database/postgres
1
0
Fork 0
mirror of https://github.com/postgres/postgres.git synced 2025-06-16 06:01:02 +03:00
Code Activity

Use an MVCC snapshot, rather than SnapshotNow, for catalog scans.

Browse Source 
SnapshotNow scans have the undesirable property that, in the face of
concurrent updates, the scan can fail to see either the old or the new
versions of the row.  In many cases, we work around this by requiring
DDL operations to hold AccessExclusiveLock on the object being
modified; in some cases, the existing locking is inadequate and random
failures occur as a result.  This commit doesn't change anything
related to locking, but will hopefully pave the way to allowing lock
strength reductions in the future.

The major issue has held us back from making this change in the past
is that taking an MVCC snapshot is significantly more expensive than
using a static special snapshot such as SnapshotNow.  However, testing
of various worst-case scenarios reveals that this problem is not
severe except under fairly extreme workloads.  To mitigate those
problems, we avoid retaking the MVCC snapshot for each new scan;
instead, we take a new snapshot only when invalidation messages have
been processed.  The catcache machinery already requires that
invalidation messages be sent before releasing the related heavyweight
lock; else other backends might rely on locally-cached data rather
than scanning the catalog at all.  Thus, making snapshot reuse
dependent on the same guarantees shouldn't break anything that wasn't
already subtly broken.

Patch by me.  Review by Michael Paquier and Andres Freund.
This commit is contained in:
Robert Haas
2013-07-02 09:47:01 -04:00
parent 384f933046
commit 568d4138c6
69 changed files with 617 additions and 353 deletions
Download Patch File Download Diff File Expand all files Collapse all files

