On 3/7/18, 9:34 AM, "Bossart, Nathan" <bossartn@amazon.com> wrote:
> On 3/6/18, 11:04 PM, "Michael Paquier" <michael@paquier.xyz> wrote:
>> + if (!(options & VACOPT_SKIP_LOCKED))
>> + relid = RangeVarGetRelid(vrel->relation, AccessShareLock,
>> false);
>> + else
>> + {
>> + relid = RangeVarGetRelid(vrel->relation, NoLock, false);
>> Yeah, I agree with Andres that getting all this logic done in
>> RangeVarGetRelidExtended would be cleaner. Let's see where the other
>> thread leads us to:
>> https://www.postgresql.org/message-id/20180306005349.b65whmvj7z6hbe2y%40alap3.anarazel.de
>
> I had missed that thread. Thanks for pointing it out.
I've noticed one more problem with ACCESS EXCLUSIVE. If an ACCESS
EXCLUSIVE lock is held on a child relation of a partitioned table,
an ANALYZE on the partitioned table will be blocked at
acquire_inherited_sample_rows().
static int
acquire_inherited_sample_rows(Relation onerel, int elevel,
HeapTuple *rows, int targrows,
double *totalrows, double *totaldeadrows)
{
...
/*
* Find all members of inheritance set. We only need AccessShareLock on
* the children.
*/
tableOIDs =
find_all_inheritors(RelationGetRelid(onerel), AccessShareLock, NULL);
It also seems possible for the call to vac_open_indexes() in
do_analyze_rel() to block.
/*
* Open all indexes of the relation, and see if there are any analyzable
* columns in the indexes. We do not analyze index columns if there was
* an explicit column list in the ANALYZE command, however. If we are
* doing a recursive scan, we don't want to touch the parent's indexes at
* all.
*/
if (!inh)
vac_open_indexes(onerel, AccessShareLock, &nindexes, &Irel);
else
{
Irel = NULL;
nindexes = 0;
}
I think the most straightforward approach for fixing this is to add
skip-locked functionality in find_all_inheritors(),
find_inheritance_children(), and vac_open_indexes(), but I am curious
what others think.
Nathan
