Andres Freund <andres(at)anarazel(dot)de> writes:
> On April 29, 2019 9:37:33 PM PDT, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>> Seems like putting reindexes of pg_class into a test script that runs
>> in parallel with other DDL wasn't a hot idea.
> Saw that. Will try to reproduce (and if necessary either run separately or revert). But isn't that somewhat broken? They're not run in a transaction, so the locking shouldn't be deadlock prone.
Hm? REINDEX INDEX is deadlock-prone by definition, because it starts
by opening/locking the index and then it has to open/lock the index's
table. Every other operation locks tables before their indexes.
regards, tom lane