database/postgres
database/postgres
1
0
Fork 0
mirror of https://github.com/postgres/postgres.git synced 2025-04-24 10:47:04 +03:00
Code

Divide the lock manager's shared state into 'partitions', so as to

Browse Source 
reduce contention for the former single LockMgrLock.  Per my recent
proposal.  I set it up for 16 partitions, but on a pgbench test this
gives only a marginal further improvement over 4 partitions --- we need
to test more scenarios to choose the number of partitions.
This commit is contained in:
Tom Lane 2005-12-11 21:02:18 +00:00
parent be8100d64e
commit ec0baf949e
10 changed files with 643 additions and 415 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
src/backend/access/transam/twophase.c
View File

@ -7,7 +7,7 @@
* Portions Copyright (c) 1994, Regents of the University of California * Portions Copyright (c) 1994, Regents of the University of California
* *
* IDENTIFICATION * IDENTIFICATION
* $PostgreSQL: pgsql/src/backend/access/transam/twophase.c,v 1.17 2005/11/22 18:17:07 momjian Exp $ * $PostgreSQL: pgsql/src/backend/access/transam/twophase.c,v 1.18 2005/12/11 21:02:17 tgl Exp $
* *
* NOTES * NOTES
* Each global transaction is associated with a global transaction * Each global transaction is associated with a global transaction
@ -284,7 +284,8 @@ MarkAsPreparing(TransactionId xid, const char *gid,
gxact->proc.lwWaitLink = NULL; gxact->proc.lwWaitLink = NULL;
gxact->proc.waitLock = NULL; gxact->proc.waitLock = NULL;
gxact->proc.waitProcLock = NULL; gxact->proc.waitProcLock = NULL;
SHMQueueInit(&(gxact->proc.procLocks)); for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
SHMQueueInit(&(gxact->proc.myProcLocks[i]));
/* subxid data must be filled later by GXactLoadSubxactData */ /* subxid data must be filled later by GXactLoadSubxactData */
gxact->proc.subxids.overflowed = false; gxact->proc.subxids.overflowed = false;
gxact->proc.subxids.nxids = 0; gxact->proc.subxids.nxids = 0;

6
src/backend/storage/ipc/procarray.c
View File

@ -14,8 +14,8 @@
* *
* The process array now also includes PGPROC structures representing * The process array now also includes PGPROC structures representing
* prepared transactions. The xid and subxids fields of these are valid, * prepared transactions. The xid and subxids fields of these are valid,
* as is the procLocks list. They can be distinguished from regular backend * as are the myProcLocks lists. They can be distinguished from regular
* PGPROCs at need by checking for pid == 0. * backend PGPROCs at need by checking for pid == 0.
* *
* *
* Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group * Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group
@ -23,7 +23,7 @@
* *
* *
* IDENTIFICATION * IDENTIFICATION
* $PostgreSQL: pgsql/src/backend/storage/ipc/procarray.c,v 1.8 2005/11/22 18:17:20 momjian Exp $ * $PostgreSQL: pgsql/src/backend/storage/ipc/procarray.c,v 1.9 2005/12/11 21:02:18 tgl Exp $
* *
*------------------------------------------------------------------------- *-------------------------------------------------------------------------
*/ */

89
src/backend/storage/lmgr/README
View File

@ -1,4 +1,4 @@
$PostgreSQL: pgsql/src/backend/storage/lmgr/README,v 1.18 2005/12/09 01:22:04 tgl Exp $ $PostgreSQL: pgsql/src/backend/storage/lmgr/README,v 1.19 2005/12/11 21:02:18 tgl Exp $
LOCKING OVERVIEW LOCKING OVERVIEW
@ -51,8 +51,11 @@ Lock methods describe the overall locking behavior. Currently there are
two lock methods: DEFAULT and USER. (USER locks are non-blocking.) two lock methods: DEFAULT and USER. (USER locks are non-blocking.)
Lock modes describe the type of the lock (read/write or shared/exclusive). Lock modes describe the type of the lock (read/write or shared/exclusive).
See src/tools/backend/index.html and src/include/storage/lock.h for more In principle, each lock method can have its own set of lock modes with
details. different conflict rules, but currently DEFAULT and USER methods use
identical lock mode sets. See src/tools/backend/index.html and
src/include/storage/lock.h for more details. (Lock modes are also called
lock types in some places in the code and documentation.)
There are two fundamental lock structures in shared memory: the There are two fundamental lock structures in shared memory: the
per-lockable-object LOCK struct, and the per-lock-and-requestor PROCLOCK per-lockable-object LOCK struct, and the per-lock-and-requestor PROCLOCK
@ -67,7 +70,7 @@ be made per lockable object/lock mode/backend. Internally to a backend,
however, the same lock may be requested and perhaps released multiple times however, the same lock may be requested and perhaps released multiple times
in a transaction, and it can also be held both transactionally and session- in a transaction, and it can also be held both transactionally and session-
wide. The internal request counts are held in LOCALLOCK so that the shared wide. The internal request counts are held in LOCALLOCK so that the shared
LockMgrLock need not be obtained to alter them. data structures need not be accessed to alter them.
--------------------------------------------------------------------------- ---------------------------------------------------------------------------
@ -103,10 +106,10 @@ procLocks -
be waiting for more!). be waiting for more!).
waitProcs - waitProcs -
This is a shared memory queue of all process structures corresponding to This is a shared memory queue of all PGPROC structures corresponding to
a backend that is waiting (sleeping) until another backend releases this backends that are waiting (sleeping) until another backend releases this
lock. The process structure holds the information needed to determine lock. The process structure holds the information needed to determine
if it should be woken up when this lock is released. if it should be woken up when the lock is released.
nRequested - nRequested -
Keeps a count of how many times this lock has been attempted to be Keeps a count of how many times this lock has been attempted to be
@ -131,12 +134,12 @@ nGranted -
granted - granted -
Keeps count of how many locks of each type are currently held. Once again Keeps count of how many locks of each type are currently held. Once again
only elements 1 through MAX_LOCKMODES-1 are used (0 is not). Also, like only elements 1 through MAX_LOCKMODES-1 are used (0 is not). Also, like
requested, summing the values of granted should total to the value requested[], summing the values of granted[] should total to the value
of nGranted. of nGranted.
We should always have 0 <= nGranted <= nRequested, and We should always have 0 <= nGranted <= nRequested, and
0 <= granted[i] <= requested[i] for each i. If the request counts go to 0 <= granted[i] <= requested[i] for each i. When all the request counts
zero, the lock object is no longer needed and can be freed. go to zero, the LOCK object is no longer needed and can be freed.
--------------------------------------------------------------------------- ---------------------------------------------------------------------------
@ -154,15 +157,16 @@ tag -
SHMEM offset of PGPROC of backend process that owns this PROCLOCK. SHMEM offset of PGPROC of backend process that owns this PROCLOCK.
holdMask - holdMask -
A bitmask for the lock types successfully acquired by this PROCLOCK. A bitmask for the lock modes successfully acquired by this PROCLOCK.
This should be a subset of the LOCK object's grantMask, and also a This should be a subset of the LOCK object's grantMask, and also a
subset of the PGPROC object's heldLocks mask. subset of the PGPROC object's heldLocks mask (if the PGPROC is
currently waiting for another lock mode on this lock).
releaseMask - releaseMask -
A bitmask for the lock types due to be released during LockReleaseAll. A bitmask for the lock modes due to be released during LockReleaseAll.
This must be a subset of the holdMask. Note that it is modified without This must be a subset of the holdMask. Note that it is modified without
taking the LockMgrLock, and therefore it is unsafe for any backend except taking the partition LWLock, and therefore it is unsafe for any
the one owning the PROCLOCK to examine/change it. backend except the one owning the PROCLOCK to examine/change it.
lockLink - lockLink -
List link for shared memory queue of all the PROCLOCK objects for the List link for shared memory queue of all the PROCLOCK objects for the
@ -174,7 +178,60 @@ procLink -
--------------------------------------------------------------------------- ---------------------------------------------------------------------------
The deadlock detection algorithm:
LOCK MANAGER INTERNAL LOCKING
Before PostgreSQL 8.2, all of the shared-memory data structures used by
the lock manager were protected by a single LWLock, the LockMgrLock;
any operation involving these data structures had to exclusively lock
LockMgrLock. Not too surprisingly, this became a contention bottleneck.
To reduce contention, the lock manager's data structures have been split
into multiple "partitions", each protected by an independent LWLock.
Most operations only need to lock the single partition they are working in.
Here are the details:
* Each possible lock is assigned to one partition according to a hash of
its LOCKTAG value (see LockTagToPartition()). The partition's LWLock is
considered to protect all the LOCK objects of that partition as well as
their subsidiary PROCLOCKs. The shared-memory hash tables for LOCKs and
PROCLOCKs are divided into separate hash tables for each partition, and
operations on each hash table are likewise protected by the partition
lock.
* Formerly, each PGPROC had a single list of PROCLOCKs belonging to it.
This has now been split into per-partition lists, so that access to a
particular PROCLOCK list can be protected by the associated partition's
LWLock. (This is not strictly necessary at the moment, because at this
writing a PGPROC's PROCLOCK list is only accessed by the owning backend
anyway. But it seems forward-looking to maintain a convention for how
other backends could access it. In any case LockReleaseAll needs to be
able to quickly determine which partition each LOCK belongs to, and
for the currently contemplated number of partitions, this way takes less
shared memory than explicitly storing a partition number in LOCK structs
would require.)
* The other lock-related fields of a PGPROC are only interesting when
the PGPROC is waiting for a lock, so we consider that they are protected
by the partition LWLock of the awaited lock.
For normal lock acquisition and release, it is sufficient to lock the
partition containing the desired lock. Deadlock checking needs to touch
multiple partitions in general; for simplicity, we just make it lock all
the partitions in partition-number order. (To prevent LWLock deadlock,
we establish the rule that any backend needing to lock more than one
partition at once must lock them in partition-number order.) It's
possible that deadlock checking could be done without touching every
partition in typical cases, but since in a properly functioning system
deadlock checking should not occur often enough to be performance-critical,
trying to make this work does not seem a productive use of effort.
A backend's internal LOCALLOCK hash table is not partitioned. We do store
the partition number in LOCALLOCK table entries, but this is a straight
speed-for-space tradeoff: we could instead recalculate the partition
number from the LOCKTAG when needed.
THE DEADLOCK DETECTION ALGORITHM
Since we allow user transactions to request locks in any order, deadlock Since we allow user transactions to request locks in any order, deadlock
is possible. We use a deadlock detection/breaking algorithm that is is possible. We use a deadlock detection/breaking algorithm that is

