I'm experiencing an issue with both Fedora and Centos that I don't
understand. If I send a -TERM signal to the beam/beam.smp process and
RESPAWN_TIMEOUT > 0 the crash handler is being triggered and the process
is started again. This is happening under Fedora and Centos, but not
Amazon Linux. I have varying different Erlang versions, both R14B04 and
R15B. With Centos 6, I've installed in an openvz containter, virtualbox
and on hardware. Same result. This does not happen with Amazon Linux
however. I'm wondering if this might be some erlang issue they've
patched for that's not landed in EPEL? Any feedback on this would be
helpful.
A quick example of behavior I'm experiencing:
$ ./test/javascript/run attachment_paths
Apache CouchDB has started, time to relax.
1..1
ok 1 attachment_paths
All tests passed
Apache CouchDB crashed, restarting in 1 second.
Apache CouchDB has been shutdown.
$ ps -A
15398 pts/2 00:00:00 run
15463 pts/2 00:00:00 run
15464 pts/2 00:00:00 beam.smp
15492 ? 00:00:00 heart
$ cat tmp/run/couchdb/couchdb.pid
15464
$ kill -TERM `cat tmp/run/couchdb/couchdb.pid`
Apache CouchDB crashed, restarting in 1 second.
$ ./utils/run -k
Apache CouchDB crashed, restarting in 1 second.
Apache CouchDB has been killed.
Changing respawn timeout to 0 in ./test/javascript/run, process exits
normally.
Any help here debugging would be appreciated.
Thanks,
Wendall
