Doing this in the "monitor" event is unsafe because it causes the node
health status to flip-flop. At the moment when a node goes unhealthy
it is failed out, IPs are released and the monitor event handles the
reconfigure, returning 0 even though the service failure is
unresolved.
This change was made in the master branch a long time ago.
Signed-off-by: Martin Schwenke <martin@meltin.net>
;;
monitor)
- if ctdb_service_needs_reconfigure ; then
- ctdb_service_reconfigure
- exit 0
- fi
-
update_tickles 2049
# check that statd responds to rpc requests
;;
ipreallocated)
+ if ctdb_service_needs_reconfigure ; then
+ ctdb_service_reconfigure
+ exit 0
+ fi
+
# if the ips have been reallocated, we must restart the lockmanager
# across all nodes and ping all statd listeners
[ -x $CTDB_BASE/statd-callout ] && {