On 08/27/2014 02:20 PM, Vasil Valchev wrote:
Hello Steve,
If I just increase token_retransmits_before_loss, without increasing
token, won't it just send more tokens during the same time? For
example 8 in 30s instead of 4?
The last few times the network interruption wasn't longer than a
minute, last time the cluster was even going to reform, but the
fencing was already initiated by fenced.
I want to allow a bit more time in which the nodes can resume
communication and though increasing token timeout should do it. Do you
mean to also increase the token_retransmits_before_loss proportionally?
Vasil,
I am not sure why you would have network disruption for 4 lost tokens,
but transmitting 8 gives better chance they will reach. UDP (the
transport used) can lose those retransmitted tokens. Increasing the
token timer will allow more time for whatever action your doing on the
network that takes it out of service to repair.
Regards,
-steve
BR,
Vasil
On Thu, Aug 28, 2014 at 12:03 AM, Steven Dake <[email protected]
<mailto:[email protected]>> wrote:
On 08/26/2014 07:17 AM, Vasil Valchev wrote:
Hello all,
I have a RHEL 5 (openais) cluster with intermittent issues on the
heartbeat network, and was thinking to increase the totem token
value to 90s (currently is 30s).
Are there any negative effects from this change, apart from the
cluster taking longer to detect a node is failed - can this cause
data corruption for example or something like that?
BR,
Vasil Valchev
Vasil,
I doubt going from 30s to 90s would make a difference with
healthchecking performed. You may be better off increasing
token_retransmits_before_loss_const. Also make sure your running
the latest z stream.
Regards,
-steve
_______________________________________________
Openais mailing list
[email protected]
<mailto:[email protected]>
https://lists.linuxfoundation.org/mailman/listinfo/openais
_______________________________________________
Openais mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/openais
