On 10 May 2015, at 06:19, Reuben Farrelly reuben-dovecot@reub.net wrote:
I'm now seeing these crashes in the indexing code:
May 10 12:22:32 tornado.reub.net dovecot: master: Dovecot v2.2.17.rc1 (3903badc4ee0+) starting up for imap, lmtp, sieve ... May 10 13:07:50 tornado.reub.net dovecot: indexer: Error: Indexer worker disconnected, discarding 1 requests for reuben May 10 13:07:50 tornado.reub.net dovecot: indexer-worker(reuben): Fatal: master: service(indexer-worker): child 18532 killed with signal 11 (core not dumped)
Every time a delivery happens or mail is moved the indexer crashes out. It doesn't seem to have any visible end user impact though.
What's your current doveconf -n output? I did a quick test with fts-lucene and it was working.
How do I go about debugging this further? How do I go about getting a backtrace on what is essentially a child process that doesn't produce a core? (Note that my installation normally does produce cores when it crashes, so there's something different about these indexer processes..)
One thing that often works is:
mkdir /var/core; chmod 0777 /var/core echo "/var/core/%p" > /proc/sys/kernel/core_pattern
If that doesn't help, you can also try: sysctl -w fs.suid_dumpable=2 (and disable afterwards)
Also if you can get indexer-worker to start without crashing (new mail delivery works?) you can attach gdb into it and wait for it to crash (gdb -p <indexer-worker pid>, cont).