Le lun 7 août 2006 22:19, Krzysztof Krzyzaniak a écrit : > Pierre Habouzit wrote: > > Package: lighttpd > > Version: 1.4.11-7 > > Severity: grave > > Justification: renders package unusable > > > > > > I use lighttpd on many servers, and sometimes, because of > > logrotate, the server stops, but never starts again (presumably > > because the sleep 5 is not enough, logs shows that it took 5 > > seconds to stop: > > > > 2006-07-27 06:25:20: (server.c.1216) [note] graceful shutdown > > started 2006-07-27 06:25:24: (log.c.135) server stopped > > > > (1) I'm surprised that lighttpd is not able to understand a kill > > -HUP to reload. > > It's because lighty dropping privileges after binding to port 80, > there is discussion [1] and preliminary, experimental patch.
I understand that but ... > > (2) even if it doesn't, it should be possible to wait until it's > > really stopped before starting it again, instead of a crude > > sleep. > > > > given that the logrotate fires the stop around 06:25, it causes > > serious downtime, hence the severity. it should still able to achieve that: having a way to tell *when* lighty is properly shut down. because that's the problem here: you try to restart lighty *before* it's actually shut down, and the second instance fails because the first still listen on the :80. IMHO, that could be done like that: at shutdown time, the master lighty process just close ALL the binded sockets, so that the new instance can be started immediately. a clever way would be that some tool connects itself to an unix lighty control socket, asks for removal, and that lighty says when it shutted all the sockets down just before closing that socket. when that client retunrs, you can safely start a new lighty. that's simple and safe. -- ·O· Pierre Habouzit ··O [EMAIL PROTECTED] OOO http://www.madism.org
pgppE0r9X9uXN.pgp
Description: PGP signature
