Fix index-only scan plans, take 2.

Commit 4ace45677 failed to fix the problem fully, because the
same issue of attempting to fetch a non-returnable index column
can occur when rechecking the indexqual after using a lossy index
operator.  Moreover, it broke EXPLAIN for such indexquals (which
indicates a gap in our test cases :-().

Revert the code changes of 4ace45677 in favor of adding a new field
to struct IndexOnlyScan, containing a version of the indexqual that
can be executed against the index-returned tuple without using any
non-returnable columns.  (The restrictions imposed by check_index_only
guarantee this is possible, although we may have to recompute indexed
expressions.)  Support construction of that during setrefs.c
processing by marking IndexOnlyScan.indextlist entries as resjunk
if they can't be returned, rather than removing them entirely.
(We could alternatively require setrefs.c to look up the IndexOptInfo
again, but abusing resjunk this way seems like a reasonably safe way
to avoid needing to do that.)

This solution isn't great from an API-stability standpoint: if there
are any extensions out there that build IndexOnlyScan structs directly,
they'll be broken in the next minor releases.  However, only a very
invasive extension would be likely to do such a thing.  There's no
change in the Path representation, so typical planner extensions
shouldn't have a problem.

As before, back-patch to all supported branches.

Discussion: https://postgr.es/m/3179992.1641150853@sss.pgh.pa.us
Discussion: https://postgr.es/m/17350-b5bdcf476e5badbb@postgresql.org
This commit is contained in:
Tom Lane 2022-01-03 15:42:27 -05:00
parent 4b160492b9
commit 9a3ddeb519
12 changed files with 124 additions and 66 deletions

View File

@ -1746,7 +1746,7 @@ ExplainNode(PlanState *planstate, List *ancestors,
case T_IndexOnlyScan: case T_IndexOnlyScan:
show_scan_qual(((IndexOnlyScan *) plan)->indexqual, show_scan_qual(((IndexOnlyScan *) plan)->indexqual,
"Index Cond", planstate, ancestors, es); "Index Cond", planstate, ancestors, es);
if (((IndexOnlyScan *) plan)->indexqual) if (((IndexOnlyScan *) plan)->recheckqual)
show_instrumentation_count("Rows Removed by Index Recheck", 2, show_instrumentation_count("Rows Removed by Index Recheck", 2,
planstate, es); planstate, es);
show_scan_qual(((IndexOnlyScan *) plan)->indexorderby, show_scan_qual(((IndexOnlyScan *) plan)->indexorderby,

View File

@ -214,13 +214,11 @@ IndexOnlyNext(IndexOnlyScanState *node)
/* /*
* If the index was lossy, we have to recheck the index quals. * If the index was lossy, we have to recheck the index quals.
* (Currently, this can never happen, but we should support the case
* for possible future use, eg with GiST indexes.)
*/ */
if (scandesc->xs_recheck) if (scandesc->xs_recheck)
{ {
econtext->ecxt_scantuple = slot; econtext->ecxt_scantuple = slot;
if (!ExecQualAndReset(node->indexqual, econtext)) if (!ExecQualAndReset(node->recheckqual, econtext))
{ {
/* Fails recheck, so drop it and loop back for another */ /* Fails recheck, so drop it and loop back for another */
InstrCountFiltered2(node, 1); InstrCountFiltered2(node, 1);
@ -555,8 +553,8 @@ ExecInitIndexOnlyScan(IndexOnlyScan *node, EState *estate, int eflags)
*/ */
indexstate->ss.ps.qual = indexstate->ss.ps.qual =
ExecInitQual(node->scan.plan.qual, (PlanState *) indexstate); ExecInitQual(node->scan.plan.qual, (PlanState *) indexstate);
indexstate->indexqual = indexstate->recheckqual =
ExecInitQual(node->indexqual, (PlanState *) indexstate); ExecInitQual(node->recheckqual, (PlanState *) indexstate);
/* /*
* If we are just doing EXPLAIN (ie, aren't going to run the plan), stop * If we are just doing EXPLAIN (ie, aren't going to run the plan), stop

View File

@ -519,6 +519,7 @@ _copyIndexOnlyScan(const IndexOnlyScan *from)
*/ */
COPY_SCALAR_FIELD(indexid); COPY_SCALAR_FIELD(indexid);
COPY_NODE_FIELD(indexqual); COPY_NODE_FIELD(indexqual);
COPY_NODE_FIELD(recheckqual);
COPY_NODE_FIELD(indexorderby); COPY_NODE_FIELD(indexorderby);
COPY_NODE_FIELD(indextlist); COPY_NODE_FIELD(indextlist);
COPY_SCALAR_FIELD(indexorderdir); COPY_SCALAR_FIELD(indexorderdir);

