1
0
mirror of https://github.com/postgres/postgres.git synced 2025-07-02 09:02:37 +03:00

Allow HOT updates for some expression indexes

If the value of an index expression is unchanged after UPDATE,
allow HOT updates where previously we disallowed them, giving
a significant performance boost in those cases.

Particularly useful for indexes such as JSON->>field where the
JSON value changes but the indexed value does not.

Submitted as "surjective indexes" patch, now enabled by use
of new "recheck_on_update" parameter.

Author: Konstantin Knizhnik
Reviewer: Simon Riggs, with much wordsmithing and some cleanup
This commit is contained in:
Simon Riggs
2018-03-27 19:57:02 +01:00
parent 1944cdc982
commit c203d6cf81
13 changed files with 395 additions and 22 deletions

View File

@ -129,6 +129,15 @@ static relopt_bool boolRelOpts[] =
},
true
},
{
{
"recheck_on_update",
"Recheck functional index expression for changed value after update",
RELOPT_KIND_INDEX,
ShareUpdateExclusiveLock /* since only applies to later UPDATEs */
},
true
},
{
{
"security_barrier",
@ -1310,7 +1319,7 @@ fillRelOptions(void *rdopts, Size basesize,
break;
}
}
if (validate && !found)
if (validate && !found && options[i].gen->kinds != RELOPT_KIND_INDEX)
elog(ERROR, "reloption \"%s\" not found in parse table",
options[i].gen->name);
}
@ -1466,6 +1475,40 @@ index_reloptions(amoptions_function amoptions, Datum reloptions, bool validate)
return amoptions(reloptions, validate);
}
/*
* Parse generic options for all indexes.
*
* reloptions options as text[] datum
* validate error flag
*/
bytea *
index_generic_reloptions(Datum reloptions, bool validate)
{
int numoptions;
GenericIndexOpts *idxopts;
relopt_value *options;
static const relopt_parse_elt tab[] = {
{"recheck_on_update", RELOPT_TYPE_BOOL, offsetof(GenericIndexOpts, recheck_on_update)}
};
options = parseRelOptions(reloptions, validate,
RELOPT_KIND_INDEX,
&numoptions);
/* if none set, we're done */
if (numoptions == 0)
return NULL;
idxopts = allocateReloptStruct(sizeof(GenericIndexOpts), options, numoptions);
fillRelOptions((void *)idxopts, sizeof(GenericIndexOpts), options, numoptions,
validate, tab, lengthof(tab));
pfree(options);
return (bytea*) idxopts;
}
/*
* Option parser for attribute reloptions
*/

