Re: write_pipe_chunks patch messes up early error message output

Andrew Dunstan <andrew(at)dunslane(dot)net> writes:
> Tom Lane wrote:
>> Yeah, that would work. You'd have to get rid of the current ad-hoc
>> method by which it is propagated to the syslogger child process
>> (EXEC_BACKEND case), because now it will have to be propagated to all
>> children; so postmaster.c should handle it in BackendParameters.

> The problem with this as it stands is that the syslogger itself is
> forked before the redirection is done.

Which is entirely correct. Re-read what I said about first launch vs
relaunch of the syslogger. Its stderr will be connected differently in
the two cases, and should be handled differently --- we want the first
launch to try to report problems on its own stderr, but there's no point
after a relaunch. That's why we pass down redirection_done to it.

regards, tom lane