : * If past the autovacuum force point, immediately signal an autovac
2106 : : * request. The reason for this is that autovac only processes one
2107 : : * database per invocation. Once it's finished cleaning up the oldest
2108 : : * database, it'll call here, and we'll signal the postmaster to start
2109 : : * another iteration immediately if there are still any old databases.
2110 : : */
2111 [ - + # # ]: 6 : if (MultiXactIdPrecedes(multiVacLimit, curMulti) && IsUnderPostmaster)
2112 : 0 : SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
2113 : :
2114 : : /* Give an immediate warning if past the wrap warn point */
2115 [ + - ]: 6 : if (MultiXactIdPrecedes(multiWarnLimit, curMulti))
2116 : : {
2117 : 0 : char *oldest_datname;
2118 : :
2119 : : /*
2120 : : * We can be called when not inside a transaction, for example during
2121 : : * StartupXLOG(). In such a case we cannot do database access, so we
2122 : : * must just report the oldest DB's OID.
2123 : : *
2124 : : * Note: it's also possible that get_database_name fails and returns
2125 : : * NULL, for example because the database just got dropped. We'll
2126 : : * still warn, even though the warning might now be unnecessary.
2127 : : */
2128 [ # # ]: 0 : if (IsTransactionState())
2129 : 0 : oldest_datname = get_database_name(oldest_datoid);
2130 : : else
2131 : 0 : oldest_datname = NULL;
2132 : :
2133 [ # # ]: 0 : if (oldest_datname)
2134 [ # # # # ]: 0 : ereport(WARNING,
2135 : : (errmsg_plural("database \"%s\" must be vacuumed before %u more MultiXactId is used",
2136 : : "database \"%s\" must be vacuumed before %u more MultiXactIds are used",
2137 : : multiWrapLimit - curMulti,
2138 : : oldest_datname,
2139 : : multiWrapLimit - curMulti),
2140 : : errhint("To avoid MultiXactId assignment failures, execute a database-wide VACUUM in that database.\n"
2141 : : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
2142 : : else
2143 [ # # # # ]: 0 : ereport(WARNING,
2144 : : (errmsg_plural("database with OID %u must be vacuumed before %u more MultiXactId is used",
2145 : : "database with OID %u must be vacuumed before %u more MultiXactIds are used",
2146 : : multiWrapLimit - curMulti,
2147 : : oldest_datoid,
2148 : : multiWrapLimit - curMulti),
2149 : : errhint("To avoid MultiXactId assignment failures, execute a database-wide VACUUM in that database.\n"
2150 : : "You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
2151 : 0 : }
2152 [ - + ]: 11 : }
2153 : :
2154 : : /*
2155 : : * Ensure the next-to-be-assigned MultiXactId is at least minMulti,
2156 : : * and similarly nextOffset is at least minMultiOffset.
2157 : : *
2158 : : * This is used when we can determine minimum safe values from an XLog
2159 : : * record (either an on-line checkpoint or an mxact creation log entry).
2160 : : * Although this is only called during XLog replay, we take the lock in case
2161 : : * any hot-standby backends are examining the values.
2162 : : */
2163 : : void
2164 : 0 : MultiXactAdvanceNextMXact(MultiXactId minMulti,
2165 : : MultiXactOffset minMultiOffset)
2166 : : {
2167 [ # # ]: 0 : Assert(MultiXactIdIsValid(minMulti));
2168 : :
2169 : 0 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2170 [ # # ]: 0 : if (MultiXactIdPrecedes(MultiXactState->nextMXact, minMulti))
2171 : : {
2172 : : debug_elog3(DEBUG2, "MultiXact: setting next multi to %u", minMulti);
2173 : 0 : MultiXactState->nextMXact = minMulti;
2174 : 0 : }
2175 [ # # ]: 0 : if (MultiXactState->nextOffset < minMultiOffset)
2176 : : {
2177 : : debug_elog3(DEBUG2, "MultiXact: setting next offset to %" PRIu64,
2178 : : minMultiOffset);
2179 : 0 : MultiXactState->nextOffset = minMultiOffset;
2180 : 0 : }
2181 : 0 : LWLockRelease(MultiXactGenLock);
2182 : 0 : }
2183 : :
2184 : : /*
2185 : : * Update our oldestMultiXactId value, but only if it's more recent than what
2186 : : * we had.
2187 : : *
2188 : : * This may only be called during WAL replay.
2189 : : */
2190 : : void
2191 : 0 : MultiXactAdvanceOldest(MultiXactId oldestMulti, Oid oldestMultiDB)
2192 : : {
2193 [ # # ]: 0 : Assert(InRecovery);
2194 : :
2195 [ # # ]: 0 : if (MultiXactIdPrecedes(MultiXactState->oldestMultiXactId, oldestMulti))
2196 : 0 : SetMultiXactIdLimit(oldestMulti, oldestMultiDB);
2197 : 0 : }
2198 : :
2199 : : /*
2200 : : * Make sure that MultiXactOffset has room for a newly-allocated MultiXactId.
2201 : : *
2202 : : * NB: this is called while holding MultiXactGenLock. We want it to be very
2203 : : * fast most of the time; even when it's not so fast, no actual I/O need
2204 : : * happen unless we're forced to write out a dirty log or xlog page to make
2205 : : * room in shared memory.
