It is possible that a Redis server can die due to memory pressure. The OOM kills it, it segfaults or similar ill fate for it could be just around the corner. Puppet does not take too kind to services that can't figure out they need to restart themselves.

We need a way to see if the PIDFILE no longer contains a fresh process ID. The following is a patch that implements it:

# diff -u rc-redis-orig rc-redis-stale-pid-remove
--- rc-redis-orig 2012-04-21 20:34:20.022877178 +0000
+++ rc-redis-stale-pid-remove 2012-04-21 20:55:42.516233398 +0000
@@ -18,19 +18,20 @@
[ -d $WORKDIR ] || mkdir $WORKDIR

if [ -f $PIDFILE ]; then
+ PID=$(cat $PIDFILE)
+ [ -d /proc/${PID} ] && rm_daemon $daemon_name
+ fi
+
+ $EXEC $CONF >/dev/null
+ if [ $? -gt 0 ]; then
stat_fail
exit 1
else
- $EXEC $CONF >/dev/null
- if [ $? -gt 0 ]; then
- stat_fail
- exit 1
- else
- add_daemon $daemon_name
- stat_done
- fi
+ add_daemon $daemon_name
+ stat_done
fi
- ;;
+
+ ;;

stop)
stat_busy "Stopping $daemon_name"



Additional info:
Redis 2.4.11 is out :)