Emmanuel Cecchet a écrit :
Sylvain,
I wonder how to correctly use the WaitForCompletion settings. We're
running a dual controller setup with the following :
<LoadBalancer>
<RAIDb-1>
<WaitForCompletion policy="all"/>
<RAIDb-1-LeastPendingRequestsFirst/>
</RAIDb-1>
</LoadBalancer>
We understood from previous posts that this was the only supported
algorythm. But, with this setup, when I crash a controller (kill -9
or power down), I have *all* of my requests locked. Is this the
expected behavior ?
WaitForCompletion is only for the query completion locally to a
controller but has no effect on cross-controller completion.
The fact that your queries are hanging on one controller is due to
your group communication settings that do not properly detect the
other controller failure. The controller will wait until the group
communication either reports a timeout or a failure. Otherwise your
settings are correct.
OK, it's clear this way. I was not sure of that.
So I have two problems with group communication : it fails under load
(see other mail) and it fails on crash. I'll try to make it work under
load before anything else, that could solve this issue also !
Thanks !
_______________________________________________
Sequoia mailing list
[email protected]
https://forge.continuent.org/mailman/listinfo/sequoia