20
src/backend/storage/lmgr/deadlock.c
View File

@ -12,7 +12,7 @@
* *
* *
* IDENTIFICATION * IDENTIFICATION
* $PostgreSQL: pgsql/src/backend/storage/lmgr/deadlock.c,v 1.37 2005/12/09 01:22:04 tgl Exp $ * $PostgreSQL: pgsql/src/backend/storage/lmgr/deadlock.c,v 1.38 2005/12/11 21:02:18 tgl Exp $
* *
* Interface: * Interface:
* *
@ -53,9 +53,9 @@ typedef struct
* Information saved about each edge in a detected deadlock cycle. This * Information saved about each edge in a detected deadlock cycle. This
* is used to print a diagnostic message upon failure. * is used to print a diagnostic message upon failure.
* *
* Note: because we want to examine this info after releasing the LockMgrLock, * Note: because we want to examine this info after releasing the lock
* we can't just store LOCK and PGPROC pointers; we must extract out all the * manager's partition locks, we can't just store LOCK and PGPROC pointers;
* info we want to be able to print. * we must extract out all the info we want to be able to print.
*/ */
typedef struct typedef struct
{ {
@ -188,19 +188,11 @@ InitDeadLockChecking(void)
* deadlock. If resolution is impossible, return TRUE --- the caller * deadlock. If resolution is impossible, return TRUE --- the caller
* is then expected to abort the given proc's transaction. * is then expected to abort the given proc's transaction.
* *
* We can't block on user locks, so no sense testing for deadlock * Caller must already have locked all partitions of the lock tables.
* because there is no blocking, and no timer for the block. So,
* only look at regular locks.
*
* We must have already locked the master lock before being called.
* NOTE: although the lockmethod structure appears to allow each lock
* table to have a different masterLock, all locks that can block had
* better use the same LWLock, else this code will not be adequately
* interlocked!
* *
* On failure, deadlock details are recorded in deadlockDetails[] for * On failure, deadlock details are recorded in deadlockDetails[] for
* subsequent printing by DeadLockReport(). That activity is separate * subsequent printing by DeadLockReport(). That activity is separate
* because we don't want to do it while holding the master lock. * because we don't want to do it while holding all those LWLocks.
*/ */
bool bool
DeadLockCheck(PGPROC *proc) DeadLockCheck(PGPROC *proc)

