1 /*-------------------------------------------------------------------------
3 * predicate_internals.h
4 * POSTGRES internal predicate locking definitions.
7 * Portions Copyright (c) 1996-2011, PostgreSQL Global Development Group
8 * Portions Copyright (c) 1994, Regents of the University of California
10 * src/include/storage/predicate_internals.h
12 *-------------------------------------------------------------------------
14 #ifndef PREDICATE_INTERNALS_H
15 #define PREDICATE_INTERNALS_H
17 #include "storage/lock.h"
22 typedef uint64 SerCommitSeqNo;
25 * Reserved commit sequence numbers:
26 * - 0 is reserved to indicate a non-existent SLRU entry; it cannot be
27 * used as a SerCommitSeqNo, even an invalid one
28 * - InvalidSerCommitSeqNo is used to indicate a transaction that
29 * hasn't committed yet, so use a number greater than all valid
30 * ones to make comparison do the expected thing
31 * - RecoverySerCommitSeqNo is used to refer to transactions that
32 * happened before a crash/recovery, since we restart the sequence
33 * at that point. It's earlier than all normal sequence numbers,
34 * and is only used by recovered prepared transactions
36 #define InvalidSerCommitSeqNo ((SerCommitSeqNo) UINT64CONST(0xFFFFFFFFFFFFFFFF))
37 #define RecoverySerCommitSeqNo ((SerCommitSeqNo) 1)
38 #define FirstNormalSerCommitSeqNo ((SerCommitSeqNo) 2)
41 * The SERIALIZABLEXACT struct contains information needed for each
42 * serializable database transaction to support SSI techniques.
44 * A home-grown list is maintained in shared memory to manage these.
45 * An entry is used when the serializable transaction acquires a snapshot.
46 * Unless the transaction is rolled back, this entry must generally remain
47 * until all concurrent transactions have completed. (There are special
48 * optimizations for READ ONLY transactions which often allow them to be
49 * cleaned up earlier.) A transaction which is rolled back is cleaned up
50 * as soon as possible.
52 * Eligibility for cleanup of committed transactions is generally determined
53 * by comparing the transaction's finishedBefore field to
54 * SerializableGlobalXmin.
56 typedef struct SERIALIZABLEXACT
58 VirtualTransactionId vxid; /* The executing process always has one of
60 SerCommitSeqNo commitSeqNo;
61 union /* these values are not both interesting at
64 SerCommitSeqNo earliestOutConflictCommit; /* when committed with
66 SerCommitSeqNo lastCommitBeforeSnapshot; /* when not committed or
69 SHM_QUEUE outConflicts; /* list of write transactions whose data we
71 SHM_QUEUE inConflicts; /* list of read transactions which couldn't
73 SHM_QUEUE predicateLocks; /* list of associated PREDICATELOCK objects */
74 SHM_QUEUE finishedLink; /* list link in
75 * FinishedSerializableTransactions */
78 * for r/o transactions: list of concurrent r/w transactions that we could
79 * potentially have conflicts with, and vice versa for r/w transactions
81 SHM_QUEUE possibleUnsafeConflicts;
83 TransactionId topXid; /* top level xid for the transaction, if one
84 * exists; else invalid */
85 TransactionId finishedBefore; /* invalid means still running; else
86 * the struct expires when no
87 * serializable xids are before this. */
88 TransactionId xmin; /* the transaction's snapshot xmin */
89 uint32 flags; /* OR'd combination of values defined below */
90 int pid; /* pid of associated process */
93 #define SXACT_FLAG_COMMITTED 0x00000001 /* already committed */
94 #define SXACT_FLAG_PREPARED 0x00000002 /* about to commit */
95 #define SXACT_FLAG_ROLLED_BACK 0x00000004 /* already rolled back */
96 #define SXACT_FLAG_DOOMED 0x00000008 /* will roll back */
98 * The following flag actually means that the flagged transaction has a
99 * conflict out *to a transaction which committed ahead of it*. It's hard
100 * to get that into a name of a reasonable length.
102 #define SXACT_FLAG_CONFLICT_OUT 0x00000010
103 #define SXACT_FLAG_READ_ONLY 0x00000020
104 #define SXACT_FLAG_DEFERRABLE_WAITING 0x00000040
105 #define SXACT_FLAG_RO_SAFE 0x00000080
106 #define SXACT_FLAG_RO_UNSAFE 0x00000100
107 #define SXACT_FLAG_SUMMARY_CONFLICT_IN 0x00000200
108 #define SXACT_FLAG_SUMMARY_CONFLICT_OUT 0x00000400
111 * The following types are used to provide an ad hoc list for holding
112 * SERIALIZABLEXACT objects. An HTAB is overkill, since there is no need to
113 * access these by key -- there are direct pointers to these objects where
114 * needed. If a shared memory list is created, these types can probably be
115 * eliminated in favor of using the general solution.
