Use a ResourceOwner to track buffer pins in all cases.

Historically, we've allowed auxiliary processes to take buffer pins without
tracking them in a ResourceOwner.  However, that creates problems for error
recovery.  In particular, we've seen multiple reports of assertion crashes
in the startup process when it gets an error while holding a buffer pin,
as for example if it gets ENOSPC during a write.  In a non-assert build,
the process would simply exit without releasing the pin at all.  We've
gotten away with that so far just because a failure exit of the startup
process translates to a database crash anyhow; but any similar behavior
in other aux processes could result in stuck pins and subsequent problems
in vacuum.

To improve this, institute a policy that we must *always* have a resowner
backing any attempt to pin a buffer, which we can enforce just by removing
the previous special-case code in resowner.c.  Add infrastructure to make
it easy to create a process-lifespan AuxProcessResourceOwner and clear
out its contents at appropriate times.  Replace existing ad-hoc resowner
management in bgwriter.c and other aux processes with that.  (Thus, while
the startup process gains a resowner where it had none at all before, some
other aux process types are replacing an ad-hoc resowner with this code.)
Also use the AuxProcessResourceOwner to manage buffer pins taken during
StartupXLOG and ShutdownXLOG, even when those are being run in a bootstrap
process or a standalone backend rather than a true auxiliary process.

In passing, remove some other ad-hoc resource owner creations that had
gotten cargo-culted into various other places.  As far as I can tell
that was all unnecessary, and if it had been necessary it was incomplete,
due to lacking any provision for clearing those resowners later.
(Also worth noting in this connection is that a process that hasn't called
InitBufferPoolBackend has no business accessing buffers; so there's more
to do than just add the resowner if we want to touch buffers in processes
not covered by this patch.)

Although this fixes a very old bug, no back-patch, because there's no
evidence of any significant problem in non-assert builds.

Patch by me, pursuant to a report from Justin Pryzby.  Thanks to
Robert Haas and Kyotaro Horiguchi for reviews.

Discussion: https://postgr.es/m/20180627233939.GA10276@telsasoft.com
This commit is contained in:
Tom Lane 2018-07-18 12:15:16 -04:00
parent 6b387179ba
commit 3cb646264e
16 changed files with 142 additions and 90 deletions

View File

@ -37,7 +37,6 @@
#include "utils/guc.h"
#include "utils/inval.h"
#include "utils/memutils.h"
#include "utils/resowner.h"
#include "utils/snapmgr.h"
#include "utils/typcache.h"
@ -1220,16 +1219,19 @@ ParallelWorkerMain(Datum main_arg)
Assert(ParallelWorkerNumber == -1);
memcpy(&ParallelWorkerNumber, MyBgworkerEntry->bgw_extra, sizeof(int));
/* Set up a memory context and resource owner. */
Assert(CurrentResourceOwner == NULL);
CurrentResourceOwner = ResourceOwnerCreate(NULL, "parallel toplevel");
/* Set up a memory context to work in, just for cleanliness. */
CurrentMemoryContext = AllocSetContextCreate(TopMemoryContext,
"Parallel worker",
ALLOCSET_DEFAULT_SIZES);
/*
* Now that we have a resource owner, we can attach to the dynamic shared
* memory segment and read the table of contents.
* Attach to the dynamic shared memory segment for the parallel query, and
* find its table of contents.
*
* Note: at this point, we have not created any ResourceOwner in this
* process. This will result in our DSM mapping surviving until process
* exit, which is fine. If there were a ResourceOwner, it would acquire
* ownership of the mapping, but we have no need for that.
*/
seg = dsm_attach(DatumGetUInt32(main_arg));
if (seg == NULL)
@ -1393,7 +1395,7 @@ ParallelWorkerMain(Datum main_arg)
/* Must exit parallel mode to pop active snapshot. */
ExitParallelMode();
/* Must pop active snapshot so resowner.c doesn't complain. */
/* Must pop active snapshot so snapmgr.c doesn't complain. */
PopActiveSnapshot();
/* Shut down the parallel-worker transaction. */

