On 29.10.2012, at 17.23, Jim Knuth wrote:
I have here a problem with managesieve. With the login about webmail (roundcube) comes here in the log:
You can always easily reproduce this? Can you get the network traffic logs between Roundcube and Dovecot and reproduce it by sending those same commands manually?
--snip dovecot: managesieve-login: Fatal: pool_system_realloc(4294967296): Out of memory dovecot: managesieve-login: Fatal: master: service(managesieve-login): child 10157 returned error 83 (Out of memory (service managesieve-login { vsz_limit=1024 MB }, you may need to increase it))
Looks like there's a bug somewhere.. Doesn't it log a "raw backtrace"?
--snap I've increased of 2048M and the same above. then with 4096 M happens the following
--snip dovecot: managesieve-login: Panic: epoll_ctl(add, 61538840) failed: Bad file descriptor dovecot: managesieve-login: Fatal: master: service(managesieve-login): child 9777 killed with signal 6 (core dumps disabled) --snap
Probably related to the first error. It would be helpful to get gdb backtraces from both of them, although from the first one you couldn't without patching + recompiling Dovecot. But from the second one I think you can get a core dump with:
service managesieve-login { executable = managesieve-login -D }
Then you can do something like:
gdb /usr/lib/dovecot/managesieve-login /var/run/dovecot/login/core bt full