rdelaage
/
postgresql
mirror of https://git.postgresql.org/git/postgresql.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

autovacuum: Drop orphan temp tables more quickly but with more caution. 

Previously, we only dropped an orphan temp table when it became old
enough to threaten wraparound; instead, doing it immediately.  The
only value of waiting is that someone might be able to examine the
contents of the orphan temp table for forensic purposes, but it's
pretty difficult to actually do that and few users will wish to do so.
On the flip side, not performing the drop immediately generates log
spam and bloats pg_class.

In addition, per a report from Grigory Smolkin, if a temporary schema
contains a very large number of temporary tables, a backend attempting
to clear the temporary schema might fail due to lock table exhaustion.
It's helpful for autovacuum to clean up after such cases, and we don't
want it to wait for wraparound to threaten before doing so.  To
prevent autovacuum from failing in the same manner as a backend trying
to drop an entire temp schema, remove orphan temp tables in batches of
50, committing after each batch, so that we don't accumulate an
unbounded number of locks.  If a drop fails, retry other orphan tables
that need to be dropped up to 10 times before giving up.  With this
system, if a backend does fail to clean a temporary schema due to
lock table exhaustion, autovacuum should hopefully put things right
the next time it processes the database.

Discussion: CAB7nPqSbYT6dRwsXVgiKmBdL_ARemfDZMPA+RPeC_ge0GK70hA@mail.gmail.com

Michael Paquier, with a bunch of comment changes by me.
This commit is contained in:
Robert Haas 2016-11-21 12:54:19 -05:00
parent f24cf960d7
commit a734fd5d1c
1 changed files with 127 additions and 26 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