View File

@ -580,6 +580,7 @@ _outIndexOnlyScan(StringInfo str, const IndexOnlyScan *node)
WRITE_OID_FIELD(indexid); WRITE_OID_FIELD(indexid);
WRITE_NODE_FIELD(indexqual); WRITE_NODE_FIELD(indexqual);
WRITE_NODE_FIELD(recheckqual);
WRITE_NODE_FIELD(indexorderby); WRITE_NODE_FIELD(indexorderby);
WRITE_NODE_FIELD(indextlist); WRITE_NODE_FIELD(indextlist);
WRITE_ENUM_FIELD(indexorderdir, ScanDirection); WRITE_ENUM_FIELD(indexorderdir, ScanDirection);

View File

@ -1884,6 +1884,7 @@ _readIndexOnlyScan(void)
READ_OID_FIELD(indexid); READ_OID_FIELD(indexid);
READ_NODE_FIELD(indexqual); READ_NODE_FIELD(indexqual);
READ_NODE_FIELD(recheckqual);
READ_NODE_FIELD(indexorderby); READ_NODE_FIELD(indexorderby);
READ_NODE_FIELD(indextlist); READ_NODE_FIELD(indextlist);
READ_ENUM_FIELD(indexorderdir, ScanDirection); READ_ENUM_FIELD(indexorderdir, ScanDirection);

