codekingpro/portable-devtools

1/*-------------------------------------------------------------------------2 *3 * proc.h4 *	  per-process shared memory data structures5 *6 *7 * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group8 * Portions Copyright (c) 1994, Regents of the University of California9 *10 * src/include/storage/proc.h11 *12 *-------------------------------------------------------------------------13 */14#ifndef _PROC_H_15#define _PROC_H_16 17#include "access/clog.h"18#include "access/xlogdefs.h"19#include "lib/ilist.h"20#include "storage/latch.h"21#include "storage/lock.h"22#include "storage/pg_sema.h"23#include "storage/proclist_types.h"24 25/*26 * Each backend advertises up to PGPROC_MAX_CACHED_SUBXIDS TransactionIds27 * for non-aborted subtransactions of its current top transaction.  These28 * have to be treated as running XIDs by other backends.29 *30 * We also keep track of whether the cache overflowed (ie, the transaction has31 * generated at least one subtransaction that didn't fit in the cache).32 * If none of the caches have overflowed, we can assume that an XID that's not33 * listed anywhere in the PGPROC array is not a running transaction.  Else we34 * have to look at pg_subtrans.35 *36 * See src/test/isolation/specs/subxid-overflow.spec if you change this.37 */38#define PGPROC_MAX_CACHED_SUBXIDS 64	/* XXX guessed-at value */39 40typedef struct XidCacheStatus41{42	/* number of cached subxids, never more than PGPROC_MAX_CACHED_SUBXIDS */43	uint8		count;44	/* has PGPROC->subxids overflowed */45	bool		overflowed;46} XidCacheStatus;47 48struct XidCache49{50	TransactionId xids[PGPROC_MAX_CACHED_SUBXIDS];51};52 53/*54 * Flags for PGPROC->statusFlags and PROC_HDR->statusFlags[]55 */56#define		PROC_IS_AUTOVACUUM	0x01	/* is it an autovac worker? */57#define		PROC_IN_VACUUM		0x02	/* currently running lazy vacuum */58#define		PROC_IN_SAFE_IC		0x04	/* currently running CREATE INDEX59										 * CONCURRENTLY or REINDEX60										 * CONCURRENTLY on non-expressional,61										 * non-partial index */62#define		PROC_VACUUM_FOR_WRAPAROUND	0x08	/* set by autovac only */63#define		PROC_IN_LOGICAL_DECODING	0x10	/* currently doing logical64												 * decoding outside xact */65#define		PROC_AFFECTS_ALL_HORIZONS	0x20	/* this proc's xmin must be66												 * included in vacuum horizons67												 * in all databases */68 69/* flags reset at EOXact */70#define		PROC_VACUUM_STATE_MASK \71	(PROC_IN_VACUUM | PROC_IN_SAFE_IC | PROC_VACUUM_FOR_WRAPAROUND)72 73/*74 * Xmin-related flags. Make sure any flags that affect how the process' Xmin75 * value is interpreted by VACUUM are included here.76 */77#define		PROC_XMIN_FLAGS (PROC_IN_VACUUM | PROC_IN_SAFE_IC)78 79/*80 * We allow a small number of "weak" relation locks (AccessShareLock,81 * RowShareLock, RowExclusiveLock) to be recorded in the PGPROC structure82 * rather than the main lock table.  This eases contention on the lock83 * manager LWLocks.  See storage/lmgr/README for additional details.84 */85#define		FP_LOCK_SLOTS_PER_BACKEND 1686 87/*88 * An invalid pgprocno.  Must be larger than the maximum number of PGPROC89 * structures we could possibly have.  See comments for MAX_BACKENDS.90 */91#define INVALID_PGPROCNO		PG_INT32_MAX92 93/*94 * Flags for PGPROC.delayChkptFlags95 *96 * These flags can be used to delay the start or completion of a checkpoint97 * for short periods. A flag is in effect if the corresponding bit is set in98 * the PGPROC of any backend.99 *100 * For our purposes here, a checkpoint has three phases: (1) determine the101 * location to which the redo pointer will be moved, (2) write all the102 * data durably to disk, and (3) WAL-log the checkpoint.103 *104 * Setting DELAY_CHKPT_START prevents the system from moving from phase 1105 * to phase 2. This is useful when we are performing a WAL-logged modification106 * of data that will be flushed to disk in phase 2. By setting this