2206 : : */
2207 : : static void
2208 : 2 : ExtendMultiXactOffset(MultiXactId multi)
2209 : : {
2210 : 2 : int64 pageno;
2211 : 2 : LWLock *lock;
2212 : :
2213 : : /*
2214 : : * No work except at first MultiXactId of a page. But beware: just after
2215 : : * wraparound, the first MultiXactId of page zero is FirstMultiXactId.
2216 : : */
2217 [ + - - + ]: 2 : if (MultiXactIdToOffsetEntry(multi) != 0 &&
2218 : 2 : multi != FirstMultiXactId)
2219 : 2 : return;
2220 : :
2221 : 0 : pageno = MultiXactIdToOffsetPage(multi);
2222 : 0 : lock = SimpleLruGetBankLock(MultiXactOffsetCtl, pageno);
2223 : :
2224 : 0 : LWLockAcquire(lock, LW_EXCLUSIVE);
2225 : :
2226 : : /* Zero the page and make a WAL entry about it */
2227 : 0 : SimpleLruZeroPage(MultiXactOffsetCtl, pageno);
2228 : 0 : XLogSimpleInsertInt64(RM_MULTIXACT_ID, XLOG_MULTIXACT_ZERO_OFF_PAGE,
2229 : 0 : pageno);
2230 : :
2231 : 0 : LWLockRelease(lock);
2232 [ - + ]: 2 : }
2233 : :
2234 : : /*
2235 : : * Make sure that MultiXactMember has room for the members of a newly-
2236 : : * allocated MultiXactId.
2237 : : *
2238 : : * Like the above routine, this is called while holding MultiXactGenLock;
2239 : : * same comments apply.
2240 : : */
2241 : : static void
2242 : 2 : ExtendMultiXactMember(MultiXactOffset offset, int nmembers)
2243 : : {
2244 : : /*
2245 : : * It's possible that the members span more than one page of the members
2246 : : * file, so we loop to ensure we consider each page. The coding is not
2247 : : * optimal if the members span several pages, but that seems unusual
2248 : : * enough to not worry much about.
2249 : : */
2250 [ + + ]: 4 : while (nmembers > 0)
2251 : : {
2252 : 2 : int flagsoff;
2253 : 2 : int flagsbit;
2254 : 2 : uint32 difference;
2255 : :
2256 : : /*
2257 : : * Only zero when at first entry of a page.
2258 : : */
2259 : 2 : flagsoff = MXOffsetToFlagsOffset(offset);
2260 : 2 : flagsbit = MXOffsetToFlagsBitShift(offset);
2261 [ + - + - ]: 2 : if (flagsoff == 0 && flagsbit == 0)
2262 : : {
2263 : 0 : int64 pageno;
2264 : 0 : LWLock *lock;
2265 : :
2266 : 0 : pageno = MXOffsetToMemberPage(offset);
2267 : 0 : lock = SimpleLruGetBankLock(MultiXactMemberCtl, pageno);
2268 : :
2269 : 0 : LWLockAcquire(lock, LW_EXCLUSIVE);
2270 : :
2271 : : /* Zero the page and make a WAL entry about it */
2272 : 0 : SimpleLruZeroPage(MultiXactMemberCtl, pageno);
2273 : 0 : XLogSimpleInsertInt64(RM_MULTIXACT_ID,
2274 : 0 : XLOG_MULTIXACT_ZERO_MEM_PAGE, pageno);
2275 : :
2276 : 0 : LWLockRelease(lock);
2277 : 0 : }
2278 : :
2279 : : /* Compute the number of items till end of current page. */
2280 : 2 : difference = MULTIXACT_MEMBERS_PER_PAGE - offset % MULTIXACT_MEMBERS_PER_PAGE;
2281 : :
2282 : : /*
2283 : : * Advance to next page. OK if nmembers goes negative.
2284 : : */
2285 : 2 : nmembers -= difference;
2286 : 2 : offset += difference;
2287 : 2 : }
2288 : 2 : }
2289 : :
2290 : : /*
2291 : : * GetOldestMultiXactId
2292 : : *
2293 : : * Return the oldest MultiXactId that's still possibly still seen as live by
2294 : : * any running transaction. Older ones might still exist on disk, but they no
2295 : : * longer have any running member transaction.
2296 : : *
2297 : : * It's not safe to truncate MultiXact SLRU segments on the value returned by
2298 : : * this function; however, it can be set as the new relminmxid for any table
2299 : : * that VACUUM knows has no remaining MXIDs < the same value. It is only safe
2300 : : * to truncate SLRUs when no table can possibly still have a referencing MXID.
2301 : : */
2302 : : MultiXactId
2303 : 7324 : GetOldestMultiXactId(void)
2304 : : {
2305 : 7324 : MultiXactId oldestMXact;
2306 : 7324 : int i;
2307 : :
2308 : : /*
2309 : : * This is the oldest valid value among all the OldestMemberMXactId[] and
2310 : : * OldestVisibleMXactId[] entries, or nextMXact if none are valid.
