On 3/21/21 12:38 PM, Tom Lane wrote:
> Andrew Dunstan <andrew@dunslane.net> writes:
>> Unbreak recovery test on Windows
> Hmm, looks like this broke things on other machines.
> crake and sifaka (so far) are showing
>
> # Running: pg_ctl kill KILL 2237831
> ok 3 - killed process with KILL
> ack Broken pipe: write( 13, '\\q
> ' ) at /usr/share/perl5/vendor_perl/IPC/Run/IO.pm line 549.
>
> which looks like a race condition: did we manage to stuff the "\q"
> command down the pipe before killing the psql process, or not?
>
>
Ugh. Is there any reason we need to do those kills before we end the
psql processes? If not I'm tempted just to move them - after the psql's
are finished it should be safe. At any rate I'll go and test that.
cheers
andrew
--
Andrew Dunstan
EDB: https://www.enterprisedb.com