On Mon, Mar 09, 2020 at 03:15:58PM -0700, Andres Freund wrote:
> On 2020-03-07 22:17:09 +0100, Julien Rouhaud wrote:
>> For reindex concurrently, a SELECT FOR UPDATE on a different connection can
>> ensure that the reindex will be stuck at some point, so canceling the command
>> after a long enough timeout reproduces the original faulty behavior.
>
> That kind of thing can already be done using statement_timeout or
> lock_timeout, no?
Yep, still that's not something I would recommend to commit in the
tree as that's a double-edged sword as you already know. For slower
machines, you need a statement_timeout large enough so as you make
sure that the state you want the query to wait for is reached, which
has a cost on all other faster machines as it makes the tests slower.
--
Michael