2311 : : */
2312 : 7324 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2313 : 7324 : oldestMXact = MultiXactState->nextMXact;
2314 [ + + ]: 933442 : for (i = 0; i < MaxOldestSlot; i++)
2315 : : {
2316 : 926118 : MultiXactId thisoldest;
2317 : :
2318 : 926118 : thisoldest = OldestMemberMXactId[i];
2319 [ + + + + ]: 926118 : if (MultiXactIdIsValid(thisoldest) &&
2320 : 6932 : MultiXactIdPrecedes(thisoldest, oldestMXact))
2321 : 40 : oldestMXact = thisoldest;
2322 : 926118 : thisoldest = OldestVisibleMXactId[i];
2323 [ - + # # ]: 926118 : if (MultiXactIdIsValid(thisoldest) &&
2324 : 0 : MultiXactIdPrecedes(thisoldest, oldestMXact))
2325 : 0 : oldestMXact = thisoldest;
2326 : 926118 : }
2327 : :
2328 : 7324 : LWLockRelease(MultiXactGenLock);
2329 : :
2330 : 14648 : return oldestMXact;
2331 : 7324 : }
2332 : :
2333 : : /*
2334 : : * Calculate the oldest member offset and install it in MultiXactState, where
2335 : : * it can be used to adjust multixid freezing cutoffs.
2336 : : */
2337 : : static void
2338 : 6 : SetOldestOffset(void)
2339 : : {
2340 : 6 : MultiXactId oldestMultiXactId;
2341 : 6 : MultiXactId nextMXact;
2342 : 6 : MultiXactOffset oldestOffset = 0; /* placate compiler */
2343 : 6 : MultiXactOffset nextOffset;
2344 : 6 : bool oldestOffsetKnown = false;
2345 : :
2346 : : /*
2347 : : * NB: Have to prevent concurrent truncation, we might otherwise try to
2348 : : * lookup an oldestMulti that's concurrently getting truncated away.
2349 : : */
2350 : 6 : LWLockAcquire(MultiXactTruncationLock, LW_SHARED);
2351 : :
2352 : : /* Read relevant fields from shared memory. */
2353 : 6 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2354 : 6 : oldestMultiXactId = MultiXactState->oldestMultiXactId;
2355 : 6 : nextMXact = MultiXactState->nextMXact;
2356 : 6 : nextOffset = MultiXactState->nextOffset;
2357 [ + - ]: 6 : Assert(MultiXactState->finishedStartup);
2358 : 6 : LWLockRelease(MultiXactGenLock);
2359 : :
2360 : : /*
2361 : : * Determine the offset of the oldest multixact. Normally, we can read
2362 : : * the offset from the multixact itself, but there's an important special
2363 : : * case: if there are no multixacts in existence at all, oldestMXact
2364 : : * obviously can't point to one. It will instead point to the multixact
2365 : : * ID that will be assigned the next time one is needed.
2366 : : */
2367 [ + - ]: 6 : if (oldestMultiXactId == nextMXact)
2368 : : {
2369 : : /*
2370 : : * When the next multixact gets created, it will be stored at the next
2371 : : * offset.
2372 : : */
2373 : 6 : oldestOffset = nextOffset;
2374 : 6 : oldestOffsetKnown = true;
2375 : 6 : }
2376 : : else
2377 : : {
2378 : : /*
2379 : : * Look up the offset at which the oldest existing multixact's members
2380 : : * are stored. If we cannot find it, be careful not to fail, and
2381 : : * leave oldestOffset unchanged. oldestOffset is initialized to zero
2382 : : * at system startup, which prevents truncating members until a proper
2383 : : * value is calculated.
2384 : : *
2385 : : * (We had bugs in early releases of PostgreSQL 9.3.X and 9.4.X where
2386 : : * the supposedly-earliest multixact might not really exist. Those
2387 : : * should be long gone by now, so this should not fail, but let's
2388 : : * still be defensive.)
2389 : : */
2390 : 0 : oldestOffsetKnown =
2391 : 0 : find_multixact_start(oldestMultiXactId, &oldestOffset);
2392 : :
2393 [ # # ]: 0 : if (oldestOffsetKnown)
2394 [ # # # # ]: 0 : ereport(DEBUG1,
2395 : : (errmsg_internal("oldest MultiXactId member is at offset %" PRIu64,
2396 : : oldestOffset)));
2397 : : else
2398 [ # # # # ]: 0 : ereport(LOG,
2399 : : (errmsg("MultiXact member truncation is disabled because oldest checkpointed MultiXact %u does not exist on disk",
2400 : : oldestMultiXactId)));
2401 : : }
2402 : :
2403 : 6 : LWLockRelease(MultiXactTruncationLock);
2404 : :
2405 : : /* Install the computed value */
2406 [ - + ]: 6 : if (oldestOffsetKnown)
2407 : : {
2408 : 6 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2409 : 6 : MultiXactState->oldestOffset = oldestOffset;
2410 : 6 : LWLockRelease(MultiXactGenLock);
2411 : 6 : }
2412 : 6 : }
2413 : :
2414 : : /*
2415 : : * Find the starting offset of the given MultiXactId.
2416 : : *
2417 : : * Returns false if the file containing the multi does not exist on disk.
2418 : : * Otherwise, returns true and sets *result to the starting member offset.
2419 : : *
2420 : : * This function does not prevent concurrent truncation, so if that's
2421 : : * required, the caller has to protect against that.
2422 : : */
2423 : : static bool
2424 : 0 : find_multixact_start(MultiXactId multi, MultiXactOffset *result)
2425 : : {
2426 : 0 : MultiXactOffset offset;
2427 : 0 : int64 pageno;
2428 : 0 : int entryno;
2429 : 0 : int slotno;
2430 : 0 : MultiXactOffset *offptr;
2431 : :
2432 [ # # ]: 0 : Assert(MultiXactState->finishedStartup);
2433 : :
2434 : 0 : pageno = MultiXactIdToOffsetPage(multi);
2435 : 0 : entryno = MultiXactIdToOffsetEntry(multi);
2436 : :
2437 : : /*
2438 : : * Write out dirty data, so PhysicalPageExists can work correctly.
2439 : : */
2440 : 0 : SimpleLruWriteAll(MultiXactOffsetCtl, true);
2441 : 0 : SimpleLruWriteAll(MultiXactMemberCtl, true);
2442 : :
2443 [ # # ]: 0 : if (!SimpleLruDoesPhysicalPageExist(MultiXactOffsetCtl, pageno))
2444 : 0 : return false;
2445 : :
2446 : : /* lock is acquired by SimpleLruReadPage_ReadOnly */
2447 : 0 : slotno = SimpleLruReadPage_ReadOnly(MultiXactOffsetCtl, pageno, multi);
2448 : 0 : offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
2449 : 0 : offptr += entryno;
2450 : 0 : offset = *offptr;
2451 : 0 : LWLockRelease(SimpleLruGetBankLock(MultiXactOffsetCtl, pageno));
2452 : :
2453 : 0 : *result = offset;
2454 : 0 : return true;
2455 : 0 : }
2456 : :
2457 : : /*
2458 : : * GetMultiXactInfo
2459 : : *
2460 : : * Returns information about the current MultiXact state, as of:
2461 : : * multixacts: Number of MultiXacts (nextMultiXactId - oldestMultiXactId)
2462 : : * nextOffset: Next-to-be-assigned offset
2463 : : * oldestMultiXactId: Oldest MultiXact ID still in use
2464 : : * oldestOffset: Oldest offset still in use
2465 : : */
2466 : : void
2467 : 703 : GetMultiXactInfo(uint32 *multixacts, MultiXactOffset *nextOffset,
2468 : : MultiXactId *oldestMultiXactId, MultiXactOffset *oldestOffset)
2469 : : {
2470 : 703 : MultiXactId nextMultiXactId;
2471 : :
2472 : 703 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2473 : 703 : *nextOffset = MultiXactState->nextOffset;
2474 : 703 : *oldestMultiXactId = MultiXactState->oldestMultiXactId;
2475 : 703 : nextMultiXactId = MultiXactState->nextMXact;
2476 : 703 : *oldestOffset = MultiXactState->oldestOffset;
2477 : 703 : LWLockRelease(MultiXactGenLock);
2478 : :
2479 : 703 : *multixacts = nextMultiXactId - *oldestMultiXactId;
2480 : 703 : }
2481 : :
2482 : : /*
2483 : : * Multixact members can be removed once the multixacts that refer to them
2484 : : * are older than every datminmxid. autovacuum_multixact_freeze_max_age and
2485 : : * vacuum_multixact_freeze_table_age work together to make sure we never have
2486 : : * too many multixacts; we hope that, at least under normal circumstances,
2487 : : * this will also be sufficient to keep us from using too many offsets.
2488 : : * However, if the average multixact has many members, we might accumulate a
2489 : : * large amount of members, consuming disk space, while still using few enough
2490 : : * multixids that the multixid limits fail to trigger relminmxid advancement
2491 : : * by VACUUM.
2492 : : *
2493 : : * To prevent that, if the members space usage exceeds a threshold
2494 : : * (MULTIXACT_MEMBER_LOW_THRESHOLD), we effectively reduce
2495 : : * autovacuum_multixact_freeze_max_age to a value just less than the number of
2496 : : * multixacts in use. We hope that this will quickly trigger autovacuuming on
2497 : : * the table or tables with the oldest relminmxid, thus allowing datminmxid
2498 : : * values to advance and removing some members.
2499 : : *
2500 : : * As the amount of the member space in use grows, we become more aggressive
2501 : : * in clamping this value. That not only causes autovacuum to ramp up, but
2502 : : * also makes any manual vacuums the user issues more aggressive. This
2503 : : * happens because vacuum_get_cutoffs() will clamp the freeze table and the
2504 : : * minimum freeze age cutoffs based on the effective
2505 : : * autovacuum_multixact_freeze_max_age this function returns. At the extreme,
2506 : : * when the members usage reaches MULTIXACT_MEMBER_HIGH_THRESHOLD, we clamp
2507 : : * freeze_max_age to zero, and every vacuum of any table will freeze every
2508 : : * multixact.