View File

@ -6360,6 +6360,15 @@ StartupXLOG(void)
bool fast_promoted = false;
struct stat st;
/*
* We should have an aux process resource owner to use, and we should not
* be in a transaction that's installed some other resowner.
*/
Assert(AuxProcessResourceOwner != NULL);
Assert(CurrentResourceOwner == NULL ||
CurrentResourceOwner == AuxProcessResourceOwner);
CurrentResourceOwner = AuxProcessResourceOwner;
/*
* Verify XLOG status looks valid.
*/
@ -8472,6 +8481,15 @@ GetNextXidAndEpoch(TransactionId *xid, uint32 *epoch)
void
ShutdownXLOG(int code, Datum arg)
{
/*
* We should have an aux process resource owner to use, and we should not
* be in a transaction that's installed some other resowner.
*/
Assert(AuxProcessResourceOwner != NULL);
Assert(CurrentResourceOwner == NULL ||
CurrentResourceOwner == AuxProcessResourceOwner);
CurrentResourceOwner = AuxProcessResourceOwner;
/* Don't be chatty in standalone mode */
ereport(IsPostmasterEnvironment ? LOG : NOTICE,
(errmsg("shutting down")));

View File

@ -403,6 +403,13 @@ AuxiliaryProcessMain(int argc, char *argv[])
/* finish setting up bufmgr.c */
InitBufferPoolBackend();
/*
* Auxiliary processes don't run transactions, but they may need a
* resource owner anyway to manage buffer pins acquired outside
* transactions (and, perhaps, other things in future).
*/
CreateAuxProcessResourceOwner();
/* Initialize backend status information */
pgstat_initialize();
pgstat_bestart();

View File

@ -522,13 +522,9 @@ AutoVacLauncherMain(int argc, char *argv[])
pgstat_report_wait_end();
AbortBufferIO();
UnlockBuffers();
if (CurrentResourceOwner)
{
ResourceOwnerRelease(CurrentResourceOwner,
RESOURCE_RELEASE_BEFORE_LOCKS,
false, true);
/* we needn't bother with the other ResourceOwnerRelease phases */
}
/* this is probably dead code, but let's be safe: */
if (AuxProcessResourceOwner)
ReleaseAuxProcessResources(false);
AtEOXact_Buffers(false);
AtEOXact_SMgr();
AtEOXact_Files(false);

View File

@ -141,12 +141,6 @@ BackgroundWriterMain(void)
/* We allow SIGQUIT (quickdie) at all times */
sigdelset(&BlockSig, SIGQUIT);
/*
* Create a resource owner to keep track of our resources (currently only
* buffer pins).
*/
CurrentResourceOwner = ResourceOwnerCreate(NULL, "Background Writer");
/*
* We just started, assume there has been either a shutdown or
* end-of-recovery snapshot.
@ -191,11 +185,7 @@ BackgroundWriterMain(void)
ConditionVariableCancelSleep();
AbortBufferIO();
UnlockBuffers();
/* buffer pins are released here: */
ResourceOwnerRelease(CurrentResourceOwner,
RESOURCE_RELEASE_BEFORE_LOCKS,
false, true);
/* we needn't bother with the other ResourceOwnerRelease phases */
ReleaseAuxProcessResources(false);
AtEOXact_Buffers(false);
AtEOXact_SMgr();
AtEOXact_Files(false);

View File

@ -231,12 +231,6 @@ CheckpointerMain(void)
*/
last_checkpoint_time = last_xlog_switch_time = (pg_time_t) time(NULL);
/*
* Create a resource owner to keep track of our resources (currently only
* buffer pins).
*/
CurrentResourceOwner = ResourceOwnerCreate(NULL, "Checkpointer");
/*
* Create a memory context that we will do all our work in. We do this so
* that we can reset the context during error recovery and thereby avoid
@ -275,11 +269,7 @@ CheckpointerMain(void)
pgstat_report_wait_end();
AbortBufferIO();
UnlockBuffers();
/* buffer pins are released here: */
ResourceOwnerRelease(CurrentResourceOwner,
RESOURCE_RELEASE_BEFORE_LOCKS,
false, true);
/* we needn't bother with the other ResourceOwnerRelease phases */
ReleaseAuxProcessResources(false);
AtEOXact_Buffers(false);
AtEOXact_SMgr();
AtEOXact_Files(false);