59
src/backend/commands/tablecmds.c
View File

@ -2741,7 +2741,7 @@ AlterTableGetLockLevel(List *cmds)
* multiple DDL operations occur in a stream against frequently accessed
* tables.
*
* 1. Catalog tables are read using SnapshotNow, which has a race bug that
* 1. Catalog tables were read using SnapshotNow, which has a race bug that
* allows a scan to return no valid rows even when one is present in the
* case of a commit of a concurrent update of the catalog table.
* SnapshotNow also ignores transactions in progress, so takes the latest
@ -3753,6 +3753,7 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap, LOCKMODE lockmode)
MemoryContext oldCxt;
List *dropped_attrs = NIL;
ListCell *lc;
Snapshot snapshot;
if (newrel)
ereport(DEBUG1,
@ -3805,7 +3806,8 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap, LOCKMODE lockmode)
* Scan through the rows, generating a new row if needed and then
* checking all the constraints.
*/
scan = heap_beginscan(oldrel, SnapshotNow, 0, NULL);
snapshot = RegisterSnapshot(GetLatestSnapshot());
scan = heap_beginscan(oldrel, snapshot, 0, NULL);
/*
* Switch to per-tuple memory context and reset it for each tuple
@ -3906,6 +3908,7 @@ ATRewriteTable(AlteredTableInfo *tab, Oid OIDNewHeap, LOCKMODE lockmode)
MemoryContextSwitchTo(oldCxt);
heap_endscan(scan);
UnregisterSnapshot(snapshot);
ExecDropSingleTupleTableSlot(oldslot);
ExecDropSingleTupleTableSlot(newslot);
@ -4182,7 +4185,7 @@ find_composite_type_dependencies(Oid typeOid, Relation origRelation,
ObjectIdGetDatum(typeOid));
depScan = systable_beginscan(depRel, DependReferenceIndexId, true,
SnapshotNow, 2, key);
NULL, 2, key);
while (HeapTupleIsValid(depTup = systable_getnext(depScan)))
{
@ -4281,7 +4284,7 @@ find_typed_table_dependencies(Oid typeOid, const char *typeName, DropBehavior be
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(typeOid));
scan = heap_beginscan(classRel, SnapshotNow, 1, key);
scan = heap_beginscan_catalog(classRel, 1, key);
while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
{
@ -6220,7 +6223,7 @@ ATExecAlterConstraint(Relation rel, AlterTableCmd *cmd,
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(RelationGetRelid(rel)));
scan = systable_beginscan(conrel, ConstraintRelidIndexId,
true, SnapshotNow, 1, &key);
true, NULL, 1, &key);
while (HeapTupleIsValid(contuple = systable_getnext(scan)))
{
@ -6282,7 +6285,7 @@ ATExecAlterConstraint(Relation rel, AlterTableCmd *cmd,
ObjectIdGetDatum(HeapTupleGetOid(contuple)));
tgscan = systable_beginscan(tgrel, TriggerConstraintIndexId, true,
SnapshotNow, 1, &tgkey);
NULL, 1, &tgkey);
while (HeapTupleIsValid(tgtuple = systable_getnext(tgscan)))
{
@ -6343,7 +6346,7 @@ ATExecValidateConstraint(Relation rel, char *constrName, bool recurse,
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(RelationGetRelid(rel)));
scan = systable_beginscan(conrel, ConstraintRelidIndexId,
true, SnapshotNow, 1, &key);
true, NULL, 1, &key);
while (HeapTupleIsValid(tuple = systable_getnext(scan)))
{
@ -6824,6 +6827,7 @@ validateCheckConstraint(Relation rel, HeapTuple constrtup)
TupleTableSlot *slot;
Form_pg_constraint constrForm;
bool isnull;
Snapshot snapshot;
constrForm = (Form_pg_constraint) GETSTRUCT(constrtup);
@ -6849,7 +6853,8 @@ validateCheckConstraint(Relation rel, HeapTuple constrtup)
slot = MakeSingleTupleTableSlot(tupdesc);
econtext->ecxt_scantuple = slot;
scan = heap_beginscan(rel, SnapshotNow, 0, NULL);
snapshot = RegisterSnapshot(GetLatestSnapshot());
scan = heap_beginscan(rel, snapshot, 0, NULL);
/*
* Switch to per-tuple memory context and reset it for each tuple
@ -6873,6 +6878,7 @@ validateCheckConstraint(Relation rel, HeapTuple constrtup)
MemoryContextSwitchTo(oldcxt);
heap_endscan(scan);
UnregisterSnapshot(snapshot);
ExecDropSingleTupleTableSlot(slot);
FreeExecutorState(estate);
}
@ -6893,6 +6899,7 @@ validateForeignKeyConstraint(char *conname,
HeapScanDesc scan;
HeapTuple tuple;
Trigger trig;
Snapshot snapshot;
ereport(DEBUG1,
(errmsg("validating foreign key constraint \"%s\"", conname)));
@ -6924,7 +6931,8 @@ validateForeignKeyConstraint(char *conname,
* if that tuple had just been inserted. If any of those fail, it should
* ereport(ERROR) and that's that.
*/
scan = heap_beginscan(rel, SnapshotNow, 0, NULL);
snapshot = RegisterSnapshot(GetLatestSnapshot());
scan = heap_beginscan(rel, snapshot, 0, NULL);
while ((tuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
{
@ -6956,6 +6964,7 @@ validateForeignKeyConstraint(char *conname,
}
heap_endscan(scan);
UnregisterSnapshot(snapshot);
}
static void
@ -7174,7 +7183,7 @@ ATExecDropConstraint(Relation rel, const char *constrName,
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(RelationGetRelid(rel)));
scan = systable_beginscan(conrel, ConstraintRelidIndexId,
true, SnapshotNow, 1, &key);
true, NULL, 1, &key);
while (HeapTupleIsValid(tuple = systable_getnext(scan)))
{
@ -7255,7 +7264,7 @@ ATExecDropConstraint(Relation rel, const char *constrName,
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(childrelid));
