Re: [HACKERS] Re: [BUGS] syslog logging setup broken?
The man page suggests that nohup is required to init postmaster, I know this isn't true but to implement an example init file and not match up with the man page seemed foolish. I guess nohup would stop postmaster doing something awfull if it doesn't handle HUP properly but I very much doubt that you guys fail to handle HUP. Good point. postmaster in 7.1 uses HUP signal to re-read postgresql.conf. It seems we should not use nohup to start postmaster. -- Tatsuo Ishii
Re: [HACKERS] Re: [BUGS] syslog logging setup broken?
Bruce Momjian [EMAIL PROTECTED] 04-Feb-01 10:07:40 PM The bottom line is that, IMHO, writing a portable init.d style (or any other such concept) startup file that is ready for blind use is beyond practicality. It might be better to collect a few of the ones that are being used now (Red Hat-style, SuSE-style, Debian, *BSD-style) and ship them. This should be coordinated with the packagers, though. Should I remove init.d from /contrib? I'm just a postgres user but I don't agree with Peter. I think the file is valuable. The file is valuable for people not using a distribution such as Debian, etc... and also is usefull to people developing packages for distributions. I don't use a packaged postgres and it was certainly valuable to me because it served as an example of what I had to do to get postgres going quickly in the way that I wanted. I sent Peter an updated file that IMHO irons out some problems which may cause Peter to consider the file broken: - ouptut was being piped to the logger if "syslog" was on It's not necessary to do that because postgres handles the decision about syslog depending on the conf file. - the postmaster was being started without nohup - the system for setting options wasn't very usefull the system that I've replaced it with isn't terribly usefull either but it works. So anyway, my view as a user is that it's usefull and that a package specific version would come with the package anyway. Nic Ferrier Tapsell-Ferrier Limited
Re: [HACKERS] Re: [BUGS] syslog logging setup broken?
"Oliver Elphick" [EMAIL PROTECTED] writes: "Nic Ferrier" wrote: - the postmaster was being started without nohup If postmaster is being started by init, it should not need nohup, because init never exits and postmaster is not going to get shutdown unexpectedly. On the other hand, when pg_ctl is invoked by hand, it probably is a bug that it fails to use nohup. Perhaps this is a reason why pg_ctl should not be made into a substitute for a startup script? regards, tom lane
Re: [HACKERS] Re: [BUGS] syslog logging setup broken?
If postmaster is being started by init, it should not need nohup, because init never exits and postmaster is not going to get shutdown unexpectedly. On the other hand, when pg_ctl is invoked by hand, it probably is a bug that it fails to use nohup. Perhaps this is a reason why pg_ctl should not be made into a substitute for a startup script? If pg_ctl unconditionally use nohup, it might be a performance penalty as Oliver mentioned. nohup has a performance cost, in that (at least on Linux) it automatically nices (lowers the priority of) the process. You may not want the priority lowered... Moreover if postmaster detaches itself to be a deamon, nohup is not necessary at all. BTW, for the startup script, I don't think we need to use pg_ctl. Invoking postmaster directry seems enough for me. The only reason for using pg_ctl to start postmaster is waiting for postmaster up and running. In most cases the time to recover DB would not be so long. And if the recovery took too long time, we would not want to be blocked in the middle of the boot sequence anyway. Comments? -- Tatsuo Ishii
Re: [HACKERS] Re: [BUGS] syslog logging setup broken?
Tom Lane writes: Peter Eisentraut [EMAIL PROTECTED] writes: The example startup file is outdated and broken. Don't use it. Er ... shouldn't we fix it? Or remove it? See my message on -hackers, "Sparc/Linux patch" thing follow-up. The bottom line is that, IMHO, writing a portable init.d style (or any other such concept) startup file that is ready for blind use is beyond practicality. It might be better to collect a few of the ones that are being used now (Red Hat-style, SuSE-style, Debian, *BSD-style) and ship them. This should be coordinated with the packagers, though. Should I remove init.d from /contrib? -- Bruce Momjian| http://candle.pha.pa.us [EMAIL PROTECTED] | (610) 853-3000 + If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup.| Drexel Hill, Pennsylvania 19026
Re: [HACKERS] Re: [BUGS] syslog logging setup broken?
Nic Ferrier writes: Should I remove init.d from /contrib? I'm just a postgres user but I don't agree with Peter. I think the file is valuable. I didn't say it wasn't valuable, I just said it didn't work... This problem seems to call for a more general solution than a handful of example style files. -- Peter Eisentraut [EMAIL PROTECTED] http://yi.org/peter-e/