Here's a new version. The condition variable patch 0001 fixes a bug: CleanupProcSignalState() also needs to broadcast. The hunk that allows the interrupt handlers to use CVs while you're already waiting on a CV is now in a separate patch 0002. I'm thinking of going ahead and committing those two. The 0003 patch to achieve $SUBJECT needs more discussion.
From 51bb33c8755efa11d24595ed0a27fa0b387e8153 Mon Sep 17 00:00:00 2001 From: Thomas Munro <thomas.mu...@gmail.com> Date: Sat, 27 Feb 2021 15:40:11 +1300 Subject: [PATCH v4 1/3] Use condition variables for ProcSignalBarriers.
Instead of a poll/sleep loop, use a condition variable for precise wake-up whenever a backend's pss_barrierGeneration advances. Discussion: https://postgr.es/m/ca+hukgldemy2gbm80kz20gte6hnvwoere8kwcjk6-u56ost...@mail.gmail.com --- src/backend/storage/ipc/procsignal.c | 34 ++++++++++++---------------- 1 file changed, 14 insertions(+), 20 deletions(-) diff --git a/src/backend/storage/ipc/procsignal.c b/src/backend/storage/ipc/procsignal.c index c43cdd685b..aa0362216b 100644 --- a/src/backend/storage/ipc/procsignal.c +++ b/src/backend/storage/ipc/procsignal.c @@ -23,6 +23,7 @@ #include "miscadmin.h" #include "pgstat.h" #include "replication/walsender.h" +#include "storage/condition_variable.h" #include "storage/ipc.h" #include "storage/latch.h" #include "storage/proc.h" @@ -59,10 +60,11 @@ */ typedef struct { - pid_t pss_pid; - sig_atomic_t pss_signalFlags[NUM_PROCSIGNALS]; + volatile pid_t pss_pid; + volatile sig_atomic_t pss_signalFlags[NUM_PROCSIGNALS]; pg_atomic_uint64 pss_barrierGeneration; pg_atomic_uint32 pss_barrierCheckMask; + ConditionVariable pss_barrierCV; } ProcSignalSlot; /* @@ -93,7 +95,7 @@ typedef struct ((flags) &= ~(((uint32) 1) << (uint32) (type))) static ProcSignalHeader *ProcSignal = NULL; -static volatile ProcSignalSlot *MyProcSignalSlot = NULL; +static ProcSignalSlot *MyProcSignalSlot = NULL; static bool CheckProcSignal(ProcSignalReason reason); static void CleanupProcSignalState(int status, Datum arg); @@ -142,6 +144,7 @@ ProcSignalShmemInit(void) MemSet(slot->pss_signalFlags, 0, sizeof(slot->pss_signalFlags)); pg_atomic_init_u64(&slot->pss_barrierGeneration, PG_UINT64_MAX); pg_atomic_init_u32(&slot->pss_barrierCheckMask, 0); + ConditionVariableInit(&slot->pss_barrierCV); } } } @@ -156,7 +159,7 @@ ProcSignalShmemInit(void) void ProcSignalInit(int pss_idx) { - volatile ProcSignalSlot *slot; + ProcSignalSlot *slot; uint64 barrier_generation; Assert(pss_idx >= 1 && pss_idx <= NumProcSignalSlots); @@ -208,7 +211,7 @@ static void CleanupProcSignalState(int status, Datum arg) { int pss_idx = DatumGetInt32(arg); - volatile ProcSignalSlot *slot; + ProcSignalSlot *slot; slot = &ProcSignal->psh_slot[pss_idx - 1]; Assert(slot == MyProcSignalSlot); @@ -237,6 +240,7 @@ CleanupProcSignalState(int status, Datum arg) * no barrier waits block on it. */ pg_atomic_write_u64(&slot->pss_barrierGeneration, PG_UINT64_MAX); + ConditionVariableBroadcast(&slot->pss_barrierCV); slot->pss_pid = 0; } @@ -391,13 +395,11 @@ EmitProcSignalBarrier(ProcSignalBarrierType type) void WaitForProcSignalBarrier(uint64 generation) { - long timeout = 125L; - Assert(generation <= pg_atomic_read_u64(&ProcSignal->psh_barrierGeneration)); for (int i = NumProcSignalSlots - 1; i >= 0; i--) { - volatile ProcSignalSlot *slot = &ProcSignal->psh_slot[i]; + ProcSignalSlot *slot = &ProcSignal->psh_slot[i]; uint64 oldval; /* @@ -409,20 +411,11 @@ WaitForProcSignalBarrier(uint64 generation) oldval = pg_atomic_read_u64(&slot->pss_barrierGeneration); while (oldval < generation) { - int events; - - CHECK_FOR_INTERRUPTS(); - - events = - WaitLatch(MyLatch, - WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH, - timeout, WAIT_EVENT_PROC_SIGNAL_BARRIER); - ResetLatch(MyLatch); - + ConditionVariableSleep(&slot->pss_barrierCV, + WAIT_EVENT_PROC_SIGNAL_BARRIER); oldval = pg_atomic_read_u64(&slot->pss_barrierGeneration); - if (events & WL_TIMEOUT) - timeout = Min(timeout * 2, 1000L); } + ConditionVariableCancelSleep(); } /* @@ -589,6 +582,7 @@ ProcessProcSignalBarrier(void) * next called. */ pg_atomic_write_u64(&MyProcSignalSlot->pss_barrierGeneration, shared_gen); + ConditionVariableBroadcast(&MyProcSignalSlot->pss_barrierCV); } /* -- 2.30.0
From 9fbb46cd8909d8058d32421e0e41c223d710d7d4 Mon Sep 17 00:00:00 2001 From: Thomas Munro <thomas.mu...@gmail.com> Date: Sat, 27 Feb 2021 15:36:22 +1300 Subject: [PATCH v4 2/3] Allow condition variables to be used in interrupt code. Adjust the condition variable sleep loop to work correctly when code reached by its internal CHECK_FOR_INTERRUPTS() call interacts with another condition variable. There are no such cases currently, but a proposed patch would do this. Discussion: https://postgr.es/m/ca+hukgldemy2gbm80kz20gte6hnvwoere8kwcjk6-u56ost...@mail.gmail.com --- src/backend/storage/lmgr/condition_variable.c | 11 +++++++++-- 1 file changed, 9 insertions(+), 2 deletions(-) diff --git a/src/backend/storage/lmgr/condition_variable.c b/src/backend/storage/lmgr/condition_variable.c index 0a61ff0031..80d70c154c 100644 --- a/src/backend/storage/lmgr/condition_variable.c +++ b/src/backend/storage/lmgr/condition_variable.c @@ -165,8 +165,6 @@ ConditionVariableTimedSleep(ConditionVariable *cv, long timeout, /* Reset latch before examining the state of the wait list. */ ResetLatch(MyLatch); - CHECK_FOR_INTERRUPTS(); - /* * If this process has been taken out of the wait list, then we know * that it has been signaled by ConditionVariableSignal (or @@ -190,6 +188,15 @@ ConditionVariableTimedSleep(ConditionVariable *cv, long timeout, } SpinLockRelease(&cv->mutex); + /* + * Check for interrupts, and return spuriously if that caused the + * current sleep target to change (meaning that interrupt handler code + * waited for a different condition variable). + */ + CHECK_FOR_INTERRUPTS(); + if (cv != cv_sleep_target) + done = true; + /* We were signaled, so return */ if (done) return false; -- 2.30.0
From 4bfe8d6fcc05175a481b98e0f91d59ee2bf36fc4 Mon Sep 17 00:00:00 2001 From: Thomas Munro <thomas.mu...@gmail.com> Date: Tue, 2 Feb 2021 11:07:06 +1300 Subject: [PATCH v4 3/3] Use a global barrier to fix DROP TABLESPACE on Windows. Previously, it was possible for DROP TABLESPACE to fail because a table had been dropped, but other backends still had open file handles. Windows won't allow a directory containing unlinked-but-still-open files to be unlinked. Tackle this problem by forcing all backends to close all fds. No change for Unix systems, which didn't suffer from the problem, but simulate the need to do it in assertion builds just to exercise the code. XXX maybe only do this after 2nd failure? Reviewed-by: Andres Freund <and...@anarazel.de> Discussion: https://postgr.es/m/ca+hukgldemy2gbm80kz20gte6hnvwoere8kwcjk6-u56ost...@mail.gmail.com --- src/backend/commands/tablespace.c | 22 +++++++++++++++------- src/backend/storage/ipc/procsignal.c | 22 ++++++---------------- src/backend/storage/smgr/md.c | 6 ++++++ src/backend/storage/smgr/smgr.c | 12 ++++++++++++ src/include/storage/md.h | 1 + src/include/storage/procsignal.h | 7 +------ src/include/storage/smgr.h | 1 + 7 files changed, 42 insertions(+), 29 deletions(-) diff --git a/src/backend/commands/tablespace.c b/src/backend/commands/tablespace.c index 69ea155d50..8853f1b658 100644 --- a/src/backend/commands/tablespace.c +++ b/src/backend/commands/tablespace.c @@ -520,15 +520,23 @@ DropTableSpace(DropTableSpaceStmt *stmt) * but we can't tell them apart from important data files that we * mustn't delete. So instead, we force a checkpoint which will clean * out any lingering files, and try again. - * - * XXX On Windows, an unlinked file persists in the directory listing - * until no process retains an open handle for the file. The DDL - * commands that schedule files for unlink send invalidation messages - * directing other PostgreSQL processes to close the files. DROP - * TABLESPACE should not give up on the tablespace becoming empty - * until all relevant invalidation processing is complete. */ RequestCheckpoint(CHECKPOINT_IMMEDIATE | CHECKPOINT_FORCE | CHECKPOINT_WAIT); + /* + * On Windows, an unlinked file persists in the directory listing until + * no process retains an open handle for the file. The DDL + * commands that schedule files for unlink send invalidation messages + * directing other PostgreSQL processes to close the files, but nothing + * guarantees they'll be processed in time. So, we'll also use a + * global barrier to ask all backends to close all files, and wait + * until they're finished. + */ +#if defined(WIN32) || defined(USE_ASSERT_CHECKING) + LWLockRelease(TablespaceCreateLock); + WaitForProcSignalBarrier(EmitProcSignalBarrier(PROCSIGNAL_BARRIER_SMGRRELEASE)); + LWLockAcquire(TablespaceCreateLock, LW_EXCLUSIVE); +#endif + /* And now try again. */ if (!destroy_tablespace_directories(tablespaceoid, false)) { /* Still not empty, the files must be important then */ diff --git a/src/backend/storage/ipc/procsignal.c b/src/backend/storage/ipc/procsignal.c index aa0362216b..0a61a70000 100644 --- a/src/backend/storage/ipc/procsignal.c +++ b/src/backend/storage/ipc/procsignal.c @@ -28,6 +28,7 @@ #include "storage/latch.h" #include "storage/proc.h" #include "storage/shmem.h" +#include "storage/smgr.h" #include "storage/sinval.h" #include "tcop/tcopprot.h" @@ -100,7 +101,7 @@ static ProcSignalSlot *MyProcSignalSlot = NULL; static bool CheckProcSignal(ProcSignalReason reason); static void CleanupProcSignalState(int status, Datum arg); static void ResetProcSignalBarrierBits(uint32 flags); -static bool ProcessBarrierPlaceholder(void); +static bool ProcessBarrierSmgrRelease(void); /* * ProcSignalShmemSize @@ -531,8 +532,8 @@ ProcessProcSignalBarrier(void) type = (ProcSignalBarrierType) pg_rightmost_one_pos32(flags); switch (type) { - case PROCSIGNAL_BARRIER_PLACEHOLDER: - processed = ProcessBarrierPlaceholder(); + case PROCSIGNAL_BARRIER_SMGRRELEASE: + processed = ProcessBarrierSmgrRelease(); break; } @@ -599,20 +600,9 @@ ResetProcSignalBarrierBits(uint32 flags) } static bool -ProcessBarrierPlaceholder(void) +ProcessBarrierSmgrRelease(void) { - /* - * XXX. This is just a placeholder until the first real user of this - * machinery gets committed. Rename PROCSIGNAL_BARRIER_PLACEHOLDER to - * PROCSIGNAL_BARRIER_SOMETHING_ELSE where SOMETHING_ELSE is something - * appropriately descriptive. Get rid of this function and instead have - * ProcessBarrierSomethingElse. Most likely, that