flag107 * before writing WAL and clearing it after we've both written WAL and108 * performed the corresponding modification, we ensure that if the WAL record109 * is inserted prior to the new redo point, the corresponding data changes will110 * also be flushed to disk before the checkpoint can complete. (In the111 * extremely common case where the data being modified is in shared buffers112 * and we acquire an exclusive content lock on the relevant buffers before113 * writing WAL, this mechanism is not needed, because phase 2 will block114 * until we release the content lock and then flush the modified data to115 * disk.)116 *117 * Setting DELAY_CHKPT_COMPLETE prevents the system from moving from phase 2118 * to phase 3. This is useful if we are performing a WAL-logged operation that119 * might invalidate buffers, such as relation truncation. In this case, we need120 * to ensure that any buffers which were invalidated and thus not flushed by121 * the checkpoint are actually destroyed on disk. Replay can cope with a file122 * or block that doesn't exist, but not with a block that has the wrong123 * contents.124 */125#define DELAY_CHKPT_START		(1<<0)126#define DELAY_CHKPT_COMPLETE	(1<<1)127 128typedef enum129{130	PROC_WAIT_STATUS_OK,131	PROC_WAIT_STATUS_WAITING,132	PROC_WAIT_STATUS_ERROR,133} ProcWaitStatus;134 135/*136 * Each backend has a PGPROC struct in shared memory.  There is also a list of137 * currently-unused PGPROC structs that will be reallocated to new backends.138 *139 * links: list link for any list the PGPROC is in.  When waiting for a lock,140 * the PGPROC is linked into that lock's waitProcs queue.  A recycled PGPROC141 * is linked into ProcGlobal's freeProcs list.142 *143 * Note: twophase.c also sets up a dummy PGPROC struct for each currently144 * prepared transaction.  These PGPROCs appear in the ProcArray data structure145 * so that the prepared transactions appear to be still running and are146 * correctly shown as holding locks.  A prepared transaction PGPROC can be147 * distinguished from a real one at need by the fact that it has pid == 0.148 * The semaphore and lock-activity fields in a prepared-xact PGPROC are unused,149 * but its myProcLocks[] lists are valid.150 *151 * We allow many fields of this struct to be accessed without locks, such as152 * delayChkptFlags and isBackgroundWorker. However, keep in mind that writing153 * mirrored ones (see below) requires holding ProcArrayLock or XidGenLock in154 * at least shared mode, so that pgxactoff does not change concurrently.155 *156 * Mirrored fields:157 *158 * Some fields in PGPROC (see "mirrored in ..." comment) are mirrored into an159 * element of more densely packed ProcGlobal arrays. These arrays are indexed160 * by PGPROC->pgxactoff. Both copies need to be maintained coherently.161 *162 * NB: The pgxactoff indexed value can *never* be accessed without holding163 * locks.164 *165 * See PROC_HDR for details.166 */167struct PGPROC168{169	/* proc->links MUST BE FIRST IN STRUCT (see ProcSleep,ProcWakeup,etc) */170	dlist_node	links;			/* list link if process is in a list */171	dlist_head *procgloballist; /* procglobal list that owns this PGPROC */172 173	PGSemaphore sem;			/* ONE semaphore to sleep on */174	ProcWaitStatus waitStatus;175 176	Latch		procLatch;		/* generic latch for process */177 178 179	TransactionId xid;			/* id of top-level transaction currently being180								 * executed by this proc, if running and XID181								 * is assigned; else InvalidTransactionId.182								 * mirrored in ProcGlobal->xids[pgxactoff] */183 184	TransactionId xmin;			/* minimal running XID as it was when we were185								 * starting our xact, excluding LAZY VACUUM:186								 * vacuum must not remove tuples deleted by187								 * xid >= xmin ! */188 189	LocalTransactionId lxid;	/* local id of top-level transaction currently190								 * being executed by this proc, if running;191								 * else