117 typedef struct PredXactListElementData
120 SERIALIZABLEXACT sxact;
121 } PredXactListElementData;
123 typedef struct PredXactListElementData *PredXactListElement;
125 #define PredXactListElementDataSize \
126 ((Size)MAXALIGN(sizeof(PredXactListElementData)))
128 typedef struct PredXactListData
130 SHM_QUEUE availableList;
131 SHM_QUEUE activeList;
134 * These global variables are maintained when registering and cleaning up
135 * serializable transactions. They must be global across all backends,
136 * but are not needed outside the predicate.c source file. Protected by
137 * SerializableXactHashLock.
139 TransactionId SxactGlobalXmin; /* global xmin for active serializable
141 int SxactGlobalXminCount; /* how many active serializable
142 * transactions have this xmin */
143 int WritableSxactCount; /* how many non-read-only serializable
144 * transactions are active */
145 SerCommitSeqNo LastSxactCommitSeqNo; /* a strictly monotonically
146 * increasing number for
147 * commits of serializable
149 /* Protected by SerializableXactHashLock. */
150 SerCommitSeqNo CanPartialClearThrough; /* can clear predicate locks
151 * and inConflicts for
152 * committed transactions
153 * through this seq no */
154 /* Protected by SerializableFinishedListLock. */
155 SerCommitSeqNo HavePartialClearedThrough; /* have cleared through this
157 SERIALIZABLEXACT *OldCommittedSxact; /* shared copy of dummy sxact */
159 PredXactListElement element;
162 typedef struct PredXactListData *PredXactList;
164 #define PredXactListDataSize \
165 ((Size)MAXALIGN(sizeof(PredXactListData)))
169 * The following types are used to provide lists of rw-conflicts between
170 * pairs of transactions. Since exactly the same information is needed,
171 * they are also used to record possible unsafe transaction relationships
172 * for purposes of identifying safe snapshots for read-only transactions.
174 * When a RWConflictData is not in use to record either type of relationship
175 * between a pair of transactions, it is kept on an "available" list. The
176 * outLink field is used for maintaining that list.
178 typedef struct RWConflictData
180 SHM_QUEUE outLink; /* link for list of conflicts out from a sxact */
181 SHM_QUEUE inLink; /* link for list of conflicts in to a sxact */
182 SERIALIZABLEXACT *sxactOut;
183 SERIALIZABLEXACT *sxactIn;
186 typedef struct RWConflictData *RWConflict;
188 #define RWConflictDataSize \
189 ((Size)MAXALIGN(sizeof(RWConflictData)))
191 typedef struct RWConflictPoolHeaderData
193 SHM_QUEUE availableList;
195 } RWConflictPoolHeaderData;
197 typedef struct RWConflictPoolHeaderData *RWConflictPoolHeader;
199 #define RWConflictPoolHeaderDataSize \
200 ((Size)MAXALIGN(sizeof(RWConflictPoolHeaderData)))
204 * The SERIALIZABLEXIDTAG struct identifies an xid assigned to a serializable
205 * transaction or any of its subtransactions.
207 typedef struct SERIALIZABLEXIDTAG
210 } SERIALIZABLEXIDTAG;
213 * The SERIALIZABLEXID struct provides a link from a TransactionId for a
214 * serializable transaction to the related SERIALIZABLEXACT record, even if
215 * the transaction has completed and its connection has been closed.
217 * These are created as new top level transaction IDs are first assigned to
218 * transactions which are participating in predicate locking. This may
219 * never happen for a particular transaction if it doesn't write anything.
220 * They are removed with their related serializable transaction objects.
222 * The SubTransGetTopmostTransaction method is used where necessary to get
223 * from an XID which might be from a subtransaction to the top level XID.
225 typedef struct SERIALIZABLEXID
228 SERIALIZABLEXIDTAG tag;
231 SERIALIZABLEXACT *myXact; /* pointer to the top level transaction data */
236 * The PREDICATELOCKTARGETTAG struct identifies a database object which can
237 * be the target of predicate locks.