scan = systable_beginscan(conrel, ConstraintRelidIndexId,
true, SnapshotNow, 1, &key);
true, NULL, 1, &key);
/* scan for matching tuple - there should only be one */
while (HeapTupleIsValid(tuple = systable_getnext(scan)))
@ -7655,7 +7664,7 @@ ATExecAlterColumnType(AlteredTableInfo *tab, Relation rel,
Int32GetDatum((int32) attnum));
scan = systable_beginscan(depRel, DependReferenceIndexId, true,
SnapshotNow, 3, key);
NULL, 3, key);
while (HeapTupleIsValid(depTup = systable_getnext(scan)))
{
@ -7840,7 +7849,7 @@ ATExecAlterColumnType(AlteredTableInfo *tab, Relation rel,
Int32GetDatum((int32) attnum));
scan = systable_beginscan(depRel, DependDependerIndexId, true,
SnapshotNow, 3, key);
NULL, 3, key);
while (HeapTupleIsValid(depTup = systable_getnext(scan)))
{
@ -8517,7 +8526,7 @@ change_owner_fix_column_acls(Oid relationOid, Oid oldOwnerId, Oid newOwnerId)
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(relationOid));
scan = systable_beginscan(attRelation, AttributeRelidNumIndexId,
true, SnapshotNow, 1, key);
true, NULL, 1, key);
while (HeapTupleIsValid(attributeTuple = systable_getnext(scan)))
{
Form_pg_attribute att = (Form_pg_attribute) GETSTRUCT(attributeTuple);
@ -8594,7 +8603,7 @@ change_owner_recurse_to_sequences(Oid relationOid, Oid newOwnerId, LOCKMODE lock
/* we leave refobjsubid unspecified */
scan = systable_beginscan(depRel, DependReferenceIndexId, true,
SnapshotNow, 2, key);
NULL, 2, key);
while (HeapTupleIsValid(tup = systable_getnext(scan)))
{
@ -9188,7 +9197,7 @@ ATExecAddInherit(Relation child_rel, RangeVar *parent, LOCKMODE lockmode)
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(RelationGetRelid(child_rel)));
scan = systable_beginscan(catalogRelation, InheritsRelidSeqnoIndexId,
true, SnapshotNow, 1, &key);
true, NULL, 1, &key);
/* inhseqno sequences start at 1 */
inhseqno = 0;
@ -9430,7 +9439,7 @@ MergeConstraintsIntoExisting(Relation child_rel, Relation parent_rel)
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(RelationGetRelid(parent_rel)));
parent_scan = systable_beginscan(catalog_relation, ConstraintRelidIndexId,
true, SnapshotNow, 1, &parent_key);
true, NULL, 1, &parent_key);
while (HeapTupleIsValid(parent_tuple = systable_getnext(parent_scan)))
{
@ -9453,7 +9462,7 @@ MergeConstraintsIntoExisting(Relation child_rel, Relation parent_rel)
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(RelationGetRelid(child_rel)));
child_scan = systable_beginscan(catalog_relation, ConstraintRelidIndexId,
true, SnapshotNow, 1, &child_key);
true, NULL, 1, &child_key);
while (HeapTupleIsValid(child_tuple = systable_getnext(child_scan)))
{
@ -9561,7 +9570,7 @@ ATExecDropInherit(Relation rel, RangeVar *parent, LOCKMODE lockmode)
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(RelationGetRelid(rel)));
scan = systable_beginscan(catalogRelation, InheritsRelidSeqnoIndexId,
true, SnapshotNow, 1, key);
true, NULL, 1, key);
while (HeapTupleIsValid(inheritsTuple = systable_getnext(scan)))
{
@ -9595,7 +9604,7 @@ ATExecDropInherit(Relation rel, RangeVar *parent, LOCKMODE lockmode)
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(RelationGetRelid(rel)));
scan = systable_beginscan(catalogRelation, AttributeRelidNumIndexId,
true, SnapshotNow, 1, key);
true, NULL, 1, key);
while (HeapTupleIsValid(attributeTuple = systable_getnext(scan)))
{
Form_pg_attribute att = (Form_pg_attribute) GETSTRUCT(attributeTuple);
@ -9637,7 +9646,7 @@ ATExecDropInherit(Relation rel, RangeVar *parent, LOCKMODE lockmode)
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(RelationGetRelid(parent_rel)));
scan = systable_beginscan(catalogRelation, ConstraintRelidIndexId,
true, SnapshotNow, 1, key);
true, NULL, 1, key);
connames = NIL;
@ -9657,7 +9666,7 @@ ATExecDropInherit(Relation rel, RangeVar *parent, LOCKMODE lockmode)
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(RelationGetRelid(rel)));
scan = systable_beginscan(catalogRelation, ConstraintRelidIndexId,
true, SnapshotNow, 1, key);
true, NULL, 1, key);
while (HeapTupleIsValid(constraintTuple = systable_getnext(scan)))
{
@ -9749,7 +9758,7 @@ drop_parent_dependency(Oid relid, Oid refclassid, Oid refobjid)
Int32GetDatum(0));
scan = systable_beginscan(catalogRelation, DependDependerIndexId, true,
SnapshotNow, 3, key);
NULL, 3, key);
while (HeapTupleIsValid(depTuple = systable_getnext(scan)))
{
@ -9804,7 +9813,7 @@ ATExecAddOf(Relation rel, const TypeName *ofTypename, LOCKMODE lockmode)
BTEqualStrategyNumber, F_OIDEQ,
ObjectIdGetDatum(relid));
scan = systable_beginscan(inheritsRelation, InheritsRelidSeqnoIndexId,
true, SnapshotNow, 1, &key);
true, NULL, 1, &key);
if (HeapTupleIsValid(systable_getnext(scan)))
ereport(ERROR,
(errcode(ERRCODE_WRONG_OBJECT_TYPE),
@ -10260,7 +10269,7 @@ AlterSeqNamespaces(Relation classRel, Relation rel,
/* we leave refobjsubid unspecified */
scan = systable_beginscan(depRel, DependReferenceIndexId, true,
SnapshotNow, 2, key);
NULL, 2, key);
while (HeapTupleIsValid(tup = systable_getnext(scan)))
{