InvalidLocalTransactionId */192	int			pid;			/* Backend's process ID; 0 if prepared xact */193 194	int			pgxactoff;		/* offset into various ProcGlobal->arrays with195								 * data mirrored from this PGPROC */196 197	int			pgprocno;		/* Number of this PGPROC in198								 * ProcGlobal->allProcs array. This is set199								 * once by InitProcGlobal().200								 * ProcGlobal->allProcs[n].pgprocno == n */201 202	/* These fields are zero while a backend is still starting up: */203	BackendId	backendId;		/* This backend's backend ID (if assigned) */204	Oid			databaseId;		/* OID of database this backend is using */205	Oid			roleId;			/* OID of role using this backend */206 207	Oid			tempNamespaceId;	/* OID of temp schema this backend is208									 * using */209 210	bool		isBackgroundWorker; /* true if background worker. */211 212	/*213	 * While in hot standby mode, shows that a conflict signal has been sent214	 * for the current transaction. Set/cleared while holding ProcArrayLock,215	 * though not required. Accessed without lock, if needed.216	 */217	bool		recoveryConflictPending;218 219	/* Info about LWLock the process is currently waiting for, if any. */220	uint8		lwWaiting;		/* see LWLockWaitState */221	uint8		lwWaitMode;		/* lwlock mode being waited for */222	proclist_node lwWaitLink;	/* position in LW lock wait list */223 224	/* Support for condition variables. */225	proclist_node cvWaitLink;	/* position in CV wait list */226 227	/* Info about lock the process is currently waiting for, if any. */228	/* waitLock and waitProcLock are NULL if not currently waiting. */229	LOCK	   *waitLock;		/* Lock object we're sleeping on ... */230	PROCLOCK   *waitProcLock;	/* Per-holder info for awaited lock */231	LOCKMODE	waitLockMode;	/* type of lock we're waiting for */232	LOCKMASK	heldLocks;		/* bitmask for lock types already held on this233								 * lock object by this backend */234	pg_atomic_uint64 waitStart; /* time at which wait for lock acquisition235								 * started */236 237	int			delayChkptFlags;	/* for DELAY_CHKPT_* flags */238 239	uint8		statusFlags;	/* this backend's status flags, see PROC_*240								 * above. mirrored in241								 * ProcGlobal->statusFlags[pgxactoff] */242 243	/*244	 * Info to allow us to wait for synchronous replication, if needed.245	 * waitLSN is InvalidXLogRecPtr if not waiting; set only by user backend.246	 * syncRepState must not be touched except by owning process or WALSender.247	 * syncRepLinks used only while holding SyncRepLock.248	 */249	XLogRecPtr	waitLSN;		/* waiting for this LSN or higher */250	int			syncRepState;	/* wait state for sync rep */251	dlist_node	syncRepLinks;	/* list link if process is in syncrep queue */252 253	/*254	 * All PROCLOCK objects for locks held or awaited by this backend are255	 * linked into one of these lists, according to the partition number of256	 * their lock.257	 */258	dlist_head	myProcLocks[NUM_LOCK_PARTITIONS];259 260	XidCacheStatus subxidStatus;	/* mirrored with261									 * ProcGlobal->subxidStates[i] */262	struct XidCache subxids;	/* cache for subtransaction XIDs */263 264	/* Support for group XID clearing. */265	/* true, if member of ProcArray group waiting for XID clear */266	bool		procArrayGroupMember;267	/* next ProcArray group member waiting for XID clear */268	pg_atomic_uint32 procArrayGroupNext;269 270	/*271	 * latest transaction id among the transaction's main XID and272	 * subtransactions273	 */274	TransactionId procArrayGroupMemberXid;275 276	uint32		wait_event_info;	/* proc's wait information */277 278	/* Support for group transaction status update. */279	bool		clogGroupMember;	/* true, if member of clog group */280	pg_atomic_uint32 clogGroupNext; /* next clog group member */281	TransactionId clogGroupMemberXid;	/* transaction id of clog group member */282	XidStatus	clogGroupMemberXidStatus;	/* transaction status of clog283											 * group member */284	int			clogGroupMemberPage;	/* clog