View File

@ -129,12 +129,6 @@ WalWriterMain(void)
/* We allow SIGQUIT (quickdie) at all times */
sigdelset(&BlockSig, SIGQUIT);
/*
* Create a resource owner to keep track of our resources (not clear that
* we need this, but may as well have one).
*/
CurrentResourceOwner = ResourceOwnerCreate(NULL, "Wal Writer");
/*
* Create a memory context that we will do all our work in. We do this so
* that we can reset the context during error recovery and thereby avoid
@ -172,11 +166,7 @@ WalWriterMain(void)
pgstat_report_wait_end();
AbortBufferIO();
UnlockBuffers();
/* buffer pins are released here: */
ResourceOwnerRelease(CurrentResourceOwner,
RESOURCE_RELEASE_BEFORE_LOCKS,
false, true);
/* we needn't bother with the other ResourceOwnerRelease phases */
ReleaseAuxProcessResources(false);
AtEOXact_Buffers(false);
AtEOXact_SMgr();
AtEOXact_Files(false);

View File

@ -279,8 +279,6 @@ pg_logical_slot_get_changes_guts(FunctionCallInfo fcinfo, bool confirm, bool bin
*/
startptr = MyReplicationSlot->data.restart_lsn;
CurrentResourceOwner = ResourceOwnerCreate(CurrentResourceOwner, "logical decoding");
/* invalidate non-timetravel entries */
InvalidateSystemCaches();
@ -320,6 +318,11 @@ pg_logical_slot_get_changes_guts(FunctionCallInfo fcinfo, bool confirm, bool bin
tuplestore_donestoring(tupstore);
/*
* Logical decoding could have clobbered CurrentResourceOwner during
* transaction management, so restore the executor's value. (This is
* a kluge, but it's not worth cleaning up right now.)
*/
CurrentResourceOwner = old_resowner;
/*

View File

@ -1595,6 +1595,11 @@ ApplyWorkerMain(Datum main_arg)
pqsignal(SIGTERM, die);
BackgroundWorkerUnblockSignals();
/*
* We don't currently need any ResourceOwner in a walreceiver process, but
* if we did, we could call CreateAuxProcessResourceOwner here.
*/
/* Initialise stats to a sanish value */
MyLogicalRepWorker->last_send_time = MyLogicalRepWorker->last_recv_time =
MyLogicalRepWorker->reply_time = GetCurrentTimestamp();
@ -1602,10 +1607,6 @@ ApplyWorkerMain(Datum main_arg)
/* Load the libpq-specific functions */
load_file("libpqwalreceiver", false);
Assert(CurrentResourceOwner == NULL);
CurrentResourceOwner = ResourceOwnerCreate(NULL,
"logical replication apply");
/* Run as replica session replication role. */
SetConfigOption("session_replication_role", "replica",
PGC_SUSET, PGC_S_OVERRIDE);

View File

@ -365,9 +365,6 @@ pg_logical_replication_slot_advance(XLogRecPtr moveto)
logical_read_local_xlog_page,
NULL, NULL, NULL);
CurrentResourceOwner = ResourceOwnerCreate(CurrentResourceOwner,
"logical decoding");
/* invalidate non-timetravel entries */
InvalidateSystemCaches();
@ -402,6 +399,11 @@ pg_logical_replication_slot_advance(XLogRecPtr moveto)
CHECK_FOR_INTERRUPTS();
}
/*
* Logical decoding could have clobbered CurrentResourceOwner during
* transaction management, so restore the executor's value. (This is
* a kluge, but it's not worth cleaning up right now.)
*/
CurrentResourceOwner = old_resowner;
if (ctx->reader->EndRecPtr != InvalidXLogRecPtr)

View File

@ -292,12 +292,6 @@ WalReceiverMain(void)
if (WalReceiverFunctions == NULL)
elog(ERROR, "libpqwalreceiver didn't initialize correctly");
/*
* Create a resource owner to keep track of our resources (not clear that
* we need this, but may as well have one).
*/
CurrentResourceOwner = ResourceOwnerCreate(NULL, "Wal Receiver");
/* Unblock signals (they were blocked when the postmaster forked us) */
PG_SETMASK(&UnBlockSig);