View File

@ -56,6 +56,7 @@
#include "access/xlogutils.h"
#include "catalog/catalog.h"
#include "catalog/namespace.h"
#include "catalog/index.h"
#include "miscadmin.h"
#include "pgstat.h"
#include "port/atomics.h"
@ -74,7 +75,9 @@
#include "utils/snapmgr.h"
#include "utils/syscache.h"
#include "utils/tqual.h"
#include "utils/memutils.h"
#include "nodes/execnodes.h"
#include "executor/executor.h"
/* GUC variable */
bool synchronize_seqscans = true;
@ -126,6 +129,7 @@ static bool ConditionalMultiXactIdWait(MultiXactId multi, MultiXactStatus status
static XLogRecPtr log_heap_new_cid(Relation relation, HeapTuple tup);
static HeapTuple ExtractReplicaIdentity(Relation rel, HeapTuple tup, bool key_modified,
bool *copy);
static bool ProjIndexIsUnchanged(Relation relation, HeapTuple oldtup, HeapTuple newtup);
/*
@ -3508,6 +3512,7 @@ heap_update(Relation relation, ItemPointer otid, HeapTuple newtup,
HTSU_Result result;
TransactionId xid = GetCurrentTransactionId();
Bitmapset *hot_attrs;
Bitmapset *proj_idx_attrs;
Bitmapset *key_attrs;
Bitmapset *id_attrs;
Bitmapset *interesting_attrs;
@ -3571,12 +3576,11 @@ heap_update(Relation relation, ItemPointer otid, HeapTuple newtup,
* Note that we get copies of each bitmap, so we need not worry about
* relcache flush happening midway through.
*/
hot_attrs = RelationGetIndexAttrBitmap(relation, INDEX_ATTR_BITMAP_ALL);
hot_attrs = RelationGetIndexAttrBitmap(relation, INDEX_ATTR_BITMAP_HOT);
proj_idx_attrs = RelationGetIndexAttrBitmap(relation, INDEX_ATTR_BITMAP_PROJ);
key_attrs = RelationGetIndexAttrBitmap(relation, INDEX_ATTR_BITMAP_KEY);
id_attrs = RelationGetIndexAttrBitmap(relation,
INDEX_ATTR_BITMAP_IDENTITY_KEY);
block = ItemPointerGetBlockNumber(otid);
buffer = ReadBuffer(relation, block);
page = BufferGetPage(buffer);
@ -3596,6 +3600,7 @@ heap_update(Relation relation, ItemPointer otid, HeapTuple newtup,
if (!PageIsFull(page))
{
interesting_attrs = bms_add_members(interesting_attrs, hot_attrs);
interesting_attrs = bms_add_members(interesting_attrs, proj_idx_attrs);
hot_attrs_checked = true;
}
interesting_attrs = bms_add_members(interesting_attrs, key_attrs);
@ -3894,6 +3899,7 @@ l2:
if (vmbuffer != InvalidBuffer)
ReleaseBuffer(vmbuffer);
bms_free(hot_attrs);
bms_free(proj_idx_attrs);
bms_free(key_attrs);
bms_free(id_attrs);
bms_free(modified_attrs);
@ -4201,11 +4207,18 @@ l2:
/*
* Since the new tuple is going into the same page, we might be able
* to do a HOT update. Check if any of the index columns have been
* changed. If the page was already full, we may have skipped checking
* for index columns. If so, HOT update is possible.
* changed, or if we have projection functional indexes, check whether
* the old and the new values are the same. If the page was already
* full, we may have skipped checking for index columns. If so, HOT
* update is possible.
*/
if (hot_attrs_checked && !bms_overlap(modified_attrs, hot_attrs))
if (hot_attrs_checked
&& !bms_overlap(modified_attrs, hot_attrs)
&& (!bms_overlap(modified_attrs, proj_idx_attrs)
|| ProjIndexIsUnchanged(relation, &oldtup, newtup)))
{
use_hot_update = true;
}
}
else
{
@ -4367,6 +4380,7 @@ l2:
heap_freetuple(old_key_tuple);
bms_free(hot_attrs);
bms_free(proj_idx_attrs);
bms_free(key_attrs);
bms_free(id_attrs);
bms_free(modified_attrs);
@ -4453,6 +4467,83 @@ heap_tuple_attr_equals(TupleDesc tupdesc, int attrnum,
}
}
/*
* Check whether the value is unchanged after update of a projection
* functional index. Compare the new and old values of the indexed
* expression to see if we are able to use a HOT update or not.
*/
static bool ProjIndexIsUnchanged(Relation relation, HeapTuple oldtup, HeapTuple newtup)
{
ListCell *l;
List *indexoidlist = RelationGetIndexList(relation);
EState *estate = CreateExecutorState();
ExprContext *econtext = GetPerTupleExprContext(estate);
TupleTableSlot *slot = MakeSingleTupleTableSlot(RelationGetDescr(relation));
bool equals = true;
Datum old_values[INDEX_MAX_KEYS];
bool old_isnull[INDEX_MAX_KEYS];
Datum new_values[INDEX_MAX_KEYS];
bool new_isnull[INDEX_MAX_KEYS];
int indexno = 0;
econtext->ecxt_scantuple = slot;
foreach(l, indexoidlist)
{
if (bms_is_member(indexno, relation->rd_projidx))
{
Oid indexOid = lfirst_oid(l);
Relation indexDesc = index_open(indexOid, AccessShareLock);
IndexInfo *indexInfo = BuildIndexInfo(indexDesc);
int i;
ResetExprContext(econtext);
ExecStoreTuple(oldtup, slot, InvalidBuffer, false);
FormIndexDatum(indexInfo,
slot,
estate,
old_values,
old_isnull);
ExecStoreTuple(newtup, slot, InvalidBuffer, false);
FormIndexDatum(indexInfo,
slot,
estate,
new_values,
new_isnull);
for (i = 0; i < indexInfo->ii_NumIndexAttrs; i++)
{
if (old_isnull[i] != new_isnull[i])
{
equals = false;
break;
}
else if (!old_isnull[i])
{
Form_pg_attribute att = TupleDescAttr(RelationGetDescr(indexDesc), i);
if (!datumIsEqual(old_values[i], new_values[i], att->attbyval, att->attlen))
{
equals = false;
break;
}
}
}
index_close(indexDesc, AccessShareLock);
if (!equals)
{
break;
}
}
indexno += 1;
}
ExecDropSingleTupleTableSlot(slot);
FreeExecutorState(estate);
return equals;
}
/*
* Check which columns are being updated.
*

View File

@ -26,6 +26,7 @@
#include "access/amapi.h"
#include "access/multixact.h"
#include "access/relscan.h"
#include "access/reloptions.h"
#include "access/sysattr.h"
#include "access/transam.h"
#include "access/visibilitymap.h"
@ -3863,7 +3864,7 @@ reindex_relation(Oid relid, int flags, int options)
/* Ensure rd_indexattr is valid; see comments for RelationSetIndexList */
if (is_pg_class)
(void) RelationGetIndexAttrBitmap(rel, INDEX_ATTR_BITMAP_ALL);
(void) RelationGetIndexAttrBitmap(rel, INDEX_ATTR_BITMAP_HOT);
PG_TRY();
{

View File

@ -69,8 +69,10 @@
#include "miscadmin.h"
#include "nodes/nodeFuncs.h"
#include "optimizer/clauses.h"
#include "optimizer/cost.h"
#include "optimizer/prep.h"
#include "optimizer/var.h"
#include "pgstat.h"
#include "rewrite/rewriteDefine.h"
#include "rewrite/rowsecurity.h"
#include "storage/lmgr.h"
@ -2314,9 +2316,11 @@ RelationDestroyRelation(Relation relation, bool remember_tupdesc)
list_free_deep(relation->rd_fkeylist);
list_free(relation->rd_indexlist);
bms_free(relation->rd_indexattr);
bms_free(relation->rd_projindexattr);
bms_free(relation->rd_keyattr);
bms_free(relation->rd_pkattr);
bms_free(relation->rd_idattr);
bms_free(relation->rd_projidx);
if (relation->rd_pubactions)
pfree(relation->rd_pubactions);
if (relation->rd_options)
@ -4799,6 +4803,73 @@ RelationGetIndexPredicate(Relation relation)
return result;
}
#define HEURISTIC_MAX_HOT_RECHECK_EXPR_COST 1000
/*
* Check if functional index is projection: index expression returns some subset
* of its argument values. During HOT update check we handle projection indexes
* differently: instead of checking if any of attributes used in indexed
* expression were updated, we calculate and compare values of index expression
* for old and new tuple values.
*
* Decision made by this function is based on two sources:
* 1. Calculated cost of index expression: if greater than some heuristic limit
then extra comparison of index expression values is expected to be too
expensive, so we don't attempt it by default.
* 2. "recheck_on_update" index option explicitly set by user, which overrides 1)
*/
static bool IsProjectionFunctionalIndex(Relation index, IndexInfo* ii)
{
bool is_projection = false;
if (ii->ii_Expressions)
{
HeapTuple tuple;
Datum reloptions;
bool isnull;
QualCost index_expr_cost;
/* by default functional index is considered as non-injective */
is_projection = true;
cost_qual_eval(&index_expr_cost, ii->ii_Expressions, NULL);
/*
* If index expression is too expensive, then disable projection
* optimization, because extra evaluation of index expression is
* expected to be more expensive than index update. Currently the
* projection optimization has to calculate index expression twice
* when the value of index expression has not changed and three times
* when values differ because the expression is recalculated when
* inserting a new index entry for the changed value.
*/
if ((index_expr_cost.startup + index_expr_cost.per_tuple) >
HEURISTIC_MAX_HOT_RECHECK_EXPR_COST)
is_projection = false;
tuple = SearchSysCache1(RELOID, ObjectIdGetDatum(RelationGetRelid(index)));
if (!HeapTupleIsValid(tuple))
elog(ERROR, "cache lookup failed for relation %u", RelationGetRelid(index));
reloptions = SysCacheGetAttr(RELOID, tuple,
Anum_pg_class_reloptions, &isnull);
if (!isnull)
{
GenericIndexOpts *idxopts;
idxopts = (GenericIndexOpts *) index_generic_reloptions(reloptions, false);
if (idxopts != NULL)
{
is_projection = idxopts->recheck_on_update;
pfree(idxopts);
}
}
ReleaseSysCache(tuple);
}
return is_projection;
}
/*
* RelationGetIndexAttrBitmap -- get a bitmap of index attribute numbers
*
@ -4826,24 +4897,29 @@ RelationGetIndexPredicate(Relation relation)