page corresponding to285										 * transaction id of clog group member */286	XLogRecPtr	clogGroupMemberLsn; /* WAL location of commit record for clog287									 * group member */288 289	/* Lock manager data, recording fast-path locks taken by this backend. */290	LWLock		fpInfoLock;		/* protects per-backend fast-path state */291	uint64		fpLockBits;		/* lock modes held for each fast-path slot */292	Oid			fpRelId[FP_LOCK_SLOTS_PER_BACKEND]; /* slots for rel oids */293	bool		fpVXIDLock;		/* are we holding a fast-path VXID lock? */294	LocalTransactionId fpLocalTransactionId;	/* lxid for fast-path VXID295												 * lock */296 297	/*298	 * Support for lock groups.  Use LockHashPartitionLockByProc on the group299	 * leader to get the LWLock protecting these fields.300	 */301	PGPROC	   *lockGroupLeader;	/* lock group leader, if I'm a member */302	dlist_head	lockGroupMembers;	/* list of members, if I'm a leader */303	dlist_node	lockGroupLink;	/* my member link, if I'm a member */304};305 306/* NOTE: "typedef struct PGPROC PGPROC" appears in storage/lock.h. */307 308 309extern PGDLLIMPORT PGPROC *MyProc;310 311/*312 * There is one ProcGlobal struct for the whole database cluster.313 *314 * Adding/Removing an entry into the procarray requires holding *both*315 * ProcArrayLock and XidGenLock in exclusive mode (in that order). Both are316 * needed because the dense arrays (see below) are accessed from317 * GetNewTransactionId() and GetSnapshotData(), and we don't want to add318 * further contention by both using the same lock. Adding/Removing a procarray319 * entry is much less frequent.320 *321 * Some fields in PGPROC are mirrored into more densely packed arrays (e.g.322 * xids), with one entry for each backend. These arrays only contain entries323 * for PGPROCs that have been added to the shared array with ProcArrayAdd()324 * (in contrast to PGPROC array which has unused PGPROCs interspersed).325 *326 * The dense arrays are indexed by PGPROC->pgxactoff. Any concurrent327 * ProcArrayAdd() / ProcArrayRemove() can lead to pgxactoff of a procarray328 * member to change.  Therefore it is only safe to use PGPROC->pgxactoff to329 * access the dense array while holding either ProcArrayLock or XidGenLock.330 *331 * As long as a PGPROC is in the procarray, the mirrored values need to be332 * maintained in both places in a coherent manner.333 *334 * The denser separate arrays are beneficial for three main reasons: First, to335 * allow for as tight loops accessing the data as possible. Second, to prevent336 * updates of frequently changing data (e.g. xmin) from invalidating337 * cachelines also containing less frequently changing data (e.g. xid,338 * statusFlags). Third to condense frequently accessed data into as few339 * cachelines as possible.340 *341 * There are two main reasons to have the data mirrored between these dense342 * arrays and PGPROC. First, as explained above, a PGPROC's array entries can343 * only be accessed with either ProcArrayLock or XidGenLock held, whereas the344 * PGPROC entries do not require that (obviously there may still be locking345 * requirements around the individual field, separate from the concerns346 * here). That is particularly important for a backend to efficiently checks347 * it own values, which it often can safely do without locking.  Second, the348 * PGPROC fields allow to avoid unnecessary accesses and modification to the349 * dense arrays. A backend's own PGPROC is more likely to be in a local cache,350 * whereas the cachelines for the dense array will be modified by other351 * backends (often removing it from the cache for other cores/sockets). At352 * commit/abort time a check of the PGPROC value can avoid accessing/dirtying353 * the corresponding array value.354 *355 * Basically it makes sense to access the PGPROC variable when checking a356 * single backend's data, especially when already looking at the PGPROC for357 * other reasons already.  