View File

@ -90,7 +90,6 @@
#include "utils/pg_lsn.h"
#include "utils/portal.h"
#include "utils/ps_status.h"
#include "utils/resowner.h"
#include "utils/timeout.h"
#include "utils/timestamp.h"
@ -264,8 +263,10 @@ InitWalSender(void)
/* Create a per-walsender data structure in shared memory */
InitWalSenderSlot();
/* Set up resource owner */
CurrentResourceOwner = ResourceOwnerCreate(NULL, "walsender top-level resource owner");
/*
* We don't currently need any ResourceOwner in a walsender process, but
* if we did, we could call CreateAuxProcessResourceOwner here.
*/
/*
* Let postmaster know that we're a WAL sender. Once we've declared us as

View File

@ -632,8 +632,19 @@ InitPostgres(const char *in_dbname, Oid dboid, const char *username,
* We are either a bootstrap process or a standalone backend. Either
* way, start up the XLOG machinery, and register to have it closed
* down at exit.
*
* We don't yet have an aux-process resource owner, but StartupXLOG
* and ShutdownXLOG will need one. Hence, create said resource owner
* (and register a callback to clean it up after ShutdownXLOG runs).
*/
CreateAuxProcessResourceOwner();
StartupXLOG();
/* Release (and warn about) any buffer pins leaked in StartupXLOG */
ReleaseAuxProcessResources(true);
/* Reset CurrentResourceOwner to nothing for the moment */
CurrentResourceOwner = NULL;
on_shmem_exit(ShutdownXLOG, 0);
}

View File

@ -22,6 +22,7 @@
#include "access/hash.h"
#include "jit/jit.h"
#include "storage/ipc.h"
#include "storage/predicate.h"
#include "storage/proc.h"
#include "utils/memutils.h"
@ -140,6 +141,7 @@ typedef struct ResourceOwnerData
ResourceOwner CurrentResourceOwner = NULL;
ResourceOwner CurTransactionResourceOwner = NULL;
ResourceOwner TopTransactionResourceOwner = NULL;
ResourceOwner AuxProcessResourceOwner = NULL;
/*
* List of add-on callbacks for resource releasing
@ -165,6 +167,7 @@ static void ResourceOwnerReleaseInternal(ResourceOwner owner,
ResourceReleasePhase phase,
bool isCommit,
bool isTopLevel);
static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
static void PrintRelCacheLeakWarning(Relation rel);
static void PrintPlanCacheLeakWarning(CachedPlan *plan);
static void PrintTupleDescLeakWarning(TupleDesc tupdesc);
@ -823,6 +826,60 @@ UnregisterResourceReleaseCallback(ResourceReleaseCallback callback, void *arg)
}
}
/*
* Establish an AuxProcessResourceOwner for the current process.
*/
void
CreateAuxProcessResourceOwner(void)
{
Assert(AuxProcessResourceOwner == NULL);
Assert(CurrentResourceOwner == NULL);
AuxProcessResourceOwner = ResourceOwnerCreate(NULL, "AuxiliaryProcess");
CurrentResourceOwner = AuxProcessResourceOwner;
/*
* Register a shmem-exit callback for cleanup of aux-process resource
* owner. (This needs to run after, e.g., ShutdownXLOG.)
*/
on_shmem_exit(ReleaseAuxProcessResourcesCallback, 0);
}
/*
* Convenience routine to release all resources tracked in
* AuxProcessResourceOwner (but that resowner is not destroyed here).
* Warn about leaked resources if isCommit is true.
*/
void
ReleaseAuxProcessResources(bool isCommit)
{
/*
* At this writing, the only thing that could actually get released is
* buffer pins; but we may as well do the full release protocol.
*/
ResourceOwnerRelease(AuxProcessResourceOwner,
RESOURCE_RELEASE_BEFORE_LOCKS,
isCommit, true);
ResourceOwnerRelease(AuxProcessResourceOwner,
RESOURCE_RELEASE_LOCKS,
isCommit, true);
ResourceOwnerRelease(AuxProcessResourceOwner,
RESOURCE_RELEASE_AFTER_LOCKS,
isCommit, true);
}
/*
* Shmem-exit callback for the same.
* Warn about leaked resources if process exit code is zero (ie normal).
*/
static void
ReleaseAuxProcessResourcesCallback(int code, Datum arg)
{
bool isCommit = (code == 0);
ReleaseAuxProcessResources(isCommit);
}
/*
* Make sure there is room for at least one more entry in a ResourceOwner's
@ -830,15 +887,12 @@ UnregisterResourceReleaseCallback(ResourceReleaseCallback callback, void *arg)
*
* This is separate from actually inserting an entry because if we run out
* of memory, it's critical to do so *before* acquiring the resource.
*
* We allow the case owner == NULL because the bufmgr is sometimes invoked
* outside any transaction (for example, during WAL recovery).
*/
void
ResourceOwnerEnlargeBuffers(ResourceOwner owner)
{
if (owner == NULL)
return;
/* We used to allow pinning buffers without a resowner, but no more */
Assert(owner != NULL);
ResourceArrayEnlarge(&(owner->bufferarr));
}
@ -846,29 +900,19 @@ ResourceOwnerEnlargeBuffers(ResourceOwner owner)
* Remember that a buffer pin is owned by a ResourceOwner
*
* Caller must have previously done ResourceOwnerEnlargeBuffers()
*
* We allow the case owner == NULL because the bufmgr is sometimes invoked
* outside any transaction (for example, during WAL recovery).
*/
void
ResourceOwnerRememberBuffer(ResourceOwner owner, Buffer buffer)
{
if (owner == NULL)
return;
ResourceArrayAdd(&(owner->bufferarr), BufferGetDatum(buffer));
}
/*
* Forget that a buffer pin is owned by a ResourceOwner
*
* We allow the case owner == NULL because the bufmgr is sometimes invoked
* outside any transaction (for example, during WAL recovery).
*/
void
ResourceOwnerForgetBuffer(ResourceOwner owner, Buffer buffer)
{
if (owner == NULL)
return;
if (!ResourceArrayRemove(&(owner->bufferarr), BufferGetDatum(buffer)))
elog(ERROR, "buffer %d is not owned by resource owner %s",
buffer, owner->name);