View File

@ -20,7 +20,6 @@
#include <math.h> #include <math.h>
#include "access/sysattr.h" #include "access/sysattr.h"
#include "catalog/pg_am.h"
#include "catalog/pg_class.h" #include "catalog/pg_class.h"
#include "foreign/fdwapi.h" #include "foreign/fdwapi.h"
#include "miscadmin.h" #include "miscadmin.h"
@ -189,10 +188,10 @@ static IndexScan *make_indexscan(List *qptlist, List *qpqual, Index scanrelid,
ScanDirection indexscandir); ScanDirection indexscandir);
static IndexOnlyScan *make_indexonlyscan(List *qptlist, List *qpqual, static IndexOnlyScan *make_indexonlyscan(List *qptlist, List *qpqual,
Index scanrelid, Oid indexid, Index scanrelid, Oid indexid,
List *indexqual, List *indexorderby, List *indexqual, List *recheckqual,
List *indexorderby,
List *indextlist, List *indextlist,
ScanDirection indexscandir); ScanDirection indexscandir);
static List *make_indexonly_tlist(IndexOptInfo *indexinfo);
static BitmapIndexScan *make_bitmap_indexscan(Index scanrelid, Oid indexid, static BitmapIndexScan *make_bitmap_indexscan(Index scanrelid, Oid indexid,
List *indexqual, List *indexqual,
List *indexqualorig); List *indexqualorig);
@ -623,7 +622,7 @@ create_scan_plan(PlannerInfo *root, Path *best_path, int flags)
if (best_path->pathtype == T_IndexOnlyScan) if (best_path->pathtype == T_IndexOnlyScan)
{ {
/* For index-only scan, the preferred tlist is the index's */ /* For index-only scan, the preferred tlist is the index's */
tlist = copyObject(make_indexonly_tlist(((IndexPath *) best_path)->indexinfo)); tlist = copyObject(((IndexPath *) best_path)->indexinfo->indextlist);
/* /*
* Transfer sortgroupref data to the replacement tlist, if * Transfer sortgroupref data to the replacement tlist, if
@ -2934,7 +2933,8 @@ create_indexscan_plan(PlannerInfo *root,
List *indexclauses = best_path->indexclauses; List *indexclauses = best_path->indexclauses;
List *indexorderbys = best_path->indexorderbys; List *indexorderbys = best_path->indexorderbys;
Index baserelid = best_path->path.parent->relid; Index baserelid = best_path->path.parent->relid;
Oid indexoid = best_path->indexinfo->indexoid; IndexOptInfo *indexinfo = best_path->indexinfo;
Oid indexoid = indexinfo->indexoid;
List *qpqual; List *qpqual;
List *stripped_indexquals; List *stripped_indexquals;
List *fixed_indexquals; List *fixed_indexquals;
@ -3064,6 +3064,24 @@ create_indexscan_plan(PlannerInfo *root,
} }
} }
/*
* For an index-only scan, we must mark indextlist entries as resjunk if
* they are columns that the index AM can't return; this cues setrefs.c to
* not generate references to those columns.
*/
if (indexonly)
{
int i = 0;
foreach(l, indexinfo->indextlist)
{
TargetEntry *indextle = (TargetEntry *) lfirst(l);
indextle->resjunk = !indexinfo->canreturn[i];
i++;
}
}
/* Finally ready to build the plan node */ /* Finally ready to build the plan node */
if (indexonly) if (indexonly)
scan_plan = (Scan *) make_indexonlyscan(tlist, scan_plan = (Scan *) make_indexonlyscan(tlist,
@ -3071,8 +3089,9 @@ create_indexscan_plan(PlannerInfo *root,
baserelid, baserelid,
indexoid, indexoid,
fixed_indexquals, fixed_indexquals,
stripped_indexquals,
fixed_indexorderbys, fixed_indexorderbys,
make_indexonly_tlist(best_path->indexinfo), indexinfo->indextlist,
best_path->indexscandir); best_path->indexscandir);
else else
scan_plan = (Scan *) make_indexscan(tlist, scan_plan = (Scan *) make_indexscan(tlist,
@ -5444,6 +5463,7 @@ make_indexonlyscan(List *qptlist,
Index scanrelid, Index scanrelid,
Oid indexid, Oid indexid,
List *indexqual, List *indexqual,
List *recheckqual,
List *indexorderby, List *indexorderby,
List *indextlist, List *indextlist,
ScanDirection indexscandir) ScanDirection indexscandir)
@ -5458,6 +5478,7 @@ make_indexonlyscan(List *qptlist,
node->scan.scanrelid = scanrelid; node->scan.scanrelid = scanrelid;
node->indexid = indexid; node->indexid = indexid;
node->indexqual = indexqual; node->indexqual = indexqual;
node->recheckqual = recheckqual;
node->indexorderby = indexorderby; node->indexorderby = indexorderby;
node->indextlist = indextlist; node->indextlist = indextlist;
node->indexorderdir = indexscandir; node->indexorderdir = indexscandir;
@ -5465,53 +5486,6 @@ make_indexonlyscan(List *qptlist,
return node; return node;
} }
/*
* make_indexonly_tlist
*
* Construct the indextlist for an IndexOnlyScan plan node.
* We must replace any column that can't be returned by the index AM
* with a null Const of the appropriate datatype. This is necessary
* to prevent setrefs.c from trying to use the value of such a column,
* and anyway it makes the indextlist a better representative of what
* the indexscan will really return. (We do this here, not where the
* IndexOptInfo is originally constructed, because earlier planner
* steps need to know what is in such columns.)
*/
static List *
make_indexonly_tlist(IndexOptInfo *indexinfo)
{
List *result;
int i;
ListCell *lc;
/* We needn't work hard for the common case of btrees. */
if (indexinfo->relam == BTREE_AM_OID)
return indexinfo->indextlist;
result = NIL;
i = 0;
foreach(lc, indexinfo->indextlist)
{
TargetEntry *indextle = (TargetEntry *) lfirst(lc);
if (indexinfo->canreturn[i])
result = lappend(result, indextle);
else
{
TargetEntry *newtle = makeNode(TargetEntry);
Node *texpr = (Node *) indextle->expr;
memcpy(newtle, indextle, sizeof(TargetEntry));
newtle->expr = (Expr *) makeNullConst(exprType(texpr),
exprTypmod(texpr),
exprCollation(texpr));
result = lappend(result, newtle);
}
i++;
}
return result;
}
static BitmapIndexScan * static BitmapIndexScan *
make_bitmap_indexscan(Index scanrelid, make_bitmap_indexscan(Index scanrelid,
Oid indexid, Oid indexid,

