Re: pgsql: Accept that server might not be able to send error in crash reco

On 2017-09-22 13:30:14 -0400, Tom Lane wrote:
> Andres Freund <andres(at)anarazel(dot)de> writes:
> > Accept that server might not be able to send error in crash recovery test.
>
> piculet says you didn't cover all the bases:
>
> https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=piculet&dt=2017-09-22%2017%3A10%3A01
>
> Looks like you missed adding the "server closed the connection
> unexpectedly" alternative to one of the three places.

I'd hoped^Wassumed it couldn't happen in that case. I'll expand the
expected output. FWIW, I kinda like the new debugging output for this -
makes it much easier to diagnose.

Besides this individual test, I think we really need to do something
that increases the likelihood of getting these error messages to
clients.

Greetings,

Andres Freund