lwlock: Fix quadratic behavior with very long wait lists

Until now LWLockDequeueSelf() sequentially searched the list of waiters to see
if the current proc is still is on the list of waiters, or has already been
removed. In extreme workloads, where the wait lists are very long, this leads
to a quadratic behavior. #backends iterating over a list #backends
long. Additionally, the likelihood of needing to call LWLockDequeueSelf() in
the first place also increases with the increased length of the wait queue, as
it becomes more likely that a lock is released while waiting for the wait list
lock, which is held for longer during lock release.

Due to the exponential back-off in perform_spin_delay() this is surprisingly
hard to detect. We should make that easier, e.g. by adding a wait event around
the pg_usleep() - but that's a separate patch.

The fix is simple - track whether a proc is currently waiting in the wait list
or already removed but waiting to be woken up in PGPROC->lwWaiting.

In some workloads with a lot of clients contending for a small number of
lwlocks (e.g. WALWriteLock), the fix can substantially increase throughput.

This has been originally fixed for 16~ with a4adc31f6902 without a
backpatch, and we have heard complaints from users impacted by this
quadratic behavior in older versions as well.

Author: Andres Freund 
Reviewed-by: Bharath Rupireddy 
Discussion: https://postgr.es/m/20221027165914[email protected]
Discussion: https://postgr.es/m/CALj2ACXktNbG=K8Xi7PSqbofTZozavhaxjatVc14iYaLu4Maag@mail.gmail.com
Backpatch-through: 12

diff --git a/src/backend/access/transam/twophase.c b/src/backend/access/transam/twophase.c
index 076315e7a28ed47211abcdec95a5ea146f53d1f4..db68b1444319d76047959a38c196f370183a4578 100644 (file)
--- a/src/backend/access/transam/twophase.c
+++ b/src/backend/access/transam/twophase.c
@@ -482,7 +482,7 @@ MarkAsPreparingGuts(GlobalTransaction gxact, TransactionId xid, const char *gid,
    proc->roleId = owner;
    proc->tempNamespaceId = InvalidOid;
    proc->isBackgroundWorker = false;
-   proc->lwWaiting = false;
+   proc->lwWaiting = LW_WS_NOT_WAITING;
    proc->lwWaitMode = 0;
    proc->waitLock = NULL;
    proc->waitProcLock = NULL;

diff --git a/src/backend/storage/lmgr/lwlock.c b/src/backend/storage/lmgr/lwlock.c
index d63dc1b93ead39e0abdd523b0ebf2dd453ce2589..28a540961eab70e469859bb8e1bfb90252dd80ac 100644 (file)
--- a/src/backend/storage/lmgr/lwlock.c
+++ b/src/backend/storage/lmgr/lwlock.c
@@ -993,6 +993,15 @@ LWLockWakeup(LWLock *lock)
            wokeup_somebody = true;
        }
 
+       /*
+        * Signal that the process isn't on the wait list anymore. This allows
+        * LWLockDequeueSelf() to remove itself of the waitlist with a
+        * proclist_delete(), rather than having to check if it has been
+        * removed from the list.
+        */
+       Assert(waiter->lwWaiting == LW_WS_WAITING);
+       waiter->lwWaiting = LW_WS_PENDING_WAKEUP;
+
        /*
         * Once we've woken up an exclusive lock, there's no point in waking
         * up anybody else.
@@ -1050,7 +1059,7 @@ LWLockWakeup(LWLock *lock)
         * another lock.
         */
        pg_write_barrier();
-       waiter->lwWaiting = false;
+       waiter->lwWaiting = LW_WS_NOT_WAITING;
        PGSemaphoreUnlock(waiter->sem);
    }
 }
@@ -1071,7 +1080,7 @@ LWLockQueueSelf(LWLock *lock, LWLockMode mode)
    if (MyProc == NULL)
        elog(PANIC, "cannot wait without a PGPROC structure");
 
