Ben,

A few questions:

During exportfs_start, you do

* Try to export;
* if that fails right off the bat, bail out;
* otherwise, do some really complicated check whether it really worked.

Why don't you:

* Try to export;
* if that fails right off the bat, bail out;
* otherwise simply spin on monitor until it succeeds;
* if it doesn't succeed, the CRM will time you out.

That way you could eliminate duplicate code, and greatly simplify your
start op.

Same thing on stop:

* Try to unexport;
* if that fails right off the bat, bail out;
* otherwise simply spin on monitor until it returns $OCF_NOT_RUNNING;
* if that never happens, the CRM will time you out.

Other than that, can you please explain the rationale behind that
"backup" action? It's wildly unorthodox -- normally we embed those bits
into a monitor op. It's also kind of odd to have one RA spawn an
instance of itself in the background. We expect Pacemaker to shut down
_everything_ related to a resource if that resource stops or the node
goes into standby. How would that work for a process fired off into the
background?

Finally, exportfs_validate doesn't check on all required parameters. You
want to fix that too.

Cheers,
Florian

Attachment: signature.asc
Description: OpenPGP digital signature

_______________________________________________________
Linux-HA-Dev: [email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
Home Page: http://linux-ha.org/

Reply via email to