425
src/backend/storage/lmgr/lock.c
View File

@ -1,14 +1,14 @@
/*------------------------------------------------------------------------- /*-------------------------------------------------------------------------
* *
* lock.c * lock.c
* POSTGRES low-level lock mechanism * POSTGRES primary lock mechanism
* *
* Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group * Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group
* Portions Copyright (c) 1994, Regents of the University of California * Portions Copyright (c) 1994, Regents of the University of California
* *
* *
* IDENTIFICATION * IDENTIFICATION
* $PostgreSQL: pgsql/src/backend/storage/lmgr/lock.c,v 1.161 2005/12/09 01:22:04 tgl Exp $ * $PostgreSQL: pgsql/src/backend/storage/lmgr/lock.c,v 1.162 2005/12/11 21:02:18 tgl Exp $
* *
* NOTES * NOTES
* A lock table is a shared memory hash table. When * A lock table is a shared memory hash table. When
@ -163,10 +163,13 @@ typedef struct TwoPhaseLockRecord
/* /*
* Links to hash tables containing lock state * Pointers to hash tables containing lock state
*
* The LockMethodLockHash and LockMethodProcLockHash hash tables are in
* shared memory; LockMethodLocalHash is local to each backend.
*/ */
static HTAB *LockMethodLockHash; static HTAB *LockMethodLockHash[NUM_LOCK_PARTITIONS];
static HTAB *LockMethodProcLockHash; static HTAB *LockMethodProcLockHash[NUM_LOCK_PARTITIONS];
static HTAB *LockMethodLocalHash; static HTAB *LockMethodLocalHash;
@ -255,16 +258,25 @@ PROCLOCK_PRINT(const char *where, const PROCLOCK *proclockP)
static void RemoveLocalLock(LOCALLOCK *locallock); static void RemoveLocalLock(LOCALLOCK *locallock);
static void GrantLockLocal(LOCALLOCK *locallock, ResourceOwner owner); static void GrantLockLocal(LOCALLOCK *locallock, ResourceOwner owner);
static void WaitOnLock(LOCKMETHODID lockmethodid, LOCALLOCK *locallock, static void WaitOnLock(LOCALLOCK *locallock, ResourceOwner owner);
ResourceOwner owner);
static bool UnGrantLock(LOCK *lock, LOCKMODE lockmode, static bool UnGrantLock(LOCK *lock, LOCKMODE lockmode,
PROCLOCK *proclock, LockMethod lockMethodTable); PROCLOCK *proclock, LockMethod lockMethodTable);
static void CleanUpLock(LOCKMETHODID lockmethodid, LOCK *lock, static void CleanUpLock(LOCK *lock, PROCLOCK *proclock,
PROCLOCK *proclock, bool wakeupNeeded); LockMethod lockMethodTable, int partition,
bool wakeupNeeded);
/* /*
* InitLocks -- Initialize the lock module's shared memory. * InitLocks -- Initialize the lock manager's data structures.
*
* This is called from CreateSharedMemoryAndSemaphores(), which see for
* more comments. In the normal postmaster case, the shared hash tables
* are created here, as well as a locallock hash table that will remain
* unused and empty in the postmaster itself. Backends inherit the pointers
* to the shared tables via fork(), and also inherit an image of the locallock
* hash table, which they proceed to use. In the EXEC_BACKEND case, each
* backend re-executes this code to obtain pointers to the already existing
* shared hash tables and to create its locallock hash table.
*/ */
void void
InitLocks(void) InitLocks(void)
@ -274,13 +286,18 @@ InitLocks(void)
int hash_flags; int hash_flags;
long init_table_size, long init_table_size,
max_table_size; max_table_size;
int i;
/* Compute init/max size to request for lock hashtables */ /*
* Compute init/max size to request for lock hashtables. Note these
* calculations must agree with LockShmemSize!
*/
max_table_size = NLOCKENTS(); max_table_size = NLOCKENTS();
max_table_size = (max_table_size - 1) / NUM_LOCK_PARTITIONS + 1;
init_table_size = max_table_size / 2; init_table_size = max_table_size / 2;
/* /*
* allocate a hash table for LOCK structs. This is used to store * Allocate hash tables for LOCK structs. These are used to store
* per-locked-object information. * per-locked-object information.
*/ */
MemSet(&info, 0, sizeof(info)); MemSet(&info, 0, sizeof(info));
@ -289,37 +306,45 @@ InitLocks(void)
info.hash = tag_hash; info.hash = tag_hash;
hash_flags = (HASH_ELEM | HASH_FUNCTION); hash_flags = (HASH_ELEM | HASH_FUNCTION);
sprintf(shmemName, "LOCK hash"); for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
LockMethodLockHash = ShmemInitHash(shmemName, {
sprintf(shmemName, "LOCK hash %d", i);
LockMethodLockHash[i] = ShmemInitHash(shmemName,
init_table_size, init_table_size,
max_table_size, max_table_size,
&info, &info,
hash_flags); hash_flags);
if (!LockMethodLockHash[i])
if (!LockMethodLockHash)
elog(FATAL, "could not initialize lock table \"%s\"", shmemName); elog(FATAL, "could not initialize lock table \"%s\"", shmemName);
}
/* Assume an average of 2 holders per lock */
max_table_size *= 2;
init_table_size *= 2;
/* /*
* allocate a hash table for PROCLOCK structs. This is used to store * Allocate hash tables for PROCLOCK structs. These are used to store
* per-lock-holder information. * per-lock-per-holder information.
*/ */
info.keysize = sizeof(PROCLOCKTAG); info.keysize = sizeof(PROCLOCKTAG);
info.entrysize = sizeof(PROCLOCK); info.entrysize = sizeof(PROCLOCK);
info.hash = tag_hash; info.hash = tag_hash;
hash_flags = (HASH_ELEM | HASH_FUNCTION); hash_flags = (HASH_ELEM | HASH_FUNCTION);
sprintf(shmemName, "PROCLOCK hash"); for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
LockMethodProcLockHash = ShmemInitHash(shmemName, {
sprintf(shmemName, "PROCLOCK hash %d", i);
LockMethodProcLockHash[i] = ShmemInitHash(shmemName,
init_table_size, init_table_size,
max_table_size, max_table_size,
&info, &info,
hash_flags); hash_flags);
if (!LockMethodProcLockHash[i])
if (!LockMethodProcLockHash)
elog(FATAL, "could not initialize lock table \"%s\"", shmemName); elog(FATAL, "could not initialize lock table \"%s\"", shmemName);
}
/* /*
* allocate a non-shared hash table for LOCALLOCK structs. This is used * Allocate one non-shared hash table for LOCALLOCK structs. This is used
* to store lock counts and resource owner information. * to store lock counts and resource owner information.
* *
* The non-shared table could already exist in this process (this occurs * The non-shared table could already exist in this process (this occurs
@ -355,6 +380,39 @@ GetLocksMethodTable(const LOCK *lock)
} }
/*
* Given a LOCKTAG, determine which partition the lock belongs in.
*
* Basically what we want to do here is hash the locktag. However, it
* seems unwise to use hash_any() because that is the same function that
* will be used to distribute the locks within each partition's hash table;
* if we use it, we run a big risk of having uneven distribution of hash
* codes within each hash table. Instead, we use a simple linear XOR of the
* bits of the locktag.
*/
int
LockTagToPartition(const LOCKTAG *locktag)
{
const uint8 *ptr = (const uint8 *) locktag;
int result = 0;
int i;
for (i = 0; i < sizeof(LOCKTAG); i++)
result ^= *ptr++;
#if NUM_LOCK_PARTITIONS == 16
result ^= result >> 4;
result &= 0x0F;
#elif NUM_LOCK_PARTITIONS == 4
result ^= result >> 4;
result ^= result >> 2;
result &= 0x03;
#else
#error unsupported NUM_LOCK_PARTITIONS
#endif
return result;
}
/* /*
* LockAcquire -- Check for lock conflicts, sleep if conflict found, * LockAcquire -- Check for lock conflicts, sleep if conflict found,
* set lock if/when no conflicts. * set lock if/when no conflicts.
@ -397,7 +455,8 @@ LockAcquire(const LOCKTAG *locktag,
PROCLOCKTAG proclocktag; PROCLOCKTAG proclocktag;
bool found; bool found;
ResourceOwner owner; ResourceOwner owner;
LWLockId masterLock; int partition;
LWLockId partitionLock;
int status; int status;
if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods)) if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
@ -438,6 +497,7 @@ LockAcquire(const LOCKTAG *locktag,
locallock->lock = NULL; locallock->lock = NULL;
locallock->proclock = NULL; locallock->proclock = NULL;
locallock->isTempObject = isTempObject; locallock->isTempObject = isTempObject;
locallock->partition = LockTagToPartition(&(localtag.lock));
locallock->nLocks = 0; locallock->nLocks = 0;
locallock->numLockOwners = 0; locallock->numLockOwners = 0;
locallock->maxLockOwners = 8; locallock->maxLockOwners = 8;
@ -474,9 +534,10 @@ LockAcquire(const LOCKTAG *locktag,
/* /*
* Otherwise we've got to mess with the shared lock table. * Otherwise we've got to mess with the shared lock table.
*/ */
masterLock = LockMgrLock; partition = locallock->partition;
partitionLock = FirstLockMgrLock + partition;
LWLockAcquire(masterLock, LW_EXCLUSIVE); LWLockAcquire(partitionLock, LW_EXCLUSIVE);
/* /*
* Find or create a lock with this tag. * Find or create a lock with this tag.
@ -486,12 +547,12 @@ LockAcquire(const LOCKTAG *locktag,
* pointer is valid, since a lock object with no locks can go away * pointer is valid, since a lock object with no locks can go away
* anytime. * anytime.
*/ */
lock = (LOCK *) hash_search(LockMethodLockHash, lock = (LOCK *) hash_search(LockMethodLockHash[partition],
(void *) locktag, (void *) locktag,
HASH_ENTER_NULL, &found); HASH_ENTER_NULL, &found);
if (!lock) if (!lock)
{ {
LWLockRelease(masterLock); LWLockRelease(partitionLock);
ereport(ERROR, ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY), (errcode(ERRCODE_OUT_OF_MEMORY),
errmsg("out of shared memory"), errmsg("out of shared memory"),
@ -532,7 +593,7 @@ LockAcquire(const LOCKTAG *locktag,
/* /*
* Find or create a proclock entry with this tag * Find or create a proclock entry with this tag
*/ */
proclock = (PROCLOCK *) hash_search(LockMethodProcLockHash, proclock = (PROCLOCK *) hash_search(LockMethodProcLockHash[partition],
(void *) &proclocktag, (void *) &proclocktag,
HASH_ENTER_NULL, &found); HASH_ENTER_NULL, &found);
if (!proclock) if (!proclock)
@ -547,12 +608,12 @@ LockAcquire(const LOCKTAG *locktag,
* anyone to release the lock object later. * anyone to release the lock object later.
*/ */
Assert(SHMQueueEmpty(&(lock->procLocks))); Assert(SHMQueueEmpty(&(lock->procLocks)));
if (!hash_search(LockMethodLockHash, if (!hash_search(LockMethodLockHash[partition],
(void *) &(lock->tag), (void *) &(lock->tag),
HASH_REMOVE, NULL)) HASH_REMOVE, NULL))
elog(PANIC, "lock table corrupted"); elog(PANIC, "lock table corrupted");
} }
LWLockRelease(masterLock); LWLockRelease(partitionLock);
ereport(ERROR, ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY), (errcode(ERRCODE_OUT_OF_MEMORY),
errmsg("out of shared memory"), errmsg("out of shared memory"),
@ -569,7 +630,8 @@ LockAcquire(const LOCKTAG *locktag,
proclock->releaseMask = 0; proclock->releaseMask = 0;
/* Add proclock to appropriate lists */ /* Add proclock to appropriate lists */
SHMQueueInsertBefore(&lock->procLocks, &proclock->lockLink); SHMQueueInsertBefore(&lock->procLocks, &proclock->lockLink);
SHMQueueInsertBefore(&MyProc->procLocks, &proclock->procLink); SHMQueueInsertBefore(&(MyProc->myProcLocks[partition]),
&proclock->procLink);
PROCLOCK_PRINT("LockAcquire: new", proclock); PROCLOCK_PRINT("LockAcquire: new", proclock);
} }
else else
@ -666,7 +728,7 @@ LockAcquire(const LOCKTAG *locktag,
{ {
SHMQueueDelete(&proclock->lockLink); SHMQueueDelete(&proclock->lockLink);
SHMQueueDelete(&proclock->procLink); SHMQueueDelete(&proclock->procLink);
if (!hash_search(LockMethodProcLockHash, if (!hash_search(LockMethodProcLockHash[partition],
(void *) &(proclock->tag), (void *) &(proclock->tag),
HASH_REMOVE, NULL)) HASH_REMOVE, NULL))
elog(PANIC, "proclock table corrupted"); elog(PANIC, "proclock table corrupted");
@ -678,7 +740,7 @@ LockAcquire(const LOCKTAG *locktag,
LOCK_PRINT("LockAcquire: conditional lock failed", lock, lockmode); LOCK_PRINT("LockAcquire: conditional lock failed", lock, lockmode);
Assert((lock->nRequested > 0) && (lock->requested[lockmode] >= 0)); Assert((lock->nRequested > 0) && (lock->requested[lockmode] >= 0));
Assert(lock->nGranted <= lock->nRequested); Assert(lock->nGranted <= lock->nRequested);
LWLockRelease(masterLock); LWLockRelease(partitionLock);
if (locallock->nLocks == 0) if (locallock->nLocks == 0)
RemoveLocalLock(locallock); RemoveLocalLock(locallock);
return LOCKACQUIRE_NOT_AVAIL; return LOCKACQUIRE_NOT_AVAIL;
@ -692,7 +754,7 @@ LockAcquire(const LOCKTAG *locktag,
/* /*
* Sleep till someone wakes me up. * Sleep till someone wakes me up.
*/ */
WaitOnLock(lockmethodid, locallock, owner); WaitOnLock(locallock, owner);
/* /*
* NOTE: do not do any material change of state between here and * NOTE: do not do any material change of state between here and
@ -709,14 +771,14 @@ LockAcquire(const LOCKTAG *locktag,
PROCLOCK_PRINT("LockAcquire: INCONSISTENT", proclock); PROCLOCK_PRINT("LockAcquire: INCONSISTENT", proclock);
LOCK_PRINT("LockAcquire: INCONSISTENT", lock, lockmode); LOCK_PRINT("LockAcquire: INCONSISTENT", lock, lockmode);
/* Should we retry ? */ /* Should we retry ? */
LWLockRelease(masterLock); LWLockRelease(partitionLock);
elog(ERROR, "LockAcquire failed"); elog(ERROR, "LockAcquire failed");
} }
PROCLOCK_PRINT("LockAcquire: granted", proclock); PROCLOCK_PRINT("LockAcquire: granted", proclock);
LOCK_PRINT("LockAcquire: granted", lock, lockmode); LOCK_PRINT("LockAcquire: granted", lock, lockmode);
} }
LWLockRelease(masterLock); LWLockRelease(partitionLock);
return LOCKACQUIRE_OK; return LOCKACQUIRE_OK;
} }
@ -894,11 +956,12 @@ UnGrantLock(LOCK *lock, LOCKMODE lockmode,
* should be called after UnGrantLock, and wakeupNeeded is the result from * should be called after UnGrantLock, and wakeupNeeded is the result from
* UnGrantLock.) * UnGrantLock.)
* *
* The locktable's masterLock must be held at entry, and will be * The lock table's partition lock must be held at entry, and will be
* held at exit. * held at exit.
*/ */
static void static void
CleanUpLock(LOCKMETHODID lockmethodid, LOCK *lock, PROCLOCK *proclock, CleanUpLock(LOCK *lock, PROCLOCK *proclock,
LockMethod lockMethodTable, int partition,
bool wakeupNeeded) bool wakeupNeeded)
{ {
/* /*
@ -910,7 +973,7 @@ CleanUpLock(LOCKMETHODID lockmethodid, LOCK *lock, PROCLOCK *proclock,
PROCLOCK_PRINT("CleanUpLock: deleting", proclock); PROCLOCK_PRINT("CleanUpLock: deleting", proclock);
SHMQueueDelete(&proclock->lockLink); SHMQueueDelete(&proclock->lockLink);
SHMQueueDelete(&proclock->procLink); SHMQueueDelete(&proclock->procLink);
if (!hash_search(LockMethodProcLockHash, if (!hash_search(LockMethodProcLockHash[partition],
(void *) &(proclock->tag), (void *) &(proclock->tag),
HASH_REMOVE, NULL)) HASH_REMOVE, NULL))
elog(PANIC, "proclock table corrupted"); elog(PANIC, "proclock table corrupted");
@ -924,7 +987,7 @@ CleanUpLock(LOCKMETHODID lockmethodid, LOCK *lock, PROCLOCK *proclock,
*/ */
LOCK_PRINT("CleanUpLock: deleting", lock, 0); LOCK_PRINT("CleanUpLock: deleting", lock, 0);
Assert(SHMQueueEmpty(&(lock->procLocks))); Assert(SHMQueueEmpty(&(lock->procLocks)));
if (!hash_search(LockMethodLockHash, if (!hash_search(LockMethodLockHash[partition],
(void *) &(lock->tag), (void *) &(lock->tag),
HASH_REMOVE, NULL)) HASH_REMOVE, NULL))
elog(PANIC, "lock table corrupted"); elog(PANIC, "lock table corrupted");
@ -932,7 +995,7 @@ CleanUpLock(LOCKMETHODID lockmethodid, LOCK *lock, PROCLOCK *proclock,
else if (wakeupNeeded) else if (wakeupNeeded)
{ {
/* There are waiters on this lock, so wake them up. */ /* There are waiters on this lock, so wake them up. */
ProcLockWakeup(LockMethods[lockmethodid], lock); ProcLockWakeup(lockMethodTable, lock);
} }
} }
@ -988,12 +1051,12 @@ GrantAwaitedLock(void)
* Caller must have set MyProc->heldLocks to reflect locks already held * Caller must have set MyProc->heldLocks to reflect locks already held
* on the lockable object by this process. * on the lockable object by this process.
* *
* The locktable's masterLock must be held at entry. * The appropriate partition lock must be held at entry.
*/ */
static void static void
WaitOnLock(LOCKMETHODID lockmethodid, LOCALLOCK *locallock, WaitOnLock(LOCALLOCK *locallock, ResourceOwner owner)
ResourceOwner owner)
{ {
LOCKMETHODID lockmethodid = LOCALLOCK_LOCKMETHOD(*locallock);
LockMethod lockMethodTable = LockMethods[lockmethodid]; LockMethod lockMethodTable = LockMethods[lockmethodid];
const char *old_status; const char *old_status;
char *new_status; char *new_status;
@ -1025,10 +1088,7 @@ WaitOnLock(LOCKMETHODID lockmethodid, LOCALLOCK *locallock,
* will also happen in the cancel/die case. * will also happen in the cancel/die case.
*/ */
if (ProcSleep(lockMethodTable, if (ProcSleep(locallock, lockMethodTable) != STATUS_OK)
locallock->tag.mode,
locallock->lock,
locallock->proclock) != STATUS_OK)
{ {
/* /*
* We failed as a result of a deadlock, see CheckDeadLock(). Quit now. * We failed as a result of a deadlock, see CheckDeadLock(). Quit now.
@ -1036,10 +1096,10 @@ WaitOnLock(LOCKMETHODID lockmethodid, LOCALLOCK *locallock,
awaitedLock = NULL; awaitedLock = NULL;
LOCK_PRINT("WaitOnLock: aborting on lock", LOCK_PRINT("WaitOnLock: aborting on lock",
locallock->lock, locallock->tag.mode); locallock->lock, locallock->tag.mode);
LWLockRelease(LockMgrLock); LWLockRelease(FirstLockMgrLock + locallock->partition);
/* /*
* Now that we aren't holding the LockMgrLock, we can give an error * Now that we aren't holding the partition lock, we can give an error
* report including details about the detected deadlock. * report including details about the detected deadlock.
*/ */
DeadLockReport(); DeadLockReport();
@ -1059,12 +1119,12 @@ WaitOnLock(LOCKMETHODID lockmethodid, LOCALLOCK *locallock,
* Remove a proc from the wait-queue it is on * Remove a proc from the wait-queue it is on
* (caller must know it is on one). * (caller must know it is on one).
* *
* Locktable lock must be held by caller. * Appropriate partition lock must be held by caller.
* *
* NB: this does not clean up any locallock object that may exist for the lock. * NB: this does not clean up any locallock object that may exist for the lock.
*/ */
void void
RemoveFromWaitQueue(PGPROC *proc) RemoveFromWaitQueue(PGPROC *proc, int partition)
{ {
LOCK *waitLock = proc->waitLock; LOCK *waitLock = proc->waitLock;
PROCLOCK *proclock = proc->waitProcLock; PROCLOCK *proclock = proc->waitProcLock;
@ -1102,7 +1162,9 @@ RemoveFromWaitQueue(PGPROC *proc)
* LockRelease expects there to be no remaining proclocks.) Then see if * LockRelease expects there to be no remaining proclocks.) Then see if
* any other waiters for the lock can be woken up now. * any other waiters for the lock can be woken up now.
*/ */
CleanUpLock(lockmethodid, waitLock, proclock, true); CleanUpLock(waitLock, proclock,
LockMethods[lockmethodid], partition,
true);
} }
/* /*
@ -1125,7 +1187,8 @@ LockRelease(const LOCKTAG *locktag, LOCKMODE lockmode, bool sessionLock)
LOCALLOCK *locallock; LOCALLOCK *locallock;
LOCK *lock; LOCK *lock;
PROCLOCK *proclock; PROCLOCK *proclock;
LWLockId masterLock; int partition;
LWLockId partitionLock;
bool wakeupNeeded; bool wakeupNeeded;
if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods)) if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
@ -1212,9 +1275,10 @@ LockRelease(const LOCKTAG *locktag, LOCKMODE lockmode, bool sessionLock)
/* /*
* Otherwise we've got to mess with the shared lock table. * Otherwise we've got to mess with the shared lock table.
*/ */
masterLock = LockMgrLock; partition = locallock->partition;
partitionLock = FirstLockMgrLock + partition;
LWLockAcquire(masterLock, LW_EXCLUSIVE); LWLockAcquire(partitionLock, LW_EXCLUSIVE);
/* /*
* We don't need to re-find the lock or proclock, since we kept their * We don't need to re-find the lock or proclock, since we kept their
@ -1233,7 +1297,7 @@ LockRelease(const LOCKTAG *locktag, LOCKMODE lockmode, bool sessionLock)
if (!(proclock->holdMask & LOCKBIT_ON(lockmode))) if (!(proclock->holdMask & LOCKBIT_ON(lockmode)))
{ {
PROCLOCK_PRINT("LockRelease: WRONGTYPE", proclock); PROCLOCK_PRINT("LockRelease: WRONGTYPE", proclock);
LWLockRelease(masterLock); LWLockRelease(partitionLock);
elog(WARNING, "you don't own a lock of type %s", elog(WARNING, "you don't own a lock of type %s",
lockMethodTable->lockModeNames[lockmode]); lockMethodTable->lockModeNames[lockmode]);
RemoveLocalLock(locallock); RemoveLocalLock(locallock);
@ -1245,9 +1309,11 @@ LockRelease(const LOCKTAG *locktag, LOCKMODE lockmode, bool sessionLock)
*/ */
wakeupNeeded = UnGrantLock(lock, lockmode, proclock, lockMethodTable); wakeupNeeded = UnGrantLock(lock, lockmode, proclock, lockMethodTable);
CleanUpLock(lockmethodid, lock, proclock, wakeupNeeded); CleanUpLock(lock, proclock,
lockMethodTable, partition,
wakeupNeeded);
LWLockRelease(masterLock); LWLockRelease(partitionLock);
RemoveLocalLock(locallock); RemoveLocalLock(locallock);
return TRUE; return TRUE;
@ -1265,14 +1331,13 @@ void
LockReleaseAll(LOCKMETHODID lockmethodid, bool allLocks) LockReleaseAll(LOCKMETHODID lockmethodid, bool allLocks)
{ {
HASH_SEQ_STATUS status; HASH_SEQ_STATUS status;
SHM_QUEUE *procLocks = &(MyProc->procLocks);
LWLockId masterLock;
LockMethod lockMethodTable; LockMethod lockMethodTable;
int i, int i,
numLockModes; numLockModes;
LOCALLOCK *locallock; LOCALLOCK *locallock;
PROCLOCK *proclock;
LOCK *lock; LOCK *lock;
PROCLOCK *proclock;
int partition;
if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods)) if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
elog(ERROR, "unrecognized lock method: %d", lockmethodid); elog(ERROR, "unrecognized lock method: %d", lockmethodid);
@ -1284,7 +1349,6 @@ LockReleaseAll(LOCKMETHODID lockmethodid, bool allLocks)
#endif #endif
numLockModes = lockMethodTable->numLockModes; numLockModes = lockMethodTable->numLockModes;
masterLock = LockMgrLock;
/* /*
* First we run through the locallock table and get rid of unwanted * First we run through the locallock table and get rid of unwanted
@ -1351,18 +1415,30 @@ LockReleaseAll(LOCKMETHODID lockmethodid, bool allLocks)
RemoveLocalLock(locallock); RemoveLocalLock(locallock);
} }
LWLockAcquire(masterLock, LW_EXCLUSIVE); /*
* Now, scan each lock partition separately.
*/
for (partition = 0; partition < NUM_LOCK_PARTITIONS; partition++)
{
LWLockId partitionLock = FirstLockMgrLock + partition;
SHM_QUEUE *procLocks = &(MyProc->myProcLocks[partition]);
proclock = (PROCLOCK *) SHMQueueNext(procLocks, procLocks, proclock = (PROCLOCK *) SHMQueueNext(procLocks, procLocks,
offsetof(PROCLOCK, procLink)); offsetof(PROCLOCK, procLink));
if (!proclock)
continue; /* needn't examine this partition */
LWLockAcquire(partitionLock, LW_EXCLUSIVE);
while (proclock) while (proclock)
{ {
bool wakeupNeeded = false; bool wakeupNeeded = false;
PROCLOCK *nextplock; PROCLOCK *nextplock;
/* Get link first, since we may unlink/delete this proclock */ /* Get link first, since we may unlink/delete this proclock */
nextplock = (PROCLOCK *) SHMQueueNext(procLocks, &proclock->procLink, nextplock = (PROCLOCK *)
SHMQueueNext(procLocks, &proclock->procLink,
offsetof(PROCLOCK, procLink)); offsetof(PROCLOCK, procLink));
Assert(proclock->tag.proc == MAKE_OFFSET(MyProc)); Assert(proclock->tag.proc == MAKE_OFFSET(MyProc));
@ -1412,13 +1488,16 @@ LockReleaseAll(LOCKMETHODID lockmethodid, bool allLocks)
proclock->releaseMask = 0; proclock->releaseMask = 0;
/* CleanUpLock will wake up waiters if needed. */ /* CleanUpLock will wake up waiters if needed. */
CleanUpLock(lockmethodid, lock, proclock, wakeupNeeded); CleanUpLock(lock, proclock,
lockMethodTable, partition,
wakeupNeeded);
next_item: next_item:
proclock = nextplock; proclock = nextplock;
} } /* loop over PROCLOCKs within this partition */
LWLockRelease(masterLock); LWLockRelease(partitionLock);
} /* loop over partitions */
#ifdef LOCK_DEBUG #ifdef LOCK_DEBUG
if (*(lockMethodTable->trace_flag)) if (*(lockMethodTable->trace_flag))
@ -1627,19 +1706,16 @@ PostPrepare_Locks(TransactionId xid)
{ {
PGPROC *newproc = TwoPhaseGetDummyProc(xid); PGPROC *newproc = TwoPhaseGetDummyProc(xid);
HASH_SEQ_STATUS status; HASH_SEQ_STATUS status;
SHM_QUEUE *procLocks = &(MyProc->procLocks);
LWLockId masterLock;
LOCALLOCK *locallock; LOCALLOCK *locallock;
LOCK *lock;
PROCLOCK *proclock; PROCLOCK *proclock;
PROCLOCKTAG proclocktag; PROCLOCKTAG proclocktag;
bool found; bool found;
LOCK *lock; int partition;
/* This is a critical section: any error means big trouble */ /* This is a critical section: any error means big trouble */
START_CRIT_SECTION(); START_CRIT_SECTION();
masterLock = LockMgrLock;
/* /*
* First we run through the locallock table and get rid of unwanted * First we run through the locallock table and get rid of unwanted
* entries, then we scan the process's proclocks and transfer them to the * entries, then we scan the process's proclocks and transfer them to the
@ -1678,11 +1754,22 @@ PostPrepare_Locks(TransactionId xid)
RemoveLocalLock(locallock); RemoveLocalLock(locallock);
} }
LWLockAcquire(masterLock, LW_EXCLUSIVE); /*
* Now, scan each lock partition separately.
*/
for (partition = 0; partition < NUM_LOCK_PARTITIONS; partition++)
{
LWLockId partitionLock = FirstLockMgrLock + partition;
SHM_QUEUE *procLocks = &(MyProc->myProcLocks[partition]);
proclock = (PROCLOCK *) SHMQueueNext(procLocks, procLocks, proclock = (PROCLOCK *) SHMQueueNext(procLocks, procLocks,
offsetof(PROCLOCK, procLink)); offsetof(PROCLOCK, procLink));
if (!proclock)
continue; /* needn't examine this partition */
LWLockAcquire(partitionLock, LW_EXCLUSIVE);
while (proclock) while (proclock)
{ {
PROCLOCK *nextplock; PROCLOCK *nextplock;
@ -1690,7 +1777,8 @@ PostPrepare_Locks(TransactionId xid)
PROCLOCK *newproclock; PROCLOCK *newproclock;
/* Get link first, since we may unlink/delete this proclock */ /* Get link first, since we may unlink/delete this proclock */
nextplock = (PROCLOCK *) SHMQueueNext(procLocks, &proclock->procLink, nextplock = (PROCLOCK *)
SHMQueueNext(procLocks, &proclock->procLink,
offsetof(PROCLOCK, procLink)); offsetof(PROCLOCK, procLink));
Assert(proclock->tag.proc == MAKE_OFFSET(MyProc)); Assert(proclock->tag.proc == MAKE_OFFSET(MyProc));
@ -1709,7 +1797,8 @@ PostPrepare_Locks(TransactionId xid)
Assert((proclock->holdMask & ~lock->grantMask) == 0); Assert((proclock->holdMask & ~lock->grantMask) == 0);
/* /*
* Since there were no session locks, we should be releasing all locks * Since there were no session locks, we should be releasing all
* locks
*/ */
if (proclock->releaseMask != proclock->holdMask) if (proclock->releaseMask != proclock->holdMask)
elog(PANIC, "we seem to have dropped a bit somewhere"); elog(PANIC, "we seem to have dropped a bit somewhere");
@ -1717,17 +1806,18 @@ PostPrepare_Locks(TransactionId xid)
holdMask = proclock->holdMask; holdMask = proclock->holdMask;
/* /*
* We cannot simply modify proclock->tag.proc to reassign ownership of * We cannot simply modify proclock->tag.proc to reassign
* the lock, because that's part of the hash key and the proclock * ownership of the lock, because that's part of the hash key and
* would then be in the wrong hash chain. So, unlink and delete the * the proclock would then be in the wrong hash chain. So, unlink
* old proclock; create a new one with the right contents; and link it * and delete the old proclock; create a new one with the right
* into place. We do it in this order to be certain we won't run out * contents; and link it into place. We do it in this order to be
* of shared memory (the way dynahash.c works, the deleted object is * certain we won't run out of shared memory (the way dynahash.c
* certain to be available for reallocation). * works, the deleted object is certain to be available for
* reallocation).
*/ */
SHMQueueDelete(&proclock->lockLink); SHMQueueDelete(&proclock->lockLink);
SHMQueueDelete(&proclock->procLink); SHMQueueDelete(&proclock->procLink);
if (!hash_search(LockMethodProcLockHash, if (!hash_search(LockMethodProcLockHash[partition],
(void *) &(proclock->tag), (void *) &(proclock->tag),
HASH_REMOVE, NULL)) HASH_REMOVE, NULL))
elog(PANIC, "proclock table corrupted"); elog(PANIC, "proclock table corrupted");
@ -1739,7 +1829,7 @@ PostPrepare_Locks(TransactionId xid)
proclocktag.lock = MAKE_OFFSET(lock); proclocktag.lock = MAKE_OFFSET(lock);
proclocktag.proc = MAKE_OFFSET(newproc); proclocktag.proc = MAKE_OFFSET(newproc);
newproclock = (PROCLOCK *) hash_search(LockMethodProcLockHash, newproclock = (PROCLOCK *) hash_search(LockMethodProcLockHash[partition],
(void *) &proclocktag, (void *) &proclocktag,
HASH_ENTER_NULL, &found); HASH_ENTER_NULL, &found);
if (!newproclock) if (!newproclock)
@ -1757,7 +1847,8 @@ PostPrepare_Locks(TransactionId xid)
newproclock->releaseMask = 0; newproclock->releaseMask = 0;
/* Add new proclock to appropriate lists */ /* Add new proclock to appropriate lists */
SHMQueueInsertBefore(&lock->procLocks, &newproclock->lockLink); SHMQueueInsertBefore(&lock->procLocks, &newproclock->lockLink);
SHMQueueInsertBefore(&newproc->procLocks, &newproclock->procLink); SHMQueueInsertBefore(&(newproc->myProcLocks[partition]),
&newproclock->procLink);
PROCLOCK_PRINT("PostPrepare_Locks: new", newproclock); PROCLOCK_PRINT("PostPrepare_Locks: new", newproclock);
} }
else else
@ -1772,11 +1863,12 @@ PostPrepare_Locks(TransactionId xid)
Assert((newproclock->holdMask & holdMask) == 0); Assert((newproclock->holdMask & holdMask) == 0);
newproclock->holdMask |= holdMask; newproclock->holdMask |= holdMask;
next_item: next_item:
proclock = nextplock; proclock = nextplock;
} } /* loop over PROCLOCKs within this partition */
LWLockRelease(masterLock); LWLockRelease(partitionLock);
} /* loop over partitions */
END_CRIT_SECTION(); END_CRIT_SECTION();
} }
@ -1789,20 +1881,23 @@ Size
LockShmemSize(void) LockShmemSize(void)
{ {
Size size = 0; Size size = 0;
long max_table_size = NLOCKENTS(); Size tabsize;
long max_table_size;
/* lockHash table */ /* lock hash tables */
size = add_size(size, hash_estimate_size(max_table_size, sizeof(LOCK))); max_table_size = NLOCKENTS();
max_table_size = (max_table_size - 1) / NUM_LOCK_PARTITIONS + 1;
tabsize = hash_estimate_size(max_table_size, sizeof(LOCK));
size = add_size(size, mul_size(tabsize, NUM_LOCK_PARTITIONS));
/* proclockHash table */ /* proclock hash tables */
size = add_size(size, hash_estimate_size(max_table_size, sizeof(PROCLOCK))); max_table_size *= 2;
tabsize = hash_estimate_size(max_table_size, sizeof(PROCLOCK));
size = add_size(size, mul_size(tabsize, NUM_LOCK_PARTITIONS));
/* /*
* Note we count only one pair of hash tables, since the userlocks table * Since there is likely to be some space wastage due to uneven use
* actually overlays the main one. * of the partitions, add 10% safety margin.
*
* Since the lockHash entry count above is only an estimate, add 10%
* safety margin.
*/ */
size = add_size(size, size / 10); size = add_size(size, size / 10);
@ -1818,9 +1913,9 @@ LockShmemSize(void)
* copies of the same PGPROC and/or LOCK objects are likely to appear. * copies of the same PGPROC and/or LOCK objects are likely to appear.
* It is the caller's responsibility to match up duplicates if wanted. * It is the caller's responsibility to match up duplicates if wanted.
* *
* The design goal is to hold the LockMgrLock for as short a time as possible; * The design goal is to hold the LWLocks for as short a time as possible;
* thus, this function simply makes a copy of the necessary data and releases * thus, this function simply makes a copy of the necessary data and releases
* the lock, allowing the caller to contemplate and format the data for as * the locks, allowing the caller to contemplate and format the data for as
* long as it pleases. * long as it pleases.
*/ */
LockData * LockData *
@ -1830,40 +1925,67 @@ GetLockStatusData(void)
HTAB *proclockTable; HTAB *proclockTable;
PROCLOCK *proclock; PROCLOCK *proclock;
HASH_SEQ_STATUS seqstat; HASH_SEQ_STATUS seqstat;
int els;
int el;
int i; int i;
data = (LockData *) palloc(sizeof(LockData)); data = (LockData *) palloc(sizeof(LockData));
LWLockAcquire(LockMgrLock, LW_EXCLUSIVE); /*
* Acquire lock on the entire shared lock data structures. We can't
* operate one partition at a time if we want to deliver a self-consistent
* view of the state.
*
* Since this is a read-only operation, we take shared instead of exclusive
* lock. There's not a whole lot of point to this, because all the normal
* operations require exclusive lock, but it doesn't hurt anything either.
* It will at least allow two backends to do GetLockStatusData in parallel.
*
* Must grab LWLocks in partition-number order to avoid LWLock deadlock.
*
* Use same loop to count up the total number of PROCLOCK objects.
*/
els = 0;
for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
{
LWLockAcquire(FirstLockMgrLock + i, LW_SHARED);
proclockTable = LockMethodProcLockHash[i];
els += proclockTable->hctl->nentries;
}
proclockTable = LockMethodProcLockHash; data->nelements = els;
data->proclockaddrs = (SHMEM_OFFSET *) palloc(sizeof(SHMEM_OFFSET) * els);
data->proclocks = (PROCLOCK *) palloc(sizeof(PROCLOCK) * els);
data->procs = (PGPROC *) palloc(sizeof(PGPROC) * els);
data->locks = (LOCK *) palloc(sizeof(LOCK) * els);
data->nelements = i = proclockTable->hctl->nentries; el = 0;
data->proclockaddrs = (SHMEM_OFFSET *) palloc(sizeof(SHMEM_OFFSET) * i);
data->proclocks = (PROCLOCK *) palloc(sizeof(PROCLOCK) * i);
data->procs = (PGPROC *) palloc(sizeof(PGPROC) * i);
data->locks = (LOCK *) palloc(sizeof(LOCK) * i);
/* Now scan the tables to copy the data */
for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
{
proclockTable = LockMethodProcLockHash[i];
hash_seq_init(&seqstat, proclockTable); hash_seq_init(&seqstat, proclockTable);
i = 0;
while ((proclock = hash_seq_search(&seqstat))) while ((proclock = hash_seq_search(&seqstat)))
{ {
PGPROC *proc = (PGPROC *) MAKE_PTR(proclock->tag.proc); PGPROC *proc = (PGPROC *) MAKE_PTR(proclock->tag.proc);
LOCK *lock = (LOCK *) MAKE_PTR(proclock->tag.lock); LOCK *lock = (LOCK *) MAKE_PTR(proclock->tag.lock);
data->proclockaddrs[i] = MAKE_OFFSET(proclock); data->proclockaddrs[el] = MAKE_OFFSET(proclock);
memcpy(&(data->proclocks[i]), proclock, sizeof(PROCLOCK)); memcpy(&(data->proclocks[el]), proclock, sizeof(PROCLOCK));
memcpy(&(data->procs[i]), proc, sizeof(PGPROC)); memcpy(&(data->procs[el]), proc, sizeof(PGPROC));
memcpy(&(data->locks[i]), lock, sizeof(LOCK)); memcpy(&(data->locks[el]), lock, sizeof(LOCK));
i++; el++;
}
} }
LWLockRelease(LockMgrLock); /* And release locks */
for (i = NUM_LOCK_PARTITIONS; --i >= 0; )
LWLockRelease(FirstLockMgrLock + i);
Assert(i == data->nelements); Assert(el == data->nelements);
return data; return data;
} }
@ -1879,7 +2001,7 @@ GetLockmodeName(LOCKMETHODID lockmethodid, LOCKMODE mode)
#ifdef LOCK_DEBUG #ifdef LOCK_DEBUG
/* /*
* Dump all locks in the given proc's procLocks list. * Dump all locks in the given proc's myProcLocks lists.
* *
* Caller is responsible for having acquired appropriate LWLocks. * Caller is responsible for having acquired appropriate LWLocks.
*/ */
@ -1889,15 +2011,18 @@ DumpLocks(PGPROC *proc)
SHM_QUEUE *procLocks; SHM_QUEUE *procLocks;
PROCLOCK *proclock; PROCLOCK *proclock;
LOCK *lock; LOCK *lock;
int i;
if (proc == NULL) if (proc == NULL)
return; return;
procLocks = &proc->procLocks;
if (proc->waitLock) if (proc->waitLock)
LOCK_PRINT("DumpLocks: waiting on", proc->waitLock, 0); LOCK_PRINT("DumpLocks: waiting on", proc->waitLock, 0);
for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
{
procLocks = &(proc->myProcLocks[i]);
proclock = (PROCLOCK *) SHMQueueNext(procLocks, procLocks, proclock = (PROCLOCK *) SHMQueueNext(procLocks, procLocks,
offsetof(PROCLOCK, procLink)); offsetof(PROCLOCK, procLink));
@ -1910,9 +2035,11 @@ DumpLocks(PGPROC *proc)
PROCLOCK_PRINT("DumpLocks", proclock); PROCLOCK_PRINT("DumpLocks", proclock);
LOCK_PRINT("DumpLocks", lock, 0); LOCK_PRINT("DumpLocks", lock, 0);
proclock = (PROCLOCK *) SHMQueueNext(procLocks, &proclock->procLink, proclock = (PROCLOCK *)
SHMQueueNext(procLocks, &proclock->procLink,
offsetof(PROCLOCK, procLink)); offsetof(PROCLOCK, procLink));
} }
}
} }
/* /*
@ -1928,14 +2055,18 @@ DumpAllLocks(void)
LOCK *lock; LOCK *lock;
HTAB *proclockTable; HTAB *proclockTable;
HASH_SEQ_STATUS status; HASH_SEQ_STATUS status;
int i;
proc = MyProc; proc = MyProc;
proclockTable = LockMethodProcLockHash;
if (proc && proc->waitLock) if (proc && proc->waitLock)
LOCK_PRINT("DumpAllLocks: waiting on", proc->waitLock, 0); LOCK_PRINT("DumpAllLocks: waiting on", proc->waitLock, 0);
for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
{
proclockTable = LockMethodProcLockHash[i];
hash_seq_init(&status, proclockTable); hash_seq_init(&status, proclockTable);
while ((proclock = (PROCLOCK *) hash_seq_search(&status)) != NULL) while ((proclock = (PROCLOCK *) hash_seq_search(&status)) != NULL)
{ {
PROCLOCK_PRINT("DumpAllLocks", proclock); PROCLOCK_PRINT("DumpAllLocks", proclock);
@ -1948,6 +2079,7 @@ DumpAllLocks(void)
else else
elog(LOG, "DumpAllLocks: proclock->tag.lock = NULL"); elog(LOG, "DumpAllLocks: proclock->tag.lock = NULL");
} }
}
} }
#endif /* LOCK_DEBUG */ #endif /* LOCK_DEBUG */
@ -1975,7 +2107,8 @@ lock_twophase_recover(TransactionId xid, uint16 info,
PROCLOCK *proclock; PROCLOCK *proclock;
PROCLOCKTAG proclocktag; PROCLOCKTAG proclocktag;
bool found; bool found;
LWLockId masterLock; int partition;
LWLockId partitionLock;
LockMethod lockMethodTable; LockMethod lockMethodTable;
Assert(len == sizeof(TwoPhaseLockRecord)); Assert(len == sizeof(TwoPhaseLockRecord));
@ -1987,19 +2120,20 @@ lock_twophase_recover(TransactionId xid, uint16 info,
elog(ERROR, "unrecognized lock method: %d", lockmethodid); elog(ERROR, "unrecognized lock method: %d", lockmethodid);
lockMethodTable = LockMethods[lockmethodid]; lockMethodTable = LockMethods[lockmethodid];
masterLock = LockMgrLock; partition = LockTagToPartition(locktag);
partitionLock = FirstLockMgrLock + partition;
LWLockAcquire(masterLock, LW_EXCLUSIVE); LWLockAcquire(partitionLock, LW_EXCLUSIVE);
/* /*
* Find or create a lock with this tag. * Find or create a lock with this tag.
*/ */
lock = (LOCK *) hash_search(LockMethodLockHash, lock = (LOCK *) hash_search(LockMethodLockHash[partition],
(void *) locktag, (void *) locktag,
HASH_ENTER_NULL, &found); HASH_ENTER_NULL, &found);
if (!lock) if (!lock)
{ {
LWLockRelease(masterLock); LWLockRelease(partitionLock);
ereport(ERROR, ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY), (errcode(ERRCODE_OUT_OF_MEMORY),
errmsg("out of shared memory"), errmsg("out of shared memory"),
@ -2039,7 +2173,7 @@ lock_twophase_recover(TransactionId xid, uint16 info,
/* /*
* Find or create a proclock entry with this tag * Find or create a proclock entry with this tag
*/ */
proclock = (PROCLOCK *) hash_search(LockMethodProcLockHash, proclock = (PROCLOCK *) hash_search(LockMethodProcLockHash[partition],
(void *) &proclocktag, (void *) &proclocktag,
HASH_ENTER_NULL, &found); HASH_ENTER_NULL, &found);
if (!proclock) if (!proclock)
@ -2054,12 +2188,12 @@ lock_twophase_recover(TransactionId xid, uint16 info,
* anyone to release the lock object later. * anyone to release the lock object later.
*/ */
Assert(SHMQueueEmpty(&(lock->procLocks))); Assert(SHMQueueEmpty(&(lock->procLocks)));
if (!hash_search(LockMethodLockHash, if (!hash_search(LockMethodLockHash[partition],
(void *) &(lock->tag), (void *) &(lock->tag),
HASH_REMOVE, NULL)) HASH_REMOVE, NULL))
elog(PANIC, "lock table corrupted"); elog(PANIC, "lock table corrupted");
} }
LWLockRelease(masterLock); LWLockRelease(partitionLock);
ereport(ERROR, ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY), (errcode(ERRCODE_OUT_OF_MEMORY),
errmsg("out of shared memory"), errmsg("out of shared memory"),
@ -2075,7 +2209,8 @@ lock_twophase_recover(TransactionId xid, uint16 info,
proclock->releaseMask = 0; proclock->releaseMask = 0;
/* Add proclock to appropriate lists */ /* Add proclock to appropriate lists */
SHMQueueInsertBefore(&lock->procLocks, &proclock->lockLink); SHMQueueInsertBefore(&lock->procLocks, &proclock->lockLink);
SHMQueueInsertBefore(&proc->procLocks, &proclock->procLink); SHMQueueInsertBefore(&(proc->myProcLocks[partition]),
&proclock->procLink);
PROCLOCK_PRINT("lock_twophase_recover: new", proclock); PROCLOCK_PRINT("lock_twophase_recover: new", proclock);
} }
else else
@ -2106,7 +2241,7 @@ lock_twophase_recover(TransactionId xid, uint16 info,
*/ */
GrantLock(lock, proclock, lockmode); GrantLock(lock, proclock, lockmode);
LWLockRelease(masterLock); LWLockRelease(partitionLock);
} }
/* /*
@ -2123,10 +2258,11 @@ lock_twophase_postcommit(TransactionId xid, uint16 info,
LOCKTAG *locktag; LOCKTAG *locktag;
LOCKMODE lockmode; LOCKMODE lockmode;
LOCKMETHODID lockmethodid; LOCKMETHODID lockmethodid;
PROCLOCKTAG proclocktag;
LOCK *lock; LOCK *lock;
PROCLOCK *proclock; PROCLOCK *proclock;
LWLockId masterLock; PROCLOCKTAG proclocktag;
int partition;
LWLockId partitionLock;
LockMethod lockMethodTable; LockMethod lockMethodTable;
bool wakeupNeeded; bool wakeupNeeded;
@ -2139,14 +2275,15 @@ lock_twophase_postcommit(TransactionId xid, uint16 info,
elog(ERROR, "unrecognized lock method: %d", lockmethodid); elog(ERROR, "unrecognized lock method: %d", lockmethodid);
lockMethodTable = LockMethods[lockmethodid]; lockMethodTable = LockMethods[lockmethodid];
masterLock = LockMgrLock; partition = LockTagToPartition(locktag);
partitionLock = FirstLockMgrLock + partition;
LWLockAcquire(masterLock, LW_EXCLUSIVE); LWLockAcquire(partitionLock, LW_EXCLUSIVE);
/* /*
* Re-find the lock object (it had better be there). * Re-find the lock object (it had better be there).
*/ */
lock = (LOCK *) hash_search(LockMethodLockHash, lock = (LOCK *) hash_search(LockMethodLockHash[partition],
(void *) locktag, (void *) locktag,
HASH_FIND, NULL); HASH_FIND, NULL);
if (!lock) if (!lock)
@ -2158,7 +2295,7 @@ lock_twophase_postcommit(TransactionId xid, uint16 info,
MemSet(&proclocktag, 0, sizeof(PROCLOCKTAG)); /* must clear padding */ MemSet(&proclocktag, 0, sizeof(PROCLOCKTAG)); /* must clear padding */
proclocktag.lock = MAKE_OFFSET(lock); proclocktag.lock = MAKE_OFFSET(lock);
proclocktag.proc = MAKE_OFFSET(proc); proclocktag.proc = MAKE_OFFSET(proc);
proclock = (PROCLOCK *) hash_search(LockMethodProcLockHash, proclock = (PROCLOCK *) hash_search(LockMethodProcLockHash[partition],
(void *) &proclocktag, (void *) &proclocktag,
HASH_FIND, NULL); HASH_FIND, NULL);
if (!proclock) if (!proclock)
@ -2171,7 +2308,7 @@ lock_twophase_postcommit(TransactionId xid, uint16 info,
if (!(proclock->holdMask & LOCKBIT_ON(lockmode))) if (!(proclock->holdMask & LOCKBIT_ON(lockmode)))
{ {
PROCLOCK_PRINT("lock_twophase_postcommit: WRONGTYPE", proclock); PROCLOCK_PRINT("lock_twophase_postcommit: WRONGTYPE", proclock);
LWLockRelease(masterLock); LWLockRelease(partitionLock);
elog(WARNING, "you don't own a lock of type %s", elog(WARNING, "you don't own a lock of type %s",
lockMethodTable->lockModeNames[lockmode]); lockMethodTable->lockModeNames[lockmode]);
return; return;
@ -2182,9 +2319,11 @@ lock_twophase_postcommit(TransactionId xid, uint16 info,
*/ */
wakeupNeeded = UnGrantLock(lock, lockmode, proclock, lockMethodTable); wakeupNeeded = UnGrantLock(lock, lockmode, proclock, lockMethodTable);
CleanUpLock(lockmethodid, lock, proclock, wakeupNeeded); CleanUpLock(lock, proclock,
lockMethodTable, partition,
wakeupNeeded);
LWLockRelease(masterLock); LWLockRelease(partitionLock);
} }
/* /*

