* Bdale Garbee said: > In article <[🔎] 19991214104840.A17293@vip.net.pl> you wrote: > > > perhaps it would be better to just make the > > syslogd daemons restart all "affected" programs when doing an upgrade and/or > > restarting the syslog daemon? > > That's evil. Some daemons really want to be long-running regardless of what > is going on around them, like ntpd. Yes, that's correct. And losing connection to syslog is a bug of the program in question. > I can't help but ask... > > If the server needs to be so stable that bouncing named is considered harmful, > what are you doing upgrading syslogd? I wasn't talking about MY server - it has bind, but the database it manages isn't big, so the reload time doesn't matter. I am merely annoyed that I can't see bind messages in the log files anymore just because syslog has been upgraded. They are spilled all over the console - not very useful, I hope you agree with that. Also, if syslog happens to have a bug (and there have been such cases) one is better off upgrading it, right? Another thing is, that many people now will want to use syslog-ng which is (IMO) better, more secure and reliable than the other one. Yet another thing is that the syslog protocol is not meant to be persistent and error resistant. The application isn't guaranteed that the daemon will be running all the time, that the connection won't break, that the message will be written anywhere at all. That's why any syslog transaction should be open/write/close and not open/write/write/write/.../close. An application SHOULD NOT depend on the assumption that syslogd will run all the time. marek
Attachment:
pgpGaccw4qIQC.pgp
Description: PGP signature