239 * Note that the hash function being used doesn't properly respect tag
240 * length -- it will go to a four byte boundary past the end of the tag.
241 * If you change this struct, make sure any slack space is initialized,
242 * so that any random bytes in the middle or at the end are not included
245 * TODO SSI: If we always use the same fields for the same type of value, we
246 * should rename these. Holding off until it's clear there are no exceptions.
247 * Since indexes are relations with blocks and tuples, it's looking likely that
248 * the rename will be possible. If not, we may need to divide the last field
249 * and use part of it for a target type, so that we know how to interpret the
252 typedef struct PREDICATELOCKTARGETTAG
254 uint32 locktag_field1; /* a 32-bit ID field */
255 uint32 locktag_field2; /* a 32-bit ID field */
256 uint32 locktag_field3; /* a 32-bit ID field */
257 uint32 locktag_field4; /* a 32-bit ID field */
258 uint32 locktag_field5; /* a 32-bit ID field */
259 } PREDICATELOCKTARGETTAG;
262 * The PREDICATELOCKTARGET struct represents a database object on which there
263 * are predicate locks.
265 * A hash list of these objects is maintained in shared memory. An entry is
266 * added when a predicate lock is requested on an object which doesn't
267 * already have one. An entry is removed when the last lock is removed from
270 * Because a particular target might become obsolete, due to update to a new
271 * version, before the reading transaction is obsolete, we need some way to
272 * prevent errors from reuse of a tuple ID. Rather than attempting to clean
273 * up the targets as the related tuples are pruned or vacuumed, we check the
274 * xmin on access. This should be far less costly.
276 typedef struct PREDICATELOCKTARGET
279 PREDICATELOCKTARGETTAG tag; /* unique identifier of lockable object */
282 SHM_QUEUE predicateLocks; /* list of PREDICATELOCK objects assoc. with
283 * predicate lock target */
284 } PREDICATELOCKTARGET;
288 * The PREDICATELOCKTAG struct identifies an individual predicate lock.
290 * It is the combination of predicate lock target (which is a lockable
291 * object) and a serializable transaction which has acquired a lock on that
294 typedef struct PREDICATELOCKTAG
296 PREDICATELOCKTARGET *myTarget;
297 SERIALIZABLEXACT *myXact;
301 * The PREDICATELOCK struct represents an individual lock.
303 * An entry can be created here when the related database object is read, or
304 * by promotion of multiple finer-grained targets. All entries related to a
305 * serializable transaction are removed when that serializable transaction is
306 * cleaned up. Entries can also be removed when they are combined into a
307 * single coarser-grained lock entry.
309 typedef struct PREDICATELOCK
312 PREDICATELOCKTAG tag; /* unique identifier of lock */
315 SHM_QUEUE targetLink; /* list link in PREDICATELOCKTARGET's list of
317 SHM_QUEUE xactLink; /* list link in SERIALIZABLEXACT's list of
319 SerCommitSeqNo commitSeqNo; /* only used for summarized predicate locks */
324 * The LOCALPREDICATELOCK struct represents a local copy of data which is
325 * also present in the PREDICATELOCK table, organized for fast access without
326 * needing to acquire a LWLock. It is strictly for optimization.
328 * Each serializable transaction creates its own local hash table to hold a
329 * collection of these. This information is used to determine when a number
330 * of fine-grained locks should be promoted to a single coarser-grained lock.
331 * The information is maintained more-or-less in parallel to the
332 * PREDICATELOCK data, but because this data is not protected by locks and is
333 * only used in an optimization heuristic, it is allowed to drift in a few
334 * corner cases where maintaining exact data would be expensive.
336 * The hash table is created when the serializable transaction acquires its
337 * snapshot, and its memory is released upon completion of the transaction.
339 typedef struct LOCALPREDICATELOCK
342 PREDICATELOCKTARGETTAG tag; /* unique identifier of lockable object */
345 bool held; /* is lock held, or just its children? */
346 int childLocks; /* number of child locks currently held */
347 } LOCALPREDICATELOCK;
351 * The types of predicate locks which can be acquired.
353 typedef enum PredicateLockTargetType
355 PREDLOCKTAG_RELATION,
358 /* TODO SSI: Other types may be needed for index locking */
359 } PredicateLockTargetType;
363 * This structure is used to quickly capture a copy of all predicate
364 * locks. This is currently used only by the pg_lock_status function,
365 * which in turn is used by the pg_locks view.