-   if (MyProc->lwWaiting)
+   if (MyProc->lwWaiting != LW_WS_NOT_WAITING)
        elog(PANIC, "queueing for lock while waiting on another one");
 
    LWLockWaitListLock(lock);
@@ -1079,7 +1088,7 @@ LWLockQueueSelf(LWLock *lock, LWLockMode mode)
    /* setting the flag is protected by the spinlock */
    pg_atomic_fetch_or_u32(&lock->state, LW_FLAG_HAS_WAITERS);
 
-   MyProc->lwWaiting = true;
+   MyProc->lwWaiting = LW_WS_WAITING;
    MyProc->lwWaitMode = mode;
 
    /* LW_WAIT_UNTIL_FREE waiters are always at the front of the queue */
@@ -1107,8 +1116,7 @@ LWLockQueueSelf(LWLock *lock, LWLockMode mode)
 static void
 LWLockDequeueSelf(LWLock *lock)
 {
-   bool        found = false;
-   proclist_mutable_iter iter;
+   bool        on_waitlist;
 
 #ifdef LWLOCK_STATS
    lwlock_stats *lwstats;
@@ -1121,18 +1129,13 @@ LWLockDequeueSelf(LWLock *lock)
    LWLockWaitListLock(lock);
 
    /*
-    * Can't just remove ourselves from the list, but we need to iterate over
-    * all entries as somebody else could have dequeued us.
+    * Remove ourselves from the waitlist, unless we've already been
+    * removed. The removal happens with the wait list lock held, so there's
+    * no race in this check.
     */
-   proclist_foreach_modify(iter, &lock->waiters, lwWaitLink)
-   {
-       if (iter.cur == MyProc->pgprocno)
-       {
-           found = true;
-           proclist_delete(&lock->waiters, iter.cur, lwWaitLink);
-           break;
-       }
-   }
+   on_waitlist = MyProc->lwWaiting == LW_WS_WAITING;
+   if (on_waitlist)
+       proclist_delete(&lock->waiters, MyProc->pgprocno, lwWaitLink);
 
    if (proclist_is_empty(&lock->waiters) &&
        (pg_atomic_read_u32(&lock->state) & LW_FLAG_HAS_WAITERS) != 0)
@@ -1144,8 +1147,8 @@ LWLockDequeueSelf(LWLock *lock)
    LWLockWaitListUnlock(lock);
 
    /* clear waiting state again, nice for debugging */
-   if (found)
-       MyProc->lwWaiting = false;
+   if (on_waitlist)
+       MyProc->lwWaiting = LW_WS_NOT_WAITING;
    else
    {
        int         extraWaits = 0;
@@ -1169,7 +1172,7 @@ LWLockDequeueSelf(LWLock *lock)
        for (;;)
        {
            PGSemaphoreLock(MyProc->sem);
-           if (!MyProc->lwWaiting)
+           if (MyProc->lwWaiting == LW_WS_NOT_WAITING)
                break;
            extraWaits++;
        }
@@ -1320,7 +1323,7 @@ LWLockAcquire(LWLock *lock, LWLockMode mode)
        for (;;)
        {
            PGSemaphoreLock(proc->sem);
-           if (!proc->lwWaiting)
+           if (proc->lwWaiting == LW_WS_NOT_WAITING)
                break;
            extraWaits++;
        }
@@ -1485,7 +1488,7 @@ LWLockAcquireOrWait(LWLock *lock, LWLockMode mode)
            for (;;)
            {
                PGSemaphoreLock(proc->sem);
-               if (!proc->lwWaiting)
+               if (proc->lwWaiting == LW_WS_NOT_WAITING)
                    break;
                extraWaits++;
            }
@@ -1701,7 +1704,7 @@ LWLockWaitForVar(LWLock *lock, uint64 *valptr, uint64 oldval, uint64 *newval)
        for (;;)
        {
            PGSemaphoreLock(proc->sem);
-           if (!proc->lwWaiting)
+           if (proc->lwWaiting == LW_WS_NOT_WAITING)
                break;
            extraWaits++;
        }
@@ -1779,6 +1782,10 @@ LWLockUpdateVar(LWLock *lock, uint64 *valptr, uint64 val)
 
        proclist_delete(&lock->waiters, iter.cur, lwWaitLink);
        proclist_push_tail(&wakeup, iter.cur, lwWaitLink);
+
+       /* see LWLockWakeup() */
+       Assert(waiter->lwWaiting == LW_WS_WAITING);
+       waiter->lwWaiting = LW_WS_PENDING_WAKEUP;
    }
 
