mirrors
/
postgres
mirror of https://github.com/postgres/postgres
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Fix partition pruning setup during DETACH CONCURRENTLY 

When detaching partition in concurrent mode, it's possible for partition
descriptors to not match the set that was recently seen when the plan
was made, causing an assertion failure or (in production builds) failure
to construct a working plan.  The case that was reported involves
prepared statements, but I think it may be possible to hit this bug
without that too.

The problem is that CreatePartitionPruneState is constructing a
PartitionPruneState under the assumption that new partitions can be
added, but never removed, but it turns out that this isn't true: a
prepared statement gets replanned when the DETACH CONCURRENTLY session
sends out its invalidation message, but if the invalidation message
arrives after ExecInitAppend started, we would build a partition
descriptor without the partition, and then CreatePartitionPruneState
would refuse to work with it.

CreatePartitionPruneState already contains code to deal with the new
descriptor having more partitions than before (and behaving for the
extra partitions as if they had been pruned), but doesn't have code to
deal with less partitions than before, and it is naïve about the case
where the number of partitions is the same.  We could simply add that a
new stanza for less partitions than before, and in simple testing it
works to do that; but it's possible to press the test scripts even
further and hit the case where one partition is added and a partition is
removed quickly enough that we see the same number of partitions, but
they don't actually match, causing hangs during execution.

To cope with both these problems, we now memcmp() the arrays of
partition OIDs, and do a more elaborate mapping (relying on the fact
that both OID arrays are in partition-bounds order) if they're not
identical.

This fix was already pushed in backbranches earlier.

Reported-by: yajun Hu <1026592243@qq.com>
Reviewed-by: Tender Wang <tndrwang@gmail.com>
Discussion: https://postgr.es/m/18377-e0324601cfebdfe5@postgresql.org
This commit is contained in:
Alvaro Herrera 2024-06-26 13:40:26 +02:00
parent 1bf29f51fa
commit dbca3469eb
No known key found for this signature in database
GPG Key ID: 1C20ACB9D5C564AE
1 changed files with 62 additions and 52 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

114
src/backend/executor/execPartition.c
View File

@ -1942,37 +1942,31 @@ CreatePartitionPruneState(PlanState *planstate, PartitionPruneInfo *pruneinfo)
/*
* Initialize the subplan_map and subpart_map.
*
* Because we request detached partitions to be included, and
* detaching waits for old transactions, it is safe to assume that
* no partitions have disappeared since this query was planned.
* The set of partitions that exist now might not be the same that
* existed when the plan was made. The normal case is that it is;
* optimize for that case with a quick comparison, and just copy
* the subplan_map and make subpart_map point to the one in
* PruneInfo.
*
* However, new partitions may have been added.
* For the case where they aren't identical, we could have more
* partitions on either side; or even exactly the same number of
* them on both but the set of OIDs doesn't match fully. Handle
* this by creating new subplan_map and subpart_map arrays that
* corresponds to the ones in the PruneInfo where the new
* partition descriptor's OIDs match. Any that don't match can be
* set to -1, as if they were pruned. Both arrays must be in
* numerical OID order.
*/
Assert(partdesc->nparts >= pinfo->nparts);
pprune->nparts = partdesc->nparts;
pprune->subplan_map = palloc(sizeof(int) * partdesc->nparts);
if (partdesc->nparts == pinfo->nparts)
if (partdesc->nparts == pinfo->nparts &&
memcmp(partdesc->oids, pinfo->relid_map,
sizeof(int) * partdesc->nparts) == 0)
{
/*
* There are no new partitions, so this is simple. We can
* simply point to the subpart_map from the plan, but we must
* copy the subplan_map since we may change it later.
*/
pprune->subpart_map = pinfo->subpart_map;
memcpy(pprune->subplan_map, pinfo->subplan_map,
sizeof(int) * pinfo->nparts);
/*
* Double-check that the list of unpruned relations has not
* changed. (Pruned partitions are not in relid_map[].)
*/
#ifdef USE_ASSERT_CHECKING
for (int k = 0; k < pinfo->nparts; k++)
{
Assert(partdesc->oids[k] == pinfo->relid_map[k] ||
pinfo->subplan_map[k] == -1);
}
#endif
}
else
{
@ -1980,22 +1974,18 @@ CreatePartitionPruneState(PlanState *planstate, PartitionPruneInfo *pruneinfo)
int pp_idx;
/*
* Some new partitions have appeared since plan time, and
* those are reflected in our PartitionDesc but were not
* present in the one used to construct subplan_map and
* subpart_map. So we must construct new and longer arrays
* where the partitions that were originally present map to
* the same sub-structures, and any added partitions map to
* -1, as if the new partitions had been pruned.
* When the partition arrays are not identical, there could be
* some new ones but it's also possible that one was removed;
* we cope with both situations by walking the arrays and
* discarding those that don't match.
*
* Note: pinfo->relid_map[] may contain InvalidOid entries for
* partitions pruned by the planner. We cannot tell exactly
* which of the partdesc entries these correspond to, but we
* don't have to; just skip over them. The non-pruned
* relid_map entries, however, had better be a subset of the
* partdesc entries and in the same order.
* If the number of partitions on both sides match, it's still
* possible that one partition has been detached and another
* attached. Cope with that by creating a map that skips any
* mismatches.
*/
pprune->subpart_map = palloc(sizeof(int) * partdesc->nparts);
for (pp_idx = 0; pp_idx < partdesc->nparts; pp_idx++)
{
/* Skip any InvalidOid relid_map entries */
@ -2003,6 +1993,7 @@ CreatePartitionPruneState(PlanState *planstate, PartitionPruneInfo *pruneinfo)
!OidIsValid(pinfo->relid_map[pd_idx]))
pd_idx++;
recheck:
if (pd_idx < pinfo->nparts &&
pinfo->relid_map[pd_idx] == partdesc->oids[pp_idx])
{
@ -2012,24 +2003,43 @@ CreatePartitionPruneState(PlanState *planstate, PartitionPruneInfo *pruneinfo)
pprune->subpart_map[pp_idx] =
pinfo->subpart_map[pd_idx];
pd_idx++;
continue;
}
else
{
/* this partdesc entry is not in the plan */
pprune->subplan_map[pp_idx] = -1;
pprune->subpart_map[pp_idx] = -1;
}
}
/*
* It might seem that we need to skip any trailing InvalidOid
* entries in pinfo->relid_map before checking that we scanned
* all of the relid_map. But we will have skipped them above,
* because they must correspond to some partdesc->oids
* entries; we just couldn't tell which.
*/
if (pd_idx != pinfo->nparts)
elog(ERROR, "could not match partition child tables to plan elements");
/*
* There isn't an exact match in the corresponding
* positions of both arrays. Peek ahead in
* pinfo->relid_map to see if we have a match for the
* current partition in partdesc. Normally if a match
* exists it's just one element ahead, and it means the
* planner saw one extra partition that we no longer see
* now (its concurrent detach finished just in between);
* so we skip that one by updating pd_idx to the new
* location and jumping above. We can then continue to
* match the rest of the elements after skipping the OID
* with no match; no future matches are tried for the
* element that was skipped, because we know the arrays to
* be in the same order.
*
* If we don't see a match anywhere in the rest of the
* pinfo->relid_map array, that means we see an element
* now that the planner didn't see, so mark that one as
* pruned and move on.
*/
for (int pd_idx2 = pd_idx + 1; pd_idx2 < pinfo->nparts; pd_idx2++)
{
if (pd_idx2 >= pinfo->nparts)
break;
if (pinfo->relid_map[pd_idx2] == partdesc->oids[pp_idx])
{
pd_idx = pd_idx2;
goto recheck;
}
}
pprune->subpart_map[pp_idx] = -1;
pprune->subplan_map[pp_idx] = -1;
}
}
/* present_parts is also subject to later modification */