Hi all, I am running dovecot on a linux server (FC3) in conjunction with Sendmail. My configuration is quite vanilla and I am having the following problem: occasionally (once every 3-10 days) the dovecot server will become unresponsive and attempting to access mail via pop3 will produce an error message on the client side. As best as I can tell, the dovecot process dies because if I restart it with /sbin/service dovecot restart I get [failed] on the stop and [ok] on the start. Now, here's the interesting part: within 5 minutes of this happening, the whole machine becomes completely unresponsive and requires a power-cycle to restart it. In between dovecot dying and the server going down, the other applications (apache, mysql) continue to run. Naturally, there are no messages in the syslog or the maillog and it is set to log all messages of info or higher. Any suggestions on how to attack this would be greatly appreciated. Best, -cwk.
Colin Kingsbury wrote:> Hi all, > > I am running dovecot on a linux server (FC3) in conjunction with > Sendmail. My configuration is quite vanilla and I am having the > following problem: occasionally (once every 3-10 days) the dovecot > server will become unresponsive and attempting to access mail via pop3 > will produce an error message on the client side.Any chance you could mention which version of Dovecot?> As best as I can tell, the dovecot process dies because if I restart > it with /sbin/service dovecot restart I get [failed] on the stop and > [ok] on the start.Does it show up in 'ps aux'? Can you see dovecot, dovecot-auti, imap-login or pop3-login in the list?> Now, here's the interesting part: within 5 minutes of this happening, > the whole machine becomes completely unresponsive and requires a > power-cycle to restart it. In between dovecot dying and the server > going down, the other applications (apache, mysql) continue to run.My gut feeling is that Dovecot is dieing as a result of something much more serious on your system - perhaps a resource starvation like disk or RAM. -- Curtis Maloney
We have experienced similar problems with version 0.99.11 that comes with Redhat Enterprise Linux 4. If I'm not wrong, it appears to be related to the number of file descripters available to the dovecot user. To abate the situation, we stop the service, kill any remaining dovecot processes (some hang around), and then restart dovecot. Sometimes it appears that the limit of connection processes isn't honoured (as defined in the conf file) so the number can spiral out of control. I hope this helps in the absense of a workaround. Let me know if there is an obvious configuration setting to remedy this.