Problem

CFT Multi-node

NFSV4 NetApp

The prerequisites about the NFS share from the CFT documentation does not specify anything about the recover_lost_locks parameter.

Current settings:

[root@Myserver1 ~]# cat /sys/module/nfs/parameters/recover_lost_locks

N

Resolution

Locks are mainly used to protect multiple read/write of the same file to avoid corruption.

In CFT multi-node, mainly the main COM file and the uconf file are concerned during production.

In multinode, the locks are also used to test if CFTMAIN nodes are alive (files cft**.pid) and to ensure some features are handled by one single node (dscan cron with **.lck).

To counter the possible lock lost due to NFS lease time, the copmnwd process is a watchdog that requires the copilot node manager to send a message to the watchdog periodically.
That period is defined with uconf:copilot.node_manager.watchperiod.
If the watchdog does not receive the message after 2 periods, it will kill all CFTMAIN nodes that are present on the host that would be about to lose their locks.

That's why we recommend to set uconf:copilot.node_manager.watchperiod to less than half of the NFS lease time.

We also recommend not to use "recover_lost_locks". It should stay disabled.