153
src/backend/postmaster/autovacuum.c
View File

@ -130,6 +130,17 @@ int Log_autovacuum_min_duration = -1;
#define MIN_AUTOVAC_SLEEPTIME 100.0 /* milliseconds */
#define MAX_AUTOVAC_SLEEPTIME 300 /* seconds */
/*
* Maximum number of orphan temporary tables to drop in a single transaction.
* (If this is too high, we might run out of heavyweight locks.)
*/
#define MAX_ORPHAN_ITEMS 50
/*
* After this many failures, stop trying to drop orphan temporary tables.
*/
#define MAX_ORPHAN_DROP_FAILURE 10
/* Flags to tell if we are in an autovacuum process */
static bool am_autovacuum_launcher = false;
static bool am_autovacuum_worker = false;
@ -1887,6 +1898,8 @@ do_autovacuum(void)
HeapScanDesc relScan;
Form_pg_database dbForm;
List *table_oids = NIL;
List *orphan_oids = NIL;
List *pending_oids = NIL;
HASHCTL ctl;
HTAB *table_toast_map;
ListCell *volatile cell;
@ -1895,6 +1908,7 @@ do_autovacuum(void)
BufferAccessStrategy bstrategy;
ScanKeyData key;
TupleDesc pg_class_desc;
int orphan_failures;
int effective_multixact_freeze_max_age;
/*
@ -2038,33 +2052,11 @@ do_autovacuum(void)
if (backendID == MyBackendId || BackendIdGetProc(backendID) == NULL)
{
/*
* We found an orphan temp table (which was probably left
* behind by a crashed backend). If it's so old as to need
* vacuum for wraparound, forcibly drop it. Otherwise just
* log a complaint.
* We found an orphan temp table which was probably left
* behind by a crashed backend. Remember it, so we can attempt
* to drop it.
*/
if (wraparound)
{
ObjectAddress object;
ereport(LOG,
(errmsg("autovacuum: dropping orphan temp table \"%s\".\"%s\" in database \"%s\"",
get_namespace_name(classForm->relnamespace),
NameStr(classForm->relname),
get_database_name(MyDatabaseId))));
object.classId = RelationRelationId;
object.objectId = relid;
object.objectSubId = 0;
performDeletion(&object, DROP_CASCADE, PERFORM_DELETION_INTERNAL);
}
else
{
ereport(LOG,
(errmsg("autovacuum: found orphan temp table \"%s\".\"%s\" in database \"%s\"",
get_namespace_name(classForm->relnamespace),
NameStr(classForm->relname),
get_database_name(MyDatabaseId))));
}
orphan_oids = lappend_oid(orphan_oids, relid);
}
}
else
@ -2161,6 +2153,115 @@ do_autovacuum(void)
heap_endscan(relScan);
heap_close(classRel, AccessShareLock);
/*
* Loop through orphan temporary tables and drop them in batches. If
* we're unable to drop one particular table, we'll retry to see if we
* can drop others, but if we fail too many times we'll give up and proceed
* with our regular work, so that this step hopefully can't wedge
* autovacuum for too long.
*/
while (list_length(orphan_oids) > 0 &&
orphan_failures < MAX_ORPHAN_DROP_FAILURE)
{
Oid relid = linitial_oid(orphan_oids);
ObjectAddress object;
char *namespace = get_namespace_name(get_rel_namespace(relid));
char *relname = get_rel_name(relid);
orphan_oids = list_delete_first(orphan_oids);
PG_TRY();
{
ereport(LOG,
(errmsg("autovacuum: dropping orphan temp table \"%s\".\"%s\" in database \"%s\"",
namespace, relname,
get_database_name(MyDatabaseId))));
object.classId = RelationRelationId;
object.objectId = relid;
object.objectSubId = 0;
performDeletion(&object, DROP_CASCADE, PERFORM_DELETION_INTERNAL);
/*
* This orphan table has been dropped correctly, add it to the
* list of tables whose drop should be attempted again if an
* error after in the same transaction.
*/
pending_oids = lappend_oid(pending_oids, relid);
}
PG_CATCH();
{
/* Abort the current transaction. */
HOLD_INTERRUPTS();
errcontext("dropping of orphan temp table \"%s\".\"%s\" in database \"%s\"",
namespace, relname,
get_database_name(MyDatabaseId));
EmitErrorReport();
/* this resets the PGXACT flags too */
AbortOutOfAnyTransaction();
FlushErrorState();
/*
* Any tables were succesfully dropped before the failure now
* need to be dropped again. Add them back into the list, but
* don't retry the table that failed.
*/
orphan_oids = list_concat(orphan_oids, pending_oids);
orphan_failures++;
/* Start a new transaction. */
StartTransactionCommand();
/* StartTransactionCommand changed elsewhere the memory context */
MemoryContextSwitchTo(AutovacMemCxt);
RESUME_INTERRUPTS();
}
PG_END_TRY();
/*
* If we've successfully dropped quite a few tables, commit the
* transaction and begin a new one. The main point of this is to
* avoid accumulating too many locks and blowing out the lock table,
* but it also minimizes the amount of work that will have to be rolled
* back if we fail to drop some table later in the list.
*/
if (list_length(pending_oids) >= MAX_ORPHAN_ITEMS)
{
CommitTransactionCommand();
StartTransactionCommand();
/* StartTransactionCommand changed elsewhere */
MemoryContextSwitchTo(AutovacMemCxt);
list_free(pending_oids);
pending_oids = NIL;
}
pfree(relname);
pfree(namespace);
}
/*
* Commit current transaction to finish the cleanup done previously and
* restart a new one to not bloat the activity of the following steps.
* This needs to happen only if there are any items thought as previously
* pending, but are actually not as the last transaction doing the cleanup
* has been successful.
*/
if (list_length(pending_oids) > 0)
{
CommitTransactionCommand();
StartTransactionCommand();
/* StartTransactionCommand changed elsewhere */
MemoryContextSwitchTo(AutovacMemCxt);
list_free(pending_oids);
}
/*
* Create a buffer access strategy object for VACUUM to use. We want to
* use the same one across all the vacuum operations we perform, since the