2509 : : */
2510 : : int
2511 : 700 : MultiXactMemberFreezeThreshold(void)
2512 : : {
2513 : 700 : uint32 multixacts;
2514 : 700 : uint32 victim_multixacts;
2515 : 700 : double fraction;
2516 : 700 : int result;
2517 : 700 : MultiXactId oldestMultiXactId;
2518 : 700 : MultiXactOffset oldestOffset;
2519 : 700 : MultiXactOffset nextOffset;
2520 : 700 : uint64 members;
2521 : :
2522 : : /* Read the current offsets and multixact usage. */
2523 : 700 : GetMultiXactInfo(&multixacts, &nextOffset, &oldestMultiXactId, &oldestOffset);
2524 : 700 : members = nextOffset - oldestOffset;
2525 : :
2526 : : /* If member space utilization is low, no special action is required. */
2527 [ + - ]: 700 : if (members <= MULTIXACT_MEMBER_LOW_THRESHOLD)
2528 : 700 : return autovacuum_multixact_freeze_max_age;
2529 : :
2530 : : /*
2531 : : * Compute a target for relminmxid advancement. The number of multixacts
2532 : : * we try to eliminate from the system is based on how far we are past
2533 : : * MULTIXACT_MEMBER_LOW_THRESHOLD.
2534 : : *
2535 : : * The way this formula works is that when members is exactly at the low
2536 : : * threshold, fraction = 0.0, and we set freeze_max_age equal to
2537 : : * mxid_age(oldestMultiXactId). As members grows further, towards the
2538 : : * high threshold, fraction grows linearly from 0.0 to 1.0, and the result
2539 : : * shrinks from mxid_age(oldestMultiXactId) to 0. Beyond the high
2540 : : * threshold, fraction > 1.0 and the result is clamped to 0.
2541 : : */
2542 : 0 : fraction = (double) (members - MULTIXACT_MEMBER_LOW_THRESHOLD) /
2543 : : (MULTIXACT_MEMBER_HIGH_THRESHOLD - MULTIXACT_MEMBER_LOW_THRESHOLD);
2544 : :
2545 : : /* fraction could be > 1.0, but lowest possible freeze age is zero */
2546 [ # # ]: 0 : if (fraction >= 1.0)
2547 : 0 : return 0;
2548 : :
2549 : 0 : victim_multixacts = multixacts * fraction;
2550 : 0 : result = multixacts - victim_multixacts;
2551 : :
2552 : : /*
2553 : : * Clamp to autovacuum_multixact_freeze_max_age, so that we never make
2554 : : * autovacuum less aggressive than it would otherwise be.
2555 : : */
2556 [ # # ]: 0 : return Min(result, autovacuum_multixact_freeze_max_age);
2557 : 700 : }
2558 : :
2559 : :
2560 : : /*
2561 : : * Delete members segments older than newOldestOffset
2562 : : */
2563 : : static void
2564 : 0 : PerformMembersTruncation(MultiXactOffset newOldestOffset)
2565 : : {
2566 : 0 : SimpleLruTruncate(MultiXactMemberCtl,
2567 : 0 : MXOffsetToMemberPage(newOldestOffset));
2568 : 0 : }
2569 : :
2570 : : /*
2571 : : * Delete offsets segments older than newOldestMulti
2572 : : */
2573 : : static void
2574 : 0 : PerformOffsetsTruncation(MultiXactId newOldestMulti)
2575 : : {
2576 : : /*
2577 : : * We step back one multixact to avoid passing a cutoff page that hasn't
2578 : : * been created yet in the rare case that oldestMulti would be the first
2579 : : * item on a page and oldestMulti == nextMulti. In that case, if we
2580 : : * didn't subtract one, we'd trigger SimpleLruTruncate's wraparound
2581 : : * detection.
2582 : : */
2583 : 0 : SimpleLruTruncate(MultiXactOffsetCtl,
2584 : 0 : MultiXactIdToOffsetPage(PreviousMultiXactId(newOldestMulti)));
2585 : 0 : }
2586 : :
2587 : : /*
2588 : : * Remove all MultiXactOffset and MultiXactMember segments before the oldest
2589 : : * ones still of interest.
2590 : : *
2591 : : * This is only called on a primary as part of vacuum (via
2592 : : * vac_truncate_clog()). During recovery truncation is done by replaying
2593 : : * truncation WAL records logged here.
2594 : : *
2595 : : * newOldestMulti is the oldest currently required multixact, newOldestMultiDB
2596 : : * is one of the databases preventing newOldestMulti from increasing.
2597 : : */
2598 : : void
2599 : 2 : TruncateMultiXact(MultiXactId newOldestMulti, Oid newOldestMultiDB)
2600 : : {
2601 : 2 : MultiXactId oldestMulti;
2602 : 2 : MultiXactId nextMulti;
2603 : 2 : MultiXactOffset newOldestOffset;
2604 : 2 : MultiXactOffset nextOffset;
2605 : :
2606 [ + - ]: 2 : Assert(!RecoveryInProgress());
2607 [ + - ]: 2 : Assert(MultiXactState->finishedStartup);
2608 [ + - ]: 2 : Assert(MultiXactIdIsValid(newOldestMulti));
2609 : :
2610 : : /*
2611 : : * We can only allow one truncation to happen at once. Otherwise parts of
2612 : : * members might vanish while we're doing lookups or similar. There's no
2613 : : * need to have an interlock with creating new multis or such, since those
2614 : : * are constrained by the limits (which only grow, never shrink).