View File

@ -33,6 +33,7 @@ typedef struct ResourceOwnerData *ResourceOwner;
extern PGDLLIMPORT ResourceOwner CurrentResourceOwner;
extern PGDLLIMPORT ResourceOwner CurTransactionResourceOwner;
extern PGDLLIMPORT ResourceOwner TopTransactionResourceOwner;
extern PGDLLIMPORT ResourceOwner AuxProcessResourceOwner;
/*
* Resource releasing is done in three phases: pre-locks, locks, and
@ -78,5 +79,7 @@ extern void RegisterResourceReleaseCallback(ResourceReleaseCallback callback,
void *arg);
extern void UnregisterResourceReleaseCallback(ResourceReleaseCallback callback,
void *arg);
extern void CreateAuxProcessResourceOwner(void);
extern void ReleaseAuxProcessResources(bool isCommit);
#endif /* RESOWNER_H */

View File

@ -24,7 +24,6 @@
#include "storage/procarray.h"
#include "storage/shm_mq.h"
#include "storage/shm_toc.h"
#include "utils/resowner.h"
#include "test_shm_mq.h"
@ -69,13 +68,16 @@ test_shm_mq_main(Datum main_arg)
* Connect to the dynamic shared memory segment.
*
* The backend that registered this worker passed us the ID of a shared
* memory segment to which we must attach for further instructions. In
* order to attach to dynamic shared memory, we need a resource owner.
* Once we've mapped the segment in our address space, attach to the table
* of contents so we can locate the various data structures we'll need to
* memory segment to which we must attach for further instructions. Once
* we've mapped the segment in our address space, attach to the table of
* contents so we can locate the various data structures we'll need to
* find within the segment.
*
* Note: at this point, we have not created any ResourceOwner in this
* process. This will result in our DSM mapping surviving until process
* exit, which is fine. If there were a ResourceOwner, it would acquire
* ownership of the mapping, but we have no need for that.
*/
CurrentResourceOwner = ResourceOwnerCreate(NULL, "test_shm_mq worker");
seg = dsm_attach(DatumGetInt32(main_arg));
if (seg == NULL)
ereport(ERROR,
@ -133,10 +135,8 @@ test_shm_mq_main(Datum main_arg)
copy_messages(inqh, outqh);
/*
* We're done. Explicitly detach the shared memory segment so that we
* don't get a resource leak warning at commit time. This will fire any
* on_dsm_detach callbacks we've registered, as well. Once that's done,
* we can go ahead and exit.
* We're done. For cleanliness, explicitly detach from the shared memory
* segment (that would happen anyway during process exit, though).
*/
dsm_detach(seg);
proc_exit(1);