On 7/7/06, Dejan Muhamedagic <[EMAIL PROTECTED]> wrote:
Hello,

Recently, I was busy with the apache RA and there are some results
in bugzilla, under bug 1357 and this is one comment from there:

    Sometimes, I guess that that depends on how busy the node is,
    the first monitor operation fails. It is the operation
    scheduled immediately after the start operation has been
    confirmed. Hence the patch that makes start_apache wait until
    the server is actually functional.

The question is simple: Should the RA, after it has done the start
operation, wait until the resource is really active, essentially
by doing the monitor operation before reporting success?

There are exceptions to this, but in principle "start" should only
return after the resource is completely active.

The reason is that the next resource you start may depend on it.
Returning "i'm done" too early *may* end up causing the next resource
start to fail because its prerequisite was not available yet (remember
a failed start will prevent the resource from running on that node
until you clean it up).

So you're much better off waiting for start to *really* complete and
setting the timeouts accordingly.
_______________________________________________________
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