2615 : : */
2616 : 2 : LWLockAcquire(MultiXactTruncationLock, LW_EXCLUSIVE);
2617 : :
2618 : 2 : LWLockAcquire(MultiXactGenLock, LW_SHARED);
2619 : 2 : nextMulti = MultiXactState->nextMXact;
2620 : 2 : nextOffset = MultiXactState->nextOffset;
2621 : 2 : oldestMulti = MultiXactState->oldestMultiXactId;
2622 : 2 : LWLockRelease(MultiXactGenLock);
2623 : :
2624 : : /*
2625 : : * Make sure to only attempt truncation if there's values to truncate
2626 : : * away. In normal processing values shouldn't go backwards, but there's
2627 : : * some corner cases (due to bugs) where that's possible.
2628 : : */
2629 [ + - ]: 2 : if (MultiXactIdPrecedesOrEquals(newOldestMulti, oldestMulti))
2630 : : {
2631 : 2 : LWLockRelease(MultiXactTruncationLock);
2632 : 2 : return;
2633 : : }
2634 : :
2635 : : /*
2636 : : * Compute up to where to truncate MultiXactMember. Lookup the
2637 : : * corresponding member offset for newOldestMulti for that.
2638 : : */
2639 [ # # ]: 0 : if (newOldestMulti == nextMulti)
2640 : : {
2641 : : /* there are NO MultiXacts */
2642 : 0 : newOldestOffset = nextOffset;
2643 : 0 : }
2644 [ # # ]: 0 : else if (!find_multixact_start(newOldestMulti, &newOldestOffset))
2645 : : {
2646 [ # # # # ]: 0 : ereport(LOG,
2647 : : (errmsg("cannot truncate up to MultiXact %u because it does not exist on disk, skipping truncation",
2648 : : newOldestMulti)));
2649 : 0 : LWLockRelease(MultiXactTruncationLock);
2650 : 0 : return;
2651 : : }
2652 : :
2653 : : /*
2654 : : * On crash, MultiXactIdCreateFromMembers() can leave behind multixids
2655 : : * that were not yet written out and hence have zero offset on disk. If
2656 : : * such a multixid becomes oldestMulti, we won't be able to look up its
2657 : : * offset. That should be rare, so we don't try to do anything smart about
2658 : : * it. Just skip the truncation, and hope that by the next truncation
2659 : : * attempt, oldestMulti has advanced to a valid multixid.
2660 : : */
2661 [ # # ]: 0 : if (newOldestOffset == 0)
2662 : : {
2663 [ # # # # ]: 0 : ereport(LOG,
2664 : : (errmsg("cannot truncate up to MultiXact %u because it has invalid offset, skipping truncation",
2665 : : newOldestMulti)));
2666 : 0 : LWLockRelease(MultiXactTruncationLock);
2667 : 0 : return;
2668 : : }
2669 : :
2670 [ # # # # ]: 0 : elog(DEBUG1, "performing multixact truncation: "
2671 : : "oldestMulti %u (offsets segment %" PRIx64 "), "
2672 : : "oldestOffset %" PRIu64 " (members segment %" PRIx64 ")",
2673 : : newOldestMulti,
2674 : : MultiXactIdToOffsetSegment(newOldestMulti),
2675 : : newOldestOffset,
2676 : : MXOffsetToMemberSegment(newOldestOffset));
2677 : :
2678 : : /*
2679 : : * Do truncation, and the WAL logging of the truncation, in a critical
2680 : : * section. That way offsets/members cannot get out of sync anymore, i.e.
2681 : : * once consistent the newOldestMulti will always exist in members, even
2682 : : * if we crashed in the wrong moment.
2683 : : */
2684 : 0 : START_CRIT_SECTION();
2685 : :
2686 : : /*
2687 : : * Prevent checkpoints from being scheduled concurrently. This is critical
2688 : : * because otherwise a truncation record might not be replayed after a
2689 : : * crash/basebackup, even though the state of the data directory would
2690 : : * require it.
2691 : : */
2692 [ # # ]: 0 : Assert((MyProc->delayChkptFlags & DELAY_CHKPT_START) == 0);
2693 : 0 : MyProc->delayChkptFlags |= DELAY_CHKPT_START;
2694 : :
2695 : : /* WAL log truncation */
2696 : 0 : WriteMTruncateXlogRec(newOldestMultiDB, newOldestMulti, newOldestOffset);
2697 : :
2698 : : /*
2699 : : * Update in-memory limits before performing the truncation, while inside
2700 : : * the critical section: Have to do it before truncation, to prevent
2701 : : * concurrent lookups of those values. Has to be inside the critical
2702 : : * section as otherwise a future call to this function would error out,
2703 : : * while looking up the oldest member in offsets, if our caller crashes
2704 : : * before updating the limits.