    /* We are done updating shared state of the lock itself. */
@@ -1794,7 +1801,7 @@ LWLockUpdateVar(LWLock *lock, uint64 *valptr, uint64 val)
        proclist_delete(&wakeup, iter.cur, lwWaitLink);
        /* check comment in LWLockWakeup() about this barrier */
        pg_write_barrier();
-       waiter->lwWaiting = false;
+       waiter->lwWaiting = LW_WS_NOT_WAITING;
        PGSemaphoreUnlock(waiter->sem);
    }
 }

diff --git a/src/backend/storage/lmgr/proc.c b/src/backend/storage/lmgr/proc.c
index 1e06d6580ce02b3be82e1be9764981aa28f83920..f62e622a7bf1e134a015bb12d2aa4278c9f6ca1f 100644 (file)
--- a/src/backend/storage/lmgr/proc.c
+++ b/src/backend/storage/lmgr/proc.c
@@ -400,7 +400,7 @@ InitProcess(void)
    /* NB -- autovac launcher intentionally does not set IS_AUTOVACUUM */
    if (IsAutoVacuumWorkerProcess())
        MyProc->statusFlags |= PROC_IS_AUTOVACUUM;
-   MyProc->lwWaiting = false;
+   MyProc->lwWaiting = LW_WS_NOT_WAITING;
    MyProc->lwWaitMode = 0;
    MyProc->waitLock = NULL;
    MyProc->waitProcLock = NULL;
@@ -583,7 +583,7 @@ InitAuxiliaryProcess(void)
    MyProc->delayChkpt = false;
    MyProc->delayChkptEnd = false;
    MyProc->statusFlags = 0;
-   MyProc->lwWaiting = false;
+   MyProc->lwWaiting = LW_WS_NOT_WAITING;
    MyProc->lwWaitMode = 0;
    MyProc->waitLock = NULL;
    MyProc->waitProcLock = NULL;

diff --git a/src/include/storage/lwlock.h b/src/include/storage/lwlock.h
index 78afe57fb2aa53b05bd9ec9ff5a60756cb56eae3..b2c9a9dbbcc7e4444be4c07e408b004d6ff08e7d 100644 (file)
--- a/src/include/storage/lwlock.h
+++ b/src/include/storage/lwlock.h
@@ -23,6 +23,14 @@
 
 struct PGPROC;
 
+/* what state of the wait process is a backend in */
+typedef enum LWLockWaitState
+{
+   LW_WS_NOT_WAITING, /* not currently waiting / woken up */
+   LW_WS_WAITING, /* currently waiting */
+   LW_WS_PENDING_WAKEUP /* removed from waitlist, but not yet signalled */
+} LWLockWaitState;
+
 /*
  * Code outside of lwlock.c should not manipulate the contents of this
  * structure directly, but we have to declare it here to allow LWLocks to be

diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 1464fad9b943ea7aefb7b2b8a1352cc210519c5f..9cc37aa0d93fca373a5e489cb35ab723fd737e41 100644 (file)
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -180,7 +180,7 @@ struct PGPROC
    bool        recoveryConflictPending;
 
    /* Info about LWLock the process is currently waiting for, if any. */
-   bool        lwWaiting;      /* true if waiting for an LW lock */
+   uint8       lwWaiting;      /* see LWLockWaitState */
    uint8       lwWaitMode;     /* lwlock mode being waited for */
    proclist_node lwWaitLink;   /* position in LW lock wait list */