On 24.08.23 00:56, Andres Freund wrote:
> Hi,
>
> On 2023-08-23 18:32:26 -0400, Tom Lane wrote:
>> Andres Freund <andres@anarazel.de> writes:
>>> There are other potential uses for libpq in pg_regress though - I'd e.g. like
>>> to have a "monitoring" session open, which we could use to detect that the
>>> server crashed (by waiting for the FD to be become invalid). Where the
>>> connection default issue could matter more?
>>
>> Meh. I don't find that idea compelling enough to justify adding
>> restrictions on what test scenarios will work. It's seldom hard to
>> tell from the test output whether the server crashed.
>
> I find it pretty painful to wade through a several-megabyte regression.diffs
> to find the cause of a crash. I think we ought to use
> restart_after_crash=false, since after a crash there's no hope for the tests
> to succeed, but even in that case, we end up with a lot of pointless contents
> in regression.diffs. If we instead realized that we shouldn't start further
> tests, we'd limit that by a fair bit.
I once coded it up so that if the server crashes during a test, it would
wait until it recovers before running the next test. I found that
useful. I agree the current behavior is not useful in any case.