Bitmapset *
RelationGetIndexAttrBitmap(Relation relation, IndexAttrBitmapKind attrKind)
{
Bitmapset *indexattrs; /* indexed columns */
Bitmapset *indexattrs; /* columns used in non-projection indexes */
Bitmapset *projindexattrs; /* columns used in projection indexes */
Bitmapset *uindexattrs; /* columns in unique indexes */
Bitmapset *pkindexattrs; /* columns in the primary index */
Bitmapset *idindexattrs; /* columns in the replica identity */
Bitmapset *projindexes; /* projection indexes */
List *indexoidlist;
List *newindexoidlist;
Oid relpkindex;
Oid relreplindex;
ListCell *l;
MemoryContext oldcxt;
int indexno;
/* Quick exit if we already computed the result. */
if (relation->rd_indexattr != NULL)
{
switch (attrKind)
{
case INDEX_ATTR_BITMAP_ALL:
case INDEX_ATTR_BITMAP_HOT:
return bms_copy(relation->rd_indexattr);
case INDEX_ATTR_BITMAP_PROJ:
return bms_copy(relation->rd_projindexattr);
case INDEX_ATTR_BITMAP_KEY:
return bms_copy(relation->rd_keyattr);
case INDEX_ATTR_BITMAP_PRIMARY_KEY:
@ -4890,9 +4966,12 @@ restart:
* won't be returned at all by RelationGetIndexList.
*/
indexattrs = NULL;
projindexattrs = NULL;
uindexattrs = NULL;
pkindexattrs = NULL;
idindexattrs = NULL;
projindexes = NULL;
indexno = 0;
foreach(l, indexoidlist)
{
Oid indexOid = lfirst_oid(l);
@ -4943,13 +5022,22 @@ restart:
}
}
/* Collect all attributes used in expressions, too */
pull_varattnos((Node *) indexInfo->ii_Expressions, 1, &indexattrs);
/* Collect attributes used in expressions, too */
if (IsProjectionFunctionalIndex(indexDesc, indexInfo))
{
projindexes = bms_add_member(projindexes, indexno);
pull_varattnos((Node *) indexInfo->ii_Expressions, 1, &projindexattrs);
}
else
{
/* Collect all attributes used in expressions, too */
pull_varattnos((Node *) indexInfo->ii_Expressions, 1, &indexattrs);
}
/* Collect all attributes in the index predicate, too */
pull_varattnos((Node *) indexInfo->ii_Predicate, 1, &indexattrs);
index_close(indexDesc, AccessShareLock);
indexno += 1;
}
/*
@ -4976,6 +5064,8 @@ restart:
bms_free(pkindexattrs);
bms_free(idindexattrs);
bms_free(indexattrs);
bms_free(projindexattrs);
bms_free(projindexes);
goto restart;
}
@ -4983,12 +5073,16 @@ restart:
/* Don't leak the old values of these bitmaps, if any */
bms_free(relation->rd_indexattr);
relation->rd_indexattr = NULL;
bms_free(relation->rd_projindexattr);
relation->rd_projindexattr = NULL;
bms_free(relation->rd_keyattr);
relation->rd_keyattr = NULL;
bms_free(relation->rd_pkattr);
relation->rd_pkattr = NULL;
bms_free(relation->rd_idattr);
relation->rd_idattr = NULL;
bms_free(relation->rd_projidx);
relation->rd_projidx = NULL;
/*
* Now save copies of the bitmaps in the relcache entry. We intentionally
@ -5002,13 +5096,17 @@ restart:
relation->rd_pkattr = bms_copy(pkindexattrs);
relation->rd_idattr = bms_copy(idindexattrs);
relation->rd_indexattr = bms_copy(indexattrs);
relation->rd_projindexattr = bms_copy(projindexattrs);
relation->rd_projidx = bms_copy(projindexes);
MemoryContextSwitchTo(oldcxt);
/* We return our original working copy for caller to play with */
switch (attrKind)
{
case INDEX_ATTR_BITMAP_ALL:
case INDEX_ATTR_BITMAP_HOT:
return indexattrs;
case INDEX_ATTR_BITMAP_PROJ:
return projindexattrs;
case INDEX_ATTR_BITMAP_KEY:
return uindexattrs;
case INDEX_ATTR_BITMAP_PRIMARY_KEY:
@ -5632,9 +5730,11 @@ load_relcache_init_file(bool shared)
rel->rd_pkindex = InvalidOid;
rel->rd_replidindex = InvalidOid;
rel->rd_indexattr = NULL;
rel->rd_projindexattr = NULL;
rel->rd_keyattr = NULL;
rel->rd_pkattr = NULL;
rel->rd_idattr = NULL;
rel->rd_projidx = NULL;
rel->rd_pubactions = NULL;
rel->rd_statvalid = false;
rel->rd_statlist = NIL;