2705 : : */
2706 : 0 : LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
2707 : 0 : MultiXactState->oldestMultiXactId = newOldestMulti;
2708 : 0 : MultiXactState->oldestMultiXactDB = newOldestMultiDB;
2709 : 0 : MultiXactState->oldestOffset = newOldestOffset;
2710 : 0 : LWLockRelease(MultiXactGenLock);
2711 : :
2712 : : /* First truncate members */
2713 : 0 : PerformMembersTruncation(newOldestOffset);
2714 : :
2715 : : /* Then offsets */
2716 : 0 : PerformOffsetsTruncation(newOldestMulti);
2717 : :
2718 : 0 : MyProc->delayChkptFlags &= ~DELAY_CHKPT_START;
2719 : :
2720 [ # # ]: 0 : END_CRIT_SECTION();
2721 : 0 : LWLockRelease(MultiXactTruncationLock);
2722 [ - + ]: 2 : }
2723 : :
2724 : : /*
2725 : : * Decide whether a MultiXactOffset page number is "older" for truncation
2726 : : * purposes. Analogous to CLOGPagePrecedes().
2727 : : *
2728 : : * Offsetting the values is optional, because MultiXactIdPrecedes() has
2729 : : * translational symmetry.
2730 : : */
2731 : : static bool
2732 : 234 : MultiXactOffsetPagePrecedes(int64 page1, int64 page2)
2733 : : {
2734 : 234 : MultiXactId multi1;
2735 : 234 : MultiXactId multi2;
2736 : :
2737 : 234 : multi1 = ((MultiXactId) page1) * MULTIXACT_OFFSETS_PER_PAGE;
2738 : 234 : multi1 += FirstMultiXactId + 1;
2739 : 234 : multi2 = ((MultiXactId) page2) * MULTIXACT_OFFSETS_PER_PAGE;
2740 : 234 : multi2 += FirstMultiXactId + 1;
2741 : :
2742 [ + + ]: 390 : return (MultiXactIdPrecedes(multi1, multi2) &&
2743 : 312 : MultiXactIdPrecedes(multi1,
2744 : 156 : multi2 + MULTIXACT_OFFSETS_PER_PAGE - 1));
2745 : 234 : }
2746 : :
2747 : : /*
2748 : : * Decide whether a MultiXactMember page number is "older" for truncation
2749 : : * purposes. There is no "invalid offset number" and members never wrap
2750 : : * around, so use the numbers verbatim.
2751 : : */
2752 : : static bool
2753 : 0 : MultiXactMemberPagePrecedes(int64 page1, int64 page2)
2754 : : {
2755 : 0 : return page1 < page2;
2756 : : }
2757 : :
2758 : : /*
2759 : : * Decide which of two MultiXactIds is earlier.
2760 : : *
2761 : : * XXX do we need to do something special for InvalidMultiXactId?
2762 : : * (Doesn't look like it.)
2763 : : */
2764 : : bool
2765 : 135762 : MultiXactIdPrecedes(MultiXactId multi1, MultiXactId multi2)
2766 : : {
2767 : 135762 : int32 diff = (int32) (multi1 - multi2);
2768 : :
2769 : 271524 : return (diff < 0);
2770 : 135762 : }
2771 : :
2772 : : /*
2773 : : * MultiXactIdPrecedesOrEquals -- is multi1 logically <= multi2?
2774 : : *
2775 : : * XXX do we need to do something special for InvalidMultiXactId?
2776 : : * (Doesn't look like it.)
2777 : : */
2778 : : bool
2779 : 523 : MultiXactIdPrecedesOrEquals(MultiXactId multi1, MultiXactId multi2)
2780 : : {
2781 : 523 : int32 diff = (int32) (multi1 - multi2);
2782 : :
2783 : 1046 : return (diff <= 0);
2784 : 523 : }
2785 : :
2786 : :
2787 : : /*
2788 : : * Write a TRUNCATE xlog record
2789 : : *
2790 : : * We must flush the xlog record to disk before returning --- see notes in
2791 : : * TruncateCLOG().
2792 : : */
2793 : : static void
2794 : 0 : WriteMTruncateXlogRec(Oid oldestMultiDB,
2795 : : MultiXactId oldestMulti,
2796 : : MultiXactOffset oldestOffset)
2797 : : {
2798 : 0 : XLogRecPtr recptr;
2799 : 0 : xl_multixact_truncate xlrec;
2800 : :
2801 : 0 : xlrec.oldestMultiDB = oldestMultiDB;
2802 : 0 : xlrec.oldestMulti = oldestMulti;
2803 : 0 : xlrec.oldestOffset = oldestOffset;
2804 : :
2805 : 0 : XLogBeginInsert();
2806 : 0 : XLogRegisterData(&xlrec, SizeOfMultiXactTruncate);
2807 : 0 : recptr = XLogInsert(RM_MULTIXACT_ID, XLOG_MULTIXACT_TRUNCATE_ID);
2808 : 0 : XLogFlush(recptr);
2809 : 0 : }
2810 : :
2811 : : /*
2812 : : * MULTIXACT resource manager's routines
2813 : : */
2814 : : void
2815 : 0 : multixact_redo(XLogReaderState *record)
2816 : : {
2817 : 0 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
2818 : :
2819 : : /* Backup blocks are not used in multixact records */