14
src/backend/storage/lmgr/lwlock.c
View File

@ -8,14 +8,14 @@
* exclusive and shared lock modes (to support read/write and read-only * exclusive and shared lock modes (to support read/write and read-only
* access to a shared object). There are few other frammishes. User-level * access to a shared object). There are few other frammishes. User-level
* locking should be done with the full lock manager --- which depends on * locking should be done with the full lock manager --- which depends on
* an LWLock to protect its shared state. * LWLocks to protect its shared state.
* *
* *
* Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group * Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group
* Portions Copyright (c) 1994, Regents of the University of California * Portions Copyright (c) 1994, Regents of the University of California
* *
* IDENTIFICATION * IDENTIFICATION
* $PostgreSQL: pgsql/src/backend/storage/lmgr/lwlock.c,v 1.35 2005/12/06 23:08:33 tgl Exp $ * $PostgreSQL: pgsql/src/backend/storage/lmgr/lwlock.c,v 1.36 2005/12/11 21:02:18 tgl Exp $
* *
*------------------------------------------------------------------------- *-------------------------------------------------------------------------
*/ */
@ -125,7 +125,10 @@ NumLWLocks(void)
*/ */
/* Predefined LWLocks */ /* Predefined LWLocks */
numLocks = (int) NumFixedLWLocks; numLocks = (int) FirstLockMgrLock;
/* lock.c gets the ones starting at FirstLockMgrLock */
numLocks += NUM_LOCK_PARTITIONS;
/* bufmgr.c needs two for each shared buffer */ /* bufmgr.c needs two for each shared buffer */
numLocks += 2 * NBuffers; numLocks += 2 * NBuffers;
@ -204,10 +207,11 @@ CreateLWLocks(void)
/* /*
* Initialize the dynamic-allocation counter, which is stored just before * Initialize the dynamic-allocation counter, which is stored just before
* the first LWLock. * the first LWLock. The LWLocks used by lock.c are not dynamically
* allocated, it just assumes it has them.
*/ */
LWLockCounter = (int *) ((char *) LWLockArray - 2 * sizeof(int)); LWLockCounter = (int *) ((char *) LWLockArray - 2 * sizeof(int));
LWLockCounter[0] = (int) NumFixedLWLocks; LWLockCounter[0] = (int) FirstLockMgrLock + NUM_LOCK_PARTITIONS;
LWLockCounter[1] = numLocks; LWLockCounter[1] = numLocks;
} }

