Peter Eisentraut <peter_e@gmx.net> writes:
> Tom Lane writes:
>> Had you actually done anything to the database between postmaster
>> startup and kill?
> I've run the regression tests several times (parallel and serial) and
> killed the postmaster at different places, even killed a few backends in
> between, actual redo happened, yet shutdown was invariably TRUE.
Hm. Do you have an especially long intra-checkpoint interval set in
postgresql.conf? I'd expect that if you'd done anything to the db and
then waited at least a checkpoint interval (or done a manual CHECKPOINT)
before killing the postmaster, you'd find a non-shutdown checkpoint
record. That's what I get anyway.
> Another question I have is what is the significance of
> ReadRecord: record with zero length at 0/D65C00
> ? It seems to occur at the end of every redo run, perhaps it simply means
> end of records.
Yeah, that would be the normal symptom of reaching the end of the log.
> At least it's not clear to the user whether this is debug, info,
> warning, or error.
It's debug, and so labeled.
Possibly we need more elog levels than we have --- the stuff that comes
out at startup is not all of the same urgency, but DEBUG is the only
elog level we can use for it, really...
regards, tom lane