View File

@ -1142,8 +1142,26 @@ set_indexonlyscan_references(PlannerInfo *root,
int rtoffset) int rtoffset)
{ {
indexed_tlist *index_itlist; indexed_tlist *index_itlist;
List *stripped_indextlist;
ListCell *lc;
index_itlist = build_tlist_index(plan->indextlist); /*
* Vars in the plan node's targetlist, qual, and recheckqual must only
* reference columns that the index AM can actually return. To ensure
* this, remove non-returnable columns (which are marked as resjunk) from
* the indexed tlist. We can just drop them because the indexed_tlist
* machinery pays attention to TLE resnos, not physical list position.
*/
stripped_indextlist = NIL;
foreach(lc, plan->indextlist)
{
TargetEntry *indextle = (TargetEntry *) lfirst(lc);
if (!indextle->resjunk)
stripped_indextlist = lappend(stripped_indextlist, indextle);
}
index_itlist = build_tlist_index(stripped_indextlist);
plan->scan.scanrelid += rtoffset; plan->scan.scanrelid += rtoffset;
plan->scan.plan.targetlist = (List *) plan->scan.plan.targetlist = (List *)
@ -1160,6 +1178,13 @@ set_indexonlyscan_references(PlannerInfo *root,
INDEX_VAR, INDEX_VAR,
rtoffset, rtoffset,
NUM_EXEC_QUAL((Plan *) plan)); NUM_EXEC_QUAL((Plan *) plan));
plan->recheckqual = (List *)
fix_upper_expr(root,
(Node *) plan->recheckqual,
index_itlist,
INDEX_VAR,
rtoffset,
NUM_EXEC_QUAL((Plan *) plan));
/* indexqual is already transformed to reference index columns */ /* indexqual is already transformed to reference index columns */
plan->indexqual = fix_scan_list(root, plan->indexqual, plan->indexqual = fix_scan_list(root, plan->indexqual,
rtoffset, 1); rtoffset, 1);

View File

@ -2336,6 +2336,8 @@ finalize_plan(PlannerInfo *root, Plan *plan,
case T_IndexOnlyScan: case T_IndexOnlyScan:
finalize_primnode((Node *) ((IndexOnlyScan *) plan)->indexqual, finalize_primnode((Node *) ((IndexOnlyScan *) plan)->indexqual,
&context); &context);
finalize_primnode((Node *) ((IndexOnlyScan *) plan)->recheckqual,
&context);
finalize_primnode((Node *) ((IndexOnlyScan *) plan)->indexorderby, finalize_primnode((Node *) ((IndexOnlyScan *) plan)->indexorderby,
&context); &context);

View File

@ -1490,7 +1490,7 @@ typedef struct IndexScanState
/* ---------------- /* ----------------
* IndexOnlyScanState information * IndexOnlyScanState information
* *
* indexqual execution state for indexqual expressions * recheckqual execution state for recheckqual expressions
* ScanKeys Skey structures for index quals * ScanKeys Skey structures for index quals
* NumScanKeys number of ScanKeys * NumScanKeys number of ScanKeys
* OrderByKeys Skey structures for index ordering operators * OrderByKeys Skey structures for index ordering operators
@ -1509,7 +1509,7 @@ typedef struct IndexScanState
typedef struct IndexOnlyScanState typedef struct IndexOnlyScanState
{ {
ScanState ss; /* its first field is NodeTag */ ScanState ss; /* its first field is NodeTag */
ExprState *indexqual; ExprState *recheckqual;
struct ScanKeyData *ioss_ScanKeys; struct ScanKeyData *ioss_ScanKeys;
int ioss_NumScanKeys; int ioss_NumScanKeys;
struct ScanKeyData *ioss_OrderByKeys; struct ScanKeyData *ioss_OrderByKeys;

