[Linux-cluster] post_fail_delay versus deadnode_timeout

[Jeff Stoner]

> deadnode_timeout
> 
> Cons:
> - Not persistent. Need to hack cman init script to make persistent.
> 
> 
> corrections/additions welcome
> 

Thinking out loud....

Since it's executed last in the boot sequence, set it in
/etc/rc.d/rc.local instead of /etc/rc.d/init.d/cman. Updates to RHCS
won't require re-hacking the init scripts.

...but...

Bringing down the cluster services (and unloading the kernel modules)
then bringing it all back up again, without a reboot, means the setting
won't be put back in place.

...so...

Write a wrapper around stopping/starting the cluster services on the
node such that on startup, the parameter gets set in /proc/ explicitly -
use said script for cluster maintenance that doesn't involve a reboot.



--Jeff
SME - UNIX
OpSource Inc.

PGP Key ID 0x6CB364CA