It makes sense to look at the "dense" arrays if we358 * need to look at many / most entries, because we then benefit from the359 * reduced indirection and better cross-process cache-ability.360 *361 * When entering a PGPROC for 2PC transactions with ProcArrayAdd(), the data362 * in the dense arrays is initialized from the PGPROC while it already holds363 * ProcArrayLock.364 */365typedef struct PROC_HDR366{367	/* Array of PGPROC structures (not including dummies for prepared txns) */368	PGPROC	   *allProcs;369 370	/* Array mirroring PGPROC.xid for each PGPROC currently in the procarray */371	TransactionId *xids;372 373	/*374	 * Array mirroring PGPROC.subxidStatus for each PGPROC currently in the375	 * procarray.376	 */377	XidCacheStatus *subxidStates;378 379	/*380	 * Array mirroring PGPROC.statusFlags for each PGPROC currently in the381	 * procarray.382	 */383	uint8	   *statusFlags;384 385	/* Length of allProcs array */386	uint32		allProcCount;387	/* Head of list of free PGPROC structures */388	dlist_head	freeProcs;389	/* Head of list of autovacuum's free PGPROC structures */390	dlist_head	autovacFreeProcs;391	/* Head of list of bgworker free PGPROC structures */392	dlist_head	bgworkerFreeProcs;393	/* Head of list of walsender free PGPROC structures */394	dlist_head	walsenderFreeProcs;395	/* First pgproc waiting for group XID clear */396	pg_atomic_uint32 procArrayGroupFirst;397	/* First pgproc waiting for group transaction status update */398	pg_atomic_uint32 clogGroupFirst;399	/* WALWriter process's latch */400	Latch	   *walwriterLatch;401	/* Checkpointer process's latch */402	Latch	   *checkpointerLatch;403	/* Current shared estimate of appropriate spins_per_delay value */404	int			spins_per_delay;405	/* Buffer id of the buffer that Startup process waits for pin on, or -1 */406	int			startupBufferPinWaitBufId;407} PROC_HDR;408 409extern PGDLLIMPORT PROC_HDR *ProcGlobal;410 411extern PGDLLIMPORT PGPROC *PreparedXactProcs;412 413/* Accessor for PGPROC given a pgprocno. */414#define GetPGProcByNumber(n) (&ProcGlobal->allProcs[(n)])415 416/*417 * We set aside some extra PGPROC structures for auxiliary processes,418 * ie things that aren't full-fledged backends but need shmem access.419 *420 * Background writer, checkpointer, WAL writer and archiver run during normal421 * operation.  Startup process and WAL receiver also consume 2 slots, but WAL422 * writer is launched only after startup has exited, so we only need 5 slots.423 */424#define NUM_AUXILIARY_PROCS		5425 426/* configurable options */427extern PGDLLIMPORT int DeadlockTimeout;428extern PGDLLIMPORT int StatementTimeout;429extern PGDLLIMPORT int LockTimeout;430extern PGDLLIMPORT int IdleInTransactionSessionTimeout;431extern PGDLLIMPORT int IdleSessionTimeout;432extern PGDLLIMPORT bool log_lock_waits;433 434 435/*436 * Function Prototypes437 */438extern int	ProcGlobalSemas(void);439extern Size ProcGlobalShmemSize(void);440extern void InitProcGlobal(void);441extern void InitProcess(void);442extern void InitProcessPhase2(void);443extern void InitAuxiliaryProcess(void);444 445extern void SetStartupBufferPinWaitBufId(int bufid);446extern int	GetStartupBufferPinWaitBufId(void);447 448extern bool HaveNFreeProcs(int n, int *nfree);449extern void ProcReleaseLocks(bool isCommit);450 451extern ProcWaitStatus ProcSleep(LOCALLOCK *locallock, LockMethod lockMethodTable);452extern void ProcWakeup(PGPROC *proc, ProcWaitStatus waitStatus);453extern void ProcLockWakeup(LockMethod lockMethodTable, LOCK *lock);454extern void CheckDeadLockAlert(void);455extern bool IsWaitingForLock(void);456extern void LockErrorCleanup(void);457 458extern void ProcWaitForSignal(uint32 wait_event_info);459extern void ProcSendSignal(int pgprocno);460 461extern PGPROC *AuxiliaryPidGetProc(int pid);462 463extern void BecomeLockGroupLeader(void);464extern bool BecomeLockGroupMember(PGPROC *leader, int pid);465 466#endif							/* _PROC_H_ */467