View File

@ -420,15 +420,28 @@ typedef struct IndexScan
* index-only scan, in which the data comes from the index not the heap. * index-only scan, in which the data comes from the index not the heap.
* Because of this, *all* Vars in the plan node's targetlist, qual, and * Because of this, *all* Vars in the plan node's targetlist, qual, and
* index expressions reference index columns and have varno = INDEX_VAR. * index expressions reference index columns and have varno = INDEX_VAR.
* Hence we do not need separate indexqualorig and indexorderbyorig lists, *
* since their contents would be equivalent to indexqual and indexorderby. * We could almost use indexqual directly against the index's output tuple
* when rechecking lossy index operators, but that won't work for quals on
* index columns that are not retrievable. Hence, recheckqual is needed
* for rechecks: it expresses the same condition as indexqual, but using
* only index columns that are retrievable. (We will not generate an
* index-only scan if this is not possible. An example is that if an
* index has table column "x" in a retrievable index column "ind1", plus
* an expression f(x) in a non-retrievable column "ind2", an indexable
* query on f(x) will use "ind2" in indexqual and f(ind1) in recheckqual.
* Without the "ind1" column, an index-only scan would be disallowed.)
*
* We don't currently need a recheckable equivalent of indexorderby,
* because we don't support lossy operators in index ORDER BY.
* *
* To help EXPLAIN interpret the index Vars for display, we provide * To help EXPLAIN interpret the index Vars for display, we provide
* indextlist, which represents the contents of the index as a targetlist * indextlist, which represents the contents of the index as a targetlist
* with one TLE per index column. Vars appearing in this list reference * with one TLE per index column. Vars appearing in this list reference
* the base table, and this is the only field in the plan node that may * the base table, and this is the only field in the plan node that may
* contain such Vars. Note however that index columns that the AM can't * contain such Vars. Also, for the convenience of setrefs.c, TLEs in
* reconstruct are replaced by null Consts in indextlist. * indextlist are marked as resjunk if they correspond to columns that
* the index AM cannot reconstruct.
* ---------------- * ----------------
*/ */
typedef struct IndexOnlyScan typedef struct IndexOnlyScan
@ -436,6 +449,7 @@ typedef struct IndexOnlyScan
Scan scan; Scan scan;
Oid indexid; /* OID of index to scan */ Oid indexid; /* OID of index to scan */
List *indexqual; /* list of index quals (usually OpExprs) */ List *indexqual; /* list of index quals (usually OpExprs) */
List *recheckqual; /* index quals in recheckable form */
List *indexorderby; /* list of index ORDER BY exprs */ List *indexorderby; /* list of index ORDER BY exprs */
List *indextlist; /* TargetEntry list describing index's cols */ List *indextlist; /* TargetEntry list describing index's cols */
ScanDirection indexorderdir; /* forward or backward or don't care */ ScanDirection indexorderdir; /* forward or backward or don't care */

View File

@ -340,6 +340,37 @@ where p <@ box(point(5, 5), point(5.3, 5.3));
<(5.3,5.3),1> <(5.3,5.3),1>
(7 rows) (7 rows)
-- Similarly, test that index rechecks involving a non-returnable column
-- are done correctly.
explain (verbose, costs off)
select p from gist_tbl where circle(p,1) @> circle(point(0,0),0.95);
QUERY PLAN
---------------------------------------------------------------------------------------
Index Only Scan using gist_tbl_multi_index on public.gist_tbl
Output: p
Index Cond: ((circle(gist_tbl.p, '1'::double precision)) @> '<(0,0),0.95>'::circle)
(3 rows)
select p from gist_tbl where circle(p,1) @> circle(point(0,0),0.95);
p
-------
(0,0)
(1 row)
-- This case isn't supported, but it should at least EXPLAIN correctly.
explain (verbose, costs off)
select p from gist_tbl order by circle(p,1) <-> point(0,0) limit 1;
QUERY PLAN
------------------------------------------------------------------------------------
Limit
Output: p, ((circle(p, '1'::double precision) <-> '(0,0)'::point))
-> Index Only Scan using gist_tbl_multi_index on public.gist_tbl
Output: p, (circle(p, '1'::double precision) <-> '(0,0)'::point)
Order By: ((circle(gist_tbl.p, '1'::double precision)) <-> '(0,0)'::point)
(5 rows)
select p from gist_tbl order by circle(p,1) <-> point(0,0) limit 1;
ERROR: lossy distance functions are not supported in index-only scans
-- Clean up -- Clean up
reset enable_seqscan; reset enable_seqscan;
reset enable_bitmapscan; reset enable_bitmapscan;

View File

@ -153,6 +153,17 @@ where p <@ box(point(5, 5), point(5.3, 5.3));
select circle(p,1) from gist_tbl select circle(p,1) from gist_tbl
where p <@ box(point(5, 5), point(5.3, 5.3)); where p <@ box(point(5, 5), point(5.3, 5.3));
-- Similarly, test that index rechecks involving a non-returnable column
-- are done correctly.
explain (verbose, costs off)
select p from gist_tbl where circle(p,1) @> circle(point(0,0),0.95);
select p from gist_tbl where circle(p,1) @> circle(point(0,0),0.95);
-- This case isn't supported, but it should at least EXPLAIN correctly.
explain (verbose, costs off)
select p from gist_tbl order by circle(p,1) <-> point(0,0) limit 1;
select p from gist_tbl order by circle(p,1) <-> point(0,0) limit 1;
-- Clean up -- Clean up
reset enable_seqscan; reset enable_seqscan;
reset enable_bitmapscan; reset enable_bitmapscan;