126
src/backend/storage/lmgr/proc.c
View File

@ -8,7 +8,7 @@
* *
* *
* IDENTIFICATION * IDENTIFICATION
* $PostgreSQL: pgsql/src/backend/storage/lmgr/proc.c,v 1.169 2005/12/09 01:22:04 tgl Exp $ * $PostgreSQL: pgsql/src/backend/storage/lmgr/proc.c,v 1.170 2005/12/11 21:02:18 tgl Exp $
* *
*------------------------------------------------------------------------- *-------------------------------------------------------------------------
*/ */
@ -18,9 +18,8 @@
* ProcQueueAlloc() -- create a shm queue for sleeping processes * ProcQueueAlloc() -- create a shm queue for sleeping processes
* ProcQueueInit() -- create a queue without allocing memory * ProcQueueInit() -- create a queue without allocing memory
* *
* Locking and waiting for buffers can cause the backend to be * Waiting for a lock causes the backend to be put to sleep. Whoever releases
* put to sleep. Whoever releases the lock, etc. wakes the * the lock wakes the process up again (and gives it an error code so it knows
* process up again (and gives it an error code so it knows
* whether it was awoken on an error condition). * whether it was awoken on an error condition).
* *
* Interface (b): * Interface (b):
@ -65,7 +64,8 @@ NON_EXEC_STATIC slock_t *ProcStructLock = NULL;
static PROC_HDR *ProcGlobal = NULL; static PROC_HDR *ProcGlobal = NULL;
static PGPROC *DummyProcs = NULL; static PGPROC *DummyProcs = NULL;
static bool waitingForLock = false; /* If we are waiting for a lock, this points to the associated LOCALLOCK */
static LOCALLOCK *lockAwaited = NULL;
/* Mark these volatile because they can be changed by signal handler */ /* Mark these volatile because they can be changed by signal handler */
static volatile bool statement_timeout_active = false; static volatile bool statement_timeout_active = false;
@ -200,10 +200,10 @@ InitProcGlobal(void)
void void
InitProcess(void) InitProcess(void)
{ {
SHMEM_OFFSET myOffset;
/* use volatile pointer to prevent code rearrangement */ /* use volatile pointer to prevent code rearrangement */
volatile PROC_HDR *procglobal = ProcGlobal; volatile PROC_HDR *procglobal = ProcGlobal;
SHMEM_OFFSET myOffset;
int i;
/* /*
* ProcGlobal should be set by a previous call to InitProcGlobal (if we * ProcGlobal should be set by a previous call to InitProcGlobal (if we
@ -264,7 +264,8 @@ InitProcess(void)
MyProc->lwWaitLink = NULL; MyProc->lwWaitLink = NULL;
MyProc->waitLock = NULL; MyProc->waitLock = NULL;
MyProc->waitProcLock = NULL; MyProc->waitProcLock = NULL;
SHMQueueInit(&(MyProc->procLocks)); for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
SHMQueueInit(&(MyProc->myProcLocks[i]));
/* /*
* Add our PGPROC to the PGPROC array in shared memory. * Add our PGPROC to the PGPROC array in shared memory.
@ -304,6 +305,7 @@ void
InitDummyProcess(int proctype) InitDummyProcess(int proctype)
{ {
PGPROC *dummyproc; PGPROC *dummyproc;
int i;
/* /*
* ProcGlobal should be set by a previous call to InitProcGlobal (we * ProcGlobal should be set by a previous call to InitProcGlobal (we
@ -360,7 +362,8 @@ InitDummyProcess(int proctype)
MyProc->lwWaitLink = NULL; MyProc->lwWaitLink = NULL;
MyProc->waitLock = NULL; MyProc->waitLock = NULL;
MyProc->waitProcLock = NULL; MyProc->waitProcLock = NULL;
SHMQueueInit(&(MyProc->procLocks)); for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
SHMQueueInit(&(MyProc->myProcLocks[i]));
/* /*
* Arrange to clean up at process exit. * Arrange to clean up at process exit.
@ -416,21 +419,24 @@ HaveNFreeProcs(int n)
bool bool
LockWaitCancel(void) LockWaitCancel(void)
{ {
LWLockId partitionLock;
/* Nothing to do if we weren't waiting for a lock */ /* Nothing to do if we weren't waiting for a lock */
if (!waitingForLock) if (lockAwaited == NULL)
return false; return false;
/* Turn off the deadlock timer, if it's still running (see ProcSleep) */ /* Turn off the deadlock timer, if it's still running (see ProcSleep) */
disable_sig_alarm(false); disable_sig_alarm(false);
/* Unlink myself from the wait queue, if on it (might not be anymore!) */ /* Unlink myself from the wait queue, if on it (might not be anymore!) */
LWLockAcquire(LockMgrLock, LW_EXCLUSIVE); partitionLock = FirstLockMgrLock + lockAwaited->partition;
LWLockAcquire(partitionLock, LW_EXCLUSIVE);
if (MyProc->links.next != INVALID_OFFSET) if (MyProc->links.next != INVALID_OFFSET)
{ {
/* We could not have been granted the lock yet */ /* We could not have been granted the lock yet */
Assert(MyProc->waitStatus == STATUS_ERROR); Assert(MyProc->waitStatus == STATUS_ERROR);
RemoveFromWaitQueue(MyProc); RemoveFromWaitQueue(MyProc, lockAwaited->partition);
} }
else else
{ {
@ -444,9 +450,9 @@ LockWaitCancel(void)
GrantAwaitedLock(); GrantAwaitedLock();
} }
waitingForLock = false; lockAwaited = NULL;
LWLockRelease(LockMgrLock); LWLockRelease(partitionLock);
/* /*
* Reset the proc wait semaphore to zero. This is necessary in the * Reset the proc wait semaphore to zero. This is necessary in the
@ -606,18 +612,18 @@ ProcQueueInit(PROC_QUEUE *queue)
/* /*
* ProcSleep -- put a process to sleep * ProcSleep -- put a process to sleep on the specified lock
* *
* Caller must have set MyProc->heldLocks to reflect locks already held * Caller must have set MyProc->heldLocks to reflect locks already held
* on the lockable object by this process (under all XIDs). * on the lockable object by this process (under all XIDs).
* *
* Locktable's masterLock must be held at entry, and will be held * The lock table's partition lock must be held at entry, and will be held
* at exit. * at exit.
* *
* Result: STATUS_OK if we acquired the lock, STATUS_ERROR if not (deadlock). * Result: STATUS_OK if we acquired the lock, STATUS_ERROR if not (deadlock).
* *
* ASSUME: that no one will fiddle with the queue until after * ASSUME: that no one will fiddle with the queue until after
* we release the masterLock. * we release the partition lock.
* *
* NOTES: The process queue is now a priority queue for locking. * NOTES: The process queue is now a priority queue for locking.
* *
@ -625,12 +631,13 @@ ProcQueueInit(PROC_QUEUE *queue)
* semaphore is normally zero, so when we try to acquire it, we sleep. * semaphore is normally zero, so when we try to acquire it, we sleep.
*/ */
int int
ProcSleep(LockMethod lockMethodTable, ProcSleep(LOCALLOCK *locallock, LockMethod lockMethodTable)
LOCKMODE lockmode,
LOCK *lock,
PROCLOCK *proclock)
{ {
LWLockId masterLock = LockMgrLock; LOCKMODE lockmode = locallock->tag.mode;
LOCK *lock = locallock->lock;
PROCLOCK *proclock = locallock->proclock;
int partition = locallock->partition;
LWLockId partitionLock = FirstLockMgrLock + partition;
PROC_QUEUE *waitQueue = &(lock->waitProcs); PROC_QUEUE *waitQueue = &(lock->waitProcs);
LOCKMASK myHeldLocks = MyProc->heldLocks; LOCKMASK myHeldLocks = MyProc->heldLocks;
bool early_deadlock = false; bool early_deadlock = false;
@ -732,22 +739,22 @@ ProcSleep(LockMethod lockMethodTable,
*/ */
if (early_deadlock) if (early_deadlock)
{ {
RemoveFromWaitQueue(MyProc); RemoveFromWaitQueue(MyProc, partition);
return STATUS_ERROR; return STATUS_ERROR;
} }
/* mark that we are waiting for a lock */ /* mark that we are waiting for a lock */
waitingForLock = true; lockAwaited = locallock;
/* /*
* Release the locktable's masterLock. * Release the lock table's partition lock.
* *
* NOTE: this may also cause us to exit critical-section state, possibly * NOTE: this may also cause us to exit critical-section state, possibly
* allowing a cancel/die interrupt to be accepted. This is OK because we * allowing a cancel/die interrupt to be accepted. This is OK because we
* have recorded the fact that we are waiting for a lock, and so * have recorded the fact that we are waiting for a lock, and so
* LockWaitCancel will clean up if cancel/die happens. * LockWaitCancel will clean up if cancel/die happens.
*/ */
LWLockRelease(masterLock); LWLockRelease(partitionLock);
/* /*
* Set timer so we can wake up after awhile and check for a deadlock. If a * Set timer so we can wake up after awhile and check for a deadlock. If a
@ -785,16 +792,16 @@ ProcSleep(LockMethod lockMethodTable,
elog(FATAL, "could not disable timer for process wakeup"); elog(FATAL, "could not disable timer for process wakeup");
/* /*
* Re-acquire the locktable's masterLock. We have to do this to hold off * Re-acquire the lock table's partition lock. We have to do this to
* cancel/die interrupts before we can mess with waitingForLock (else we * hold off cancel/die interrupts before we can mess with lockAwaited
* might have a missed or duplicated locallock update). * (else we might have a missed or duplicated locallock update).
*/ */
LWLockAcquire(masterLock, LW_EXCLUSIVE); LWLockAcquire(partitionLock, LW_EXCLUSIVE);
/* /*
* We no longer want LockWaitCancel to do anything. * We no longer want LockWaitCancel to do anything.
*/ */
waitingForLock = false; lockAwaited = NULL;
/* /*
* If we got the lock, be sure to remember it in the locallock table. * If we got the lock, be sure to remember it in the locallock table.
@ -816,6 +823,8 @@ ProcSleep(LockMethod lockMethodTable,
* Also remove the process from the wait queue and set its links invalid. * Also remove the process from the wait queue and set its links invalid.
* RETURN: the next process in the wait queue. * RETURN: the next process in the wait queue.
* *
* The appropriate lock partition lock must be held by caller.
*
* XXX: presently, this code is only used for the "success" case, and only * XXX: presently, this code is only used for the "success" case, and only
* works correctly for that case. To clean up in failure case, would need * works correctly for that case. To clean up in failure case, would need
* to twiddle the lock's request counts too --- see RemoveFromWaitQueue. * to twiddle the lock's request counts too --- see RemoveFromWaitQueue.
@ -825,8 +834,6 @@ ProcWakeup(PGPROC *proc, int waitStatus)
{ {
PGPROC *retProc; PGPROC *retProc;
/* assume that masterLock has been acquired */
/* Proc should be sleeping ... */ /* Proc should be sleeping ... */
if (proc->links.prev == INVALID_OFFSET || if (proc->links.prev == INVALID_OFFSET ||
proc->links.next == INVALID_OFFSET) proc->links.next == INVALID_OFFSET)
@ -854,6 +861,8 @@ ProcWakeup(PGPROC *proc, int waitStatus)
* ProcLockWakeup -- routine for waking up processes when a lock is * ProcLockWakeup -- routine for waking up processes when a lock is
* released (or a prior waiter is aborted). Scan all waiters * released (or a prior waiter is aborted). Scan all waiters
* for lock, waken any that are no longer blocked. * for lock, waken any that are no longer blocked.
*
* The appropriate lock partition lock must be held by caller.
*/ */
void void
ProcLockWakeup(LockMethod lockMethodTable, LOCK *lock) ProcLockWakeup(LockMethod lockMethodTable, LOCK *lock)
@ -908,25 +917,32 @@ ProcLockWakeup(LockMethod lockMethodTable, LOCK *lock)
Assert(waitQueue->size >= 0); Assert(waitQueue->size >= 0);
} }
/* -------------------- /*
* CheckDeadLock
*
* We only get to this routine if we got SIGALRM after DeadlockTimeout * We only get to this routine if we got SIGALRM after DeadlockTimeout
* while waiting for a lock to be released by some other process. Look * while waiting for a lock to be released by some other process. Look
* to see if there's a deadlock; if not, just return and continue waiting. * to see if there's a deadlock; if not, just return and continue waiting.
* If we have a real deadlock, remove ourselves from the lock's wait queue * If we have a real deadlock, remove ourselves from the lock's wait queue
* and signal an error to ProcSleep. * and signal an error to ProcSleep.
* --------------------
*/ */
static void static void
CheckDeadLock(void) CheckDeadLock(void)
{ {
int i;
/* /*
* Acquire locktable lock. Note that the deadlock check interrupt had * Acquire exclusive lock on the entire shared lock data structures.
* better not be enabled anywhere that this process itself holds the * Must grab LWLocks in partition-number order to avoid LWLock deadlock.
* locktable lock, else this will wait forever. Also note that *
* LWLockAcquire creates a critical section, so that this routine cannot * Note that the deadlock check interrupt had better not be enabled
* be interrupted by cancel/die interrupts. * anywhere that this process itself holds lock partition locks, else this
* will wait forever. Also note that LWLockAcquire creates a critical
* section, so that this routine cannot be interrupted by cancel/die
* interrupts.
*/ */
LWLockAcquire(LockMgrLock, LW_EXCLUSIVE); for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
LWLockAcquire(FirstLockMgrLock + i, LW_EXCLUSIVE);
/* /*
* Check to see if we've been awoken by anyone in the interim. * Check to see if we've been awoken by anyone in the interim.
@ -937,14 +953,11 @@ CheckDeadLock(void)
* *
* We check by looking to see if we've been unlinked from the wait queue. * We check by looking to see if we've been unlinked from the wait queue.
* This is quicker than checking our semaphore's state, since no kernel * This is quicker than checking our semaphore's state, since no kernel
* call is needed, and it is safe because we hold the locktable lock. * call is needed, and it is safe because we hold the lock partition lock.
*/ */
if (MyProc->links.prev == INVALID_OFFSET || if (MyProc->links.prev == INVALID_OFFSET ||
MyProc->links.next == INVALID_OFFSET) MyProc->links.next == INVALID_OFFSET)
{ goto check_done;
LWLockRelease(LockMgrLock);
return;
}
#ifdef LOCK_DEBUG #ifdef LOCK_DEBUG
if (Debug_deadlocks) if (Debug_deadlocks)
@ -954,16 +967,19 @@ CheckDeadLock(void)
if (!DeadLockCheck(MyProc)) if (!DeadLockCheck(MyProc))
{ {
/* No deadlock, so keep waiting */ /* No deadlock, so keep waiting */
LWLockRelease(LockMgrLock); goto check_done;
return;
} }
/* /*
* Oops. We have a deadlock. * Oops. We have a deadlock.
* *
* Get this process out of wait state. * Get this process out of wait state. (Note: we could do this more
* efficiently by relying on lockAwaited, but use this coding to preserve
* the flexibility to kill some other transaction than the one detecting
* the deadlock.)
*/ */
RemoveFromWaitQueue(MyProc); Assert(MyProc->waitLock != NULL);
RemoveFromWaitQueue(MyProc, LockTagToPartition(&(MyProc->waitLock->tag)));
/* /*
* Set MyProc->waitStatus to STATUS_ERROR so that ProcSleep will report an * Set MyProc->waitStatus to STATUS_ERROR so that ProcSleep will report an
@ -987,7 +1003,15 @@ CheckDeadLock(void)
* them anymore. However, RemoveFromWaitQueue took care of waking up any * them anymore. However, RemoveFromWaitQueue took care of waking up any
* such processes. * such processes.
*/ */
LWLockRelease(LockMgrLock);
/*
* Release locks acquired at head of routine. Order is not critical,
* so do it back-to-front to avoid waking another CheckDeadLock instance
* before it can get all the locks.
*/
check_done:
for (i = NUM_LOCK_PARTITIONS; --i >= 0; )
LWLockRelease(FirstLockMgrLock + i);
} }