2820 [ # # ]: 0 : Assert(!XLogRecHasAnyBlockRefs(record));
2821 : :
2822 [ # # ]: 0 : if (info == XLOG_MULTIXACT_ZERO_OFF_PAGE)
2823 : : {
2824 : 0 : int64 pageno;
2825 : :
2826 : 0 : memcpy(&pageno, XLogRecGetData(record), sizeof(pageno));
2827 : 0 : SimpleLruZeroAndWritePage(MultiXactOffsetCtl, pageno);
2828 : 0 : }
2829 [ # # ]: 0 : else if (info == XLOG_MULTIXACT_ZERO_MEM_PAGE)
2830 : : {
2831 : 0 : int64 pageno;
2832 : :
2833 : 0 : memcpy(&pageno, XLogRecGetData(record), sizeof(pageno));
2834 : 0 : SimpleLruZeroAndWritePage(MultiXactMemberCtl, pageno);
2835 : 0 : }
2836 [ # # ]: 0 : else if (info == XLOG_MULTIXACT_CREATE_ID)
2837 : : {
2838 : 0 : xl_multixact_create *xlrec =
2839 : 0 : (xl_multixact_create *) XLogRecGetData(record);
2840 : 0 : TransactionId max_xid;
2841 : 0 : int i;
2842 : :
2843 : : /* Store the data back into the SLRU files */
2844 : 0 : RecordNewMultiXact(xlrec->mid, xlrec->moff, xlrec->nmembers,
2845 : 0 : xlrec->members);
2846 : :
2847 : : /* Make sure nextMXact/nextOffset are beyond what this record has */
2848 : 0 : MultiXactAdvanceNextMXact(NextMultiXactId(xlrec->mid),
2849 : 0 : xlrec->moff + xlrec->nmembers);
2850 : :
2851 : : /*
2852 : : * Make sure nextXid is beyond any XID mentioned in the record. This
2853 : : * should be unnecessary, since any XID found here ought to have other
2854 : : * evidence in the XLOG, but let's be safe.
2855 : : */
2856 : 0 : max_xid = XLogRecGetXid(record);
2857 [ # # ]: 0 : for (i = 0; i < xlrec->nmembers; i++)
2858 : : {
2859 [ # # ]: 0 : if (TransactionIdPrecedes(max_xid, xlrec->members[i].xid))
2860 : 0 : max_xid = xlrec->members[i].xid;
2861 : 0 : }
2862 : :
2863 : 0 : AdvanceNextFullTransactionIdPastXid(max_xid);
2864 : 0 : }
2865 [ # # ]: 0 : else if (info == XLOG_MULTIXACT_TRUNCATE_ID)
2866 : : {
2867 : 0 : xl_multixact_truncate xlrec;
2868 : 0 : int64 pageno;
2869 : :
2870 : 0 : memcpy(&xlrec, XLogRecGetData(record),
2871 : : SizeOfMultiXactTruncate);
2872 : :
2873 [ # # # # ]: 0 : elog(DEBUG1, "replaying multixact truncation: "
2874 : : "oldestMulti %u (offsets segment %" PRIx64 "), "
2875 : : "oldestOffset %" PRIu64 " (members segment %" PRIx64 ")",
2876 : : xlrec.oldestMulti,
2877 : : MultiXactIdToOffsetSegment(xlrec.oldestMulti),
2878 : : xlrec.oldestOffset,
2879 : : MXOffsetToMemberSegment(xlrec.oldestOffset));
2880 : :
2881 : : /* should not be required, but more than cheap enough */
2882 : 0 : LWLockAcquire(MultiXactTruncationLock, LW_EXCLUSIVE);
2883 : :
2884 : : /*
2885 : : * Advance the horizon values, so they're current at the end of
2886 : : * recovery.
2887 : : */
2888 : 0 : SetMultiXactIdLimit(xlrec.oldestMulti, xlrec.oldestMultiDB);
2889 : :
2890 : 0 : PerformMembersTruncation(xlrec.oldestOffset);
2891 : :
2892 : : /*
2893 : : * During XLOG replay, latest_page_number isn't necessarily set up
2894 : : * yet; insert a suitable value to bypass the sanity test in
2895 : : * SimpleLruTruncate.
2896 : : */
2897 : 0 : pageno = MultiXactIdToOffsetPage(xlrec.oldestMulti);
2898 : 0 : pg_atomic_write_u64(&MultiXactOffsetCtl->shared->latest_page_number,
2899 : 0 : pageno);
2900 : 0 : PerformOffsetsTruncation(xlrec.oldestMulti);
2901 : :
2902 : 0 : LWLockRelease(MultiXactTruncationLock);
2903 : 0 : }
2904 : : else
2905 [ # # # # ]: 0 : elog(PANIC, "multixact_redo: unknown op code %u", info);
2906 : 0 : }
2907 : :
2908 : : /*
2909 : : * Entrypoint for sync.c to sync offsets files.
2910 : : */
2911 : : int
2912 : 0 : multixactoffsetssyncfiletag(const FileTag *ftag, char *path)
2913 : : {
2914 : 0 : return SlruSyncFileTag(MultiXactOffsetCtl, ftag, path);
2915 : : }
2916 : :
2917 : : /*
2918 : : * Entrypoint for sync.c to sync members files.
2919 : : */
2920 : : int
2921 : 0 : multixactmemberssyncfiletag(const FileTag *ftag, char *path)
2922 : : {
2923 : 0 : return SlruSyncFileTag(MultiXactMemberCtl, ftag, path);
2924 : : }
|