function should live in - * the file pertaining to that subsystem, rather than here. - * - * The return value should be 'true' if the barrier was successfully - * absorbed and 'false' if not. Note that returning 'false' can lead to - * very frequent retries, so try hard to make that an uncommon case. - */ + smgrrelease(); return true; } diff --git a/src/backend/storage/smgr/md.c b/src/backend/storage/smgr/md.c index 1e12cfad8e..1b751205b1 100644 --- a/src/backend/storage/smgr/md.c +++ b/src/backend/storage/smgr/md.c @@ -548,6 +548,12 @@ mdclose(SMgrRelation reln, ForkNumber forknum) } } +void +mdrelease(void) +{ + closeAllVfds(); +} + /* * mdprefetch() -- Initiate asynchronous read of the specified block of a relation */ diff --git a/src/backend/storage/smgr/smgr.c b/src/backend/storage/smgr/smgr.c index 4dc24649df..b9acff73d2 100644 --- a/src/backend/storage/smgr/smgr.c +++ b/src/backend/storage/smgr/smgr.c @@ -41,6 +41,7 @@ typedef struct f_smgr { void (*smgr_init) (void); /* may be NULL */ void (*smgr_shutdown) (void); /* may be NULL */ + void (*smgr_release) (void); /* may be NULL */ void (*smgr_open) (SMgrRelation reln); void (*smgr_close) (SMgrRelation reln, ForkNumber forknum); void (*smgr_create) (SMgrRelation reln, ForkNumber forknum, @@ -69,6 +70,7 @@ static const f_smgr smgrsw[] = { { .smgr_init = mdinit, .smgr_shutdown = NULL, + .smgr_release = mdrelease, .smgr_open = mdopen, .smgr_close = mdclose, .smgr_create = mdcreate, @@ -298,6 +300,16 @@ smgrcloseall(void) smgrclose(reln); } +void +smgrrelease(void) +{ + for (int i = 0; i < NSmgr; i++) + { + if (smgrsw[i].smgr_release) + smgrsw[i].smgr_release(); + } +} + /* * smgrclosenode() -- Close SMgrRelation object for given RelFileNode, * if one exists. diff --git a/src/include/storage/md.h b/src/include/storage/md.h index 752b440864..7685367b97 100644 --- a/src/include/storage/md.h +++ b/src/include/storage/md.h @@ -23,6 +23,7 @@ extern void mdinit(void); extern void mdopen(SMgrRelation reln); extern void mdclose(SMgrRelation reln, ForkNumber forknum); +extern void mdrelease(void); extern void mdcreate(SMgrRelation reln, ForkNumber forknum, bool isRedo); extern bool mdexists(SMgrRelation reln, ForkNumber forknum); extern void mdunlink(RelFileNodeBackend rnode, ForkNumber forknum, bool isRedo); diff --git a/src/include/storage/procsignal.h b/src/include/storage/procsignal.h index 4ae7dc33b8..aaaa8ed07c 100644 --- a/src/include/storage/procsignal.h +++ b/src/include/storage/procsignal.h @@ -48,12 +48,7 @@ typedef enum typedef enum { - /* - * XXX. PROCSIGNAL_BARRIER_PLACEHOLDER should be replaced when the first - * real user of the ProcSignalBarrier mechanism is added. It's just here - * for now because we can't have an empty enum. - */ - PROCSIGNAL_BARRIER_PLACEHOLDER = 0 + PROCSIGNAL_BARRIER_SMGRRELEASE /* ask smgr to close files */ } ProcSignalBarrierType; /* diff --git a/src/include/storage/smgr.h b/src/include/storage/smgr.h index a6fbf7b6a6..1a21bd9e88 100644 --- a/src/include/storage/smgr.h +++ b/src/include/storage/smgr.h @@ -84,6 +84,7 @@ extern void smgrsetowner(SMgrRelation *owner, SMgrRelation reln); extern void smgrclearowner(SMgrRelation *owner, SMgrRelation reln); extern void smgrclose(SMgrRelation reln); extern void smgrcloseall(void); +extern void smgrrelease(void); extern void smgrclosenode(RelFileNodeBackend rnode); extern void smgrcreate(SMgrRelation reln, ForkNumber forknum, bool isRedo); extern void smgrdosyncall(SMgrRelation *rels, int nrels); -- 2.30.0