13
src/include/storage/lock.h
View File

@ -7,7 +7,7 @@
* Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group * Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group
* Portions Copyright (c) 1994, Regents of the University of California * Portions Copyright (c) 1994, Regents of the University of California
* *
* $PostgreSQL: pgsql/src/include/storage/lock.h,v 1.92 2005/12/09 01:22:04 tgl Exp $ * $PostgreSQL: pgsql/src/include/storage/lock.h,v 1.93 2005/12/11 21:02:18 tgl Exp $
* *
*------------------------------------------------------------------------- *-------------------------------------------------------------------------
*/ */
@ -19,6 +19,13 @@
#include "storage/shmem.h" #include "storage/shmem.h"
/*
* Number of partitions the shared lock tables are divided into.
*
* See LockTagToPartition() if you change this.
*/
#define NUM_LOCK_PARTITIONS 16
/* originally in procq.h */ /* originally in procq.h */
typedef struct PROC_QUEUE typedef struct PROC_QUEUE
{ {
@ -348,6 +355,7 @@ typedef struct LOCALLOCK
LOCK *lock; /* associated LOCK object in shared mem */ LOCK *lock; /* associated LOCK object in shared mem */
PROCLOCK *proclock; /* associated PROCLOCK object in shmem */ PROCLOCK *proclock; /* associated PROCLOCK object in shmem */
bool isTempObject; /* true if lock is on a temporary object */ bool isTempObject; /* true if lock is on a temporary object */
int partition; /* ID of partition containing this lock */
int nLocks; /* total number of times lock is held */ int nLocks; /* total number of times lock is held */
int numLockOwners; /* # of relevant ResourceOwners */ int numLockOwners; /* # of relevant ResourceOwners */
int maxLockOwners; /* allocated size of array */ int maxLockOwners; /* allocated size of array */
@ -389,6 +397,7 @@ typedef enum
*/ */
extern void InitLocks(void); extern void InitLocks(void);
extern LockMethod GetLocksMethodTable(const LOCK *lock); extern LockMethod GetLocksMethodTable(const LOCK *lock);
extern int LockTagToPartition(const LOCKTAG *locktag);
extern LockAcquireResult LockAcquire(const LOCKTAG *locktag, extern LockAcquireResult LockAcquire(const LOCKTAG *locktag,
bool isTempObject, bool isTempObject,
LOCKMODE lockmode, LOCKMODE lockmode,
@ -406,7 +415,7 @@ extern int LockCheckConflicts(LockMethod lockMethodTable,
LOCK *lock, PROCLOCK *proclock, PGPROC *proc); LOCK *lock, PROCLOCK *proclock, PGPROC *proc);
extern void GrantLock(LOCK *lock, PROCLOCK *proclock, LOCKMODE lockmode); extern void GrantLock(LOCK *lock, PROCLOCK *proclock, LOCKMODE lockmode);
extern void GrantAwaitedLock(void); extern void GrantAwaitedLock(void);
extern void RemoveFromWaitQueue(PGPROC *proc); extern void RemoveFromWaitQueue(PGPROC *proc, int partition);
extern Size LockShmemSize(void); extern Size LockShmemSize(void);
extern bool DeadLockCheck(PGPROC *proc); extern bool DeadLockCheck(PGPROC *proc);
extern void DeadLockReport(void); extern void DeadLockReport(void);

12
src/include/storage/lwlock.h
View File

@ -7,7 +7,7 @@
* Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group * Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group
* Portions Copyright (c) 1994, Regents of the University of California * Portions Copyright (c) 1994, Regents of the University of California
* *
* $PostgreSQL: pgsql/src/include/storage/lwlock.h,v 1.23 2005/10/15 02:49:46 momjian Exp $ * $PostgreSQL: pgsql/src/include/storage/lwlock.h,v 1.24 2005/12/11 21:02:18 tgl Exp $
* *
*------------------------------------------------------------------------- *-------------------------------------------------------------------------
*/ */
@ -16,9 +16,9 @@
/* /*
* We have a number of predefined LWLocks, plus a bunch of LWLocks that are * We have a number of predefined LWLocks, plus a bunch of LWLocks that are
* dynamically assigned (for shared buffers). The LWLock structures live * dynamically assigned (e.g., for shared buffers). The LWLock structures
* in shared memory (since they contain shared data) and are identified by * live in shared memory (since they contain shared data) and are identified
* values of this enumerated type. We abuse the notion of an enum somewhat * by values of this enumerated type. We abuse the notion of an enum somewhat
* by allowing values not listed in the enum declaration to be assigned. * by allowing values not listed in the enum declaration to be assigned.
* The extra value MaxDynamicLWLock is there to keep the compiler from * The extra value MaxDynamicLWLock is there to keep the compiler from
* deciding that the enum can be represented as char or short ... * deciding that the enum can be represented as char or short ...
@ -27,7 +27,6 @@ typedef enum LWLockId
{ {
BufMappingLock, BufMappingLock,
BufFreelistLock, BufFreelistLock,
LockMgrLock,
OidGenLock, OidGenLock,
XidGenLock, XidGenLock,
ProcArrayLock, ProcArrayLock,
@ -46,8 +45,7 @@ typedef enum LWLockId
RelCacheInitLock, RelCacheInitLock,
BgWriterCommLock, BgWriterCommLock,
TwoPhaseStateLock, TwoPhaseStateLock,
FirstLockMgrLock, /* must be last except for MaxDynamicLWLock */
NumFixedLWLocks, /* must be last except for MaxDynamicLWLock */
MaxDynamicLWLock = 1000000000 MaxDynamicLWLock = 1000000000
} LWLockId; } LWLockId;

18
src/include/storage/proc.h
View File

@ -7,7 +7,7 @@
* Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group * Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group
* Portions Copyright (c) 1994, Regents of the University of California * Portions Copyright (c) 1994, Regents of the University of California
* *
* $PostgreSQL: pgsql/src/include/storage/proc.h,v 1.84 2005/10/15 02:49:46 momjian Exp $ * $PostgreSQL: pgsql/src/include/storage/proc.h,v 1.85 2005/12/11 21:02:18 tgl Exp $
* *
*------------------------------------------------------------------------- *-------------------------------------------------------------------------
*/ */
@ -52,7 +52,8 @@ struct XidCache
* so that the prepared transactions appear to be still running and are * so that the prepared transactions appear to be still running and are
* correctly shown as holding locks. A prepared transaction PGPROC can be * correctly shown as holding locks. A prepared transaction PGPROC can be
* distinguished from a real one at need by the fact that it has pid == 0. * distinguished from a real one at need by the fact that it has pid == 0.
* The semaphore and lock-related fields in a prepared-xact PGPROC are unused. * The semaphore and lock-activity fields in a prepared-xact PGPROC are unused,
* but its myProcLocks[] lists are valid.
*/ */
struct PGPROC struct PGPROC
{ {
@ -86,8 +87,12 @@ struct PGPROC
LOCKMASK heldLocks; /* bitmask for lock types already held on this LOCKMASK heldLocks; /* bitmask for lock types already held on this
* lock object by this backend */ * lock object by this backend */
SHM_QUEUE procLocks; /* list of PROCLOCK objects for locks held or /*
* awaited by this backend */ * All PROCLOCK objects for locks held or awaited by this backend are
* linked into one of these lists, according to the partition number of
* their lock.
*/
SHM_QUEUE myProcLocks[NUM_LOCK_PARTITIONS];
struct XidCache subxids; /* cache for subtransaction XIDs */ struct XidCache subxids; /* cache for subtransaction XIDs */
}; };
@ -99,7 +104,7 @@ extern DLLIMPORT PGPROC *MyProc;
/* /*
* There is one ProcGlobal struct for the whole installation. * There is one ProcGlobal struct for the whole database cluster.
*/ */
typedef struct PROC_HDR typedef struct PROC_HDR
{ {
@ -134,8 +139,7 @@ extern bool HaveNFreeProcs(int n);
extern void ProcReleaseLocks(bool isCommit); extern void ProcReleaseLocks(bool isCommit);
extern void ProcQueueInit(PROC_QUEUE *queue); extern void ProcQueueInit(PROC_QUEUE *queue);
extern int ProcSleep(LockMethod lockMethodTable, LOCKMODE lockmode, extern int ProcSleep(LOCALLOCK *locallock, LockMethod lockMethodTable);
LOCK *lock, PROCLOCK *proclock);
extern PGPROC *ProcWakeup(PGPROC *proc, int waitStatus); extern PGPROC *ProcWakeup(PGPROC *proc, int waitStatus);
extern void ProcLockWakeup(LockMethod lockMethodTable, LOCK *lock); extern void ProcLockWakeup(LockMethod lockMethodTable, LOCK *lock);
extern bool LockWaitCancel(void); extern bool LockWaitCancel(void);