367 typedef struct PredicateLockData
370 PREDICATELOCKTARGETTAG *locktags;
371 SERIALIZABLEXACT *xacts;
376 * These macros define how we map logical IDs of lockable objects into the
377 * physical fields of PREDICATELOCKTARGETTAG. Use these to set up values,
378 * rather than accessing the fields directly. Note multiple eval of target!
380 #define SET_PREDICATELOCKTARGETTAG_RELATION(locktag,dboid,reloid) \
381 ((locktag).locktag_field1 = (dboid), \
382 (locktag).locktag_field2 = (reloid), \
383 (locktag).locktag_field3 = InvalidBlockNumber, \
384 (locktag).locktag_field4 = InvalidOffsetNumber, \
385 (locktag).locktag_field5 = InvalidTransactionId)
387 #define SET_PREDICATELOCKTARGETTAG_PAGE(locktag,dboid,reloid,blocknum) \
388 ((locktag).locktag_field1 = (dboid), \
389 (locktag).locktag_field2 = (reloid), \
390 (locktag).locktag_field3 = (blocknum), \
391 (locktag).locktag_field4 = InvalidOffsetNumber, \
392 (locktag).locktag_field5 = InvalidTransactionId)
394 #define SET_PREDICATELOCKTARGETTAG_TUPLE(locktag,dboid,reloid,blocknum,offnum,xmin) \
395 ((locktag).locktag_field1 = (dboid), \
396 (locktag).locktag_field2 = (reloid), \
397 (locktag).locktag_field3 = (blocknum), \
398 (locktag).locktag_field4 = (offnum), \
399 (locktag).locktag_field5 = (xmin))
401 #define GET_PREDICATELOCKTARGETTAG_DB(locktag) \
402 ((Oid) (locktag).locktag_field1)
403 #define GET_PREDICATELOCKTARGETTAG_RELATION(locktag) \
404 ((Oid) (locktag).locktag_field2)
405 #define GET_PREDICATELOCKTARGETTAG_PAGE(locktag) \
406 ((BlockNumber) (locktag).locktag_field3)
407 #define GET_PREDICATELOCKTARGETTAG_OFFSET(locktag) \
408 ((OffsetNumber) (locktag).locktag_field4)
409 #define GET_PREDICATELOCKTARGETTAG_XMIN(locktag) \
410 ((TransactionId) (locktag).locktag_field5)
411 #define GET_PREDICATELOCKTARGETTAG_TYPE(locktag) \
412 (((locktag).locktag_field4 != InvalidOffsetNumber) ? PREDLOCKTAG_TUPLE : \
413 (((locktag).locktag_field3 != InvalidBlockNumber) ? PREDLOCKTAG_PAGE : \
414 PREDLOCKTAG_RELATION))
417 * Two-phase commit statefile records. There are two types: for each
418 * transaction, we generate one per-transaction record and a variable
419 * number of per-predicate-lock records.
421 typedef enum TwoPhasePredicateRecordType
423 TWOPHASEPREDICATERECORD_XACT,
424 TWOPHASEPREDICATERECORD_LOCK
425 } TwoPhasePredicateRecordType;
428 * Per-transaction information to reconstruct a SERIALIZABLEXACT. Not
429 * much is needed because most of it not meaningful for a recovered
430 * prepared transaction.
432 * In particular, we do not record the in and out conflict lists for a
433 * prepared transaction because the associated SERIALIZABLEXACTs will
434 * not be available after recovery. Instead, we simply record the
435 * existence of each type of conflict by setting the transaction's
436 * summary conflict in/out flag.
438 typedef struct TwoPhasePredicateXactRecord
442 } TwoPhasePredicateXactRecord;
445 typedef struct TwoPhasePredicateLockRecord
447 PREDICATELOCKTARGETTAG target;
448 } TwoPhasePredicateLockRecord;
450 typedef struct TwoPhasePredicateRecord
452 TwoPhasePredicateRecordType type;
455 TwoPhasePredicateXactRecord xactRecord;
456 TwoPhasePredicateLockRecord lockRecord;
458 } TwoPhasePredicateRecord;
461 * Define a macro to use for an "empty" SERIALIZABLEXACT reference.
463 #define InvalidSerializableXact ((SERIALIZABLEXACT *) NULL)
467 * Function definitions for functions needing awareness of predicate
470 extern PredicateLockData *GetPredicateLockStatusData(void);
473 #endif /* PREDICATE_INTERNALS_H */