Re: Log rotation
От | Fernando Nasser |
---|---|
Тема | Re: Log rotation |
Дата | |
Msg-id | 4051C81C.80909@redhat.com обсуждение исходный текст |
Ответ на | Re: Log rotation (Tom Lane <tgl@sss.pgh.pa.us>) |
Ответы |
Re: Log rotation
Re: Log rotation |
Список | pgsql-hackers |
Tom Lane wrote: > Fernando Nasser <fnasser@redhat.com> writes: > >>Please remind me again why the postmaster cannot close and open the log >>file when it receives a SIGHUP (to re-read configuration)? > > > (a) Because it never opened it in the first place --- the log file is > whatever was passed as stderr. > > (b) Because it would not be sufficient to make the postmaster itself > close and reopen the file; every child process would have to do so also. > Doing this in any sort of synchronized fashion seems impossible. > Now I remember. Thanks for reminding me of that. > It's much cleaner to have stderr be a pipe to some separate collector > program that can handle log rotation (ie, the Apache solution). > We could also create a pipe and start a new process ("logger") and give it the other end of the pipe and the name of the log file. We could send it a SIGHUP after we reread the configuration file. But just doing a pipe on the OS level is way simpler. I don't really care on how its done, but IMO an enterprise class database must be able to do log rotation. Logging to Syslog is not an option (specially with our verbosity) -- users must be able to use flat files for logging. I never seem some many customer complaints and bug reports about a single item like the ones we have received here about logging. I think this should be the number 1 item in te TODO list. Thanks again for the clarifications. Regards, Fernando
В списке pgsql-hackers по дате отправления: