[Linux-cluster] nodes boot synchronization sensitivity

Stepan Kadlec skadlec at gk-software.com
Wed Nov 19 09:37:10 UTC 2008


anyway, still don't understand:

node1 of the two_nodes cluster boots up and becomes quorate. the other 
node2 is still down, so the fenced on node1 reports:

   Nov 19 10:11:41 node1 fenced[3559]: node2 not a
   cluster member after 6 sec post_join_delay
   Nov 19 10:11:41 node1 fenced[3559]: fencing node "node2"

and fences the node2. than node2 boots up and repeats the same scenario 
- I can't understand, why at this point the node2 can't just join the 
running cluster with node1 and instead of that reports the same "node1 
not a cluster member after 6 sec" and fences it. this oscillates forever.

is this normal behavior?

thanks for advices.
stepan

Stepan Kadlec wrote:
>     oh, I have probably misunderstood the problem - the real cause seems 
> be unsynchronized local clocks on the nodes...
>     bye stepan
> 
> 
> Stepan Kadlec wrote:
>> hello,
>> I have two_node cluster. If I synchronize the boot to the same time, 
>> both nodes join fain and everything works.
>>
>> I am trying to make it less sensitive to boot-time synchronization (to 
>> accept at least two minutes difference) but the nodes never join and 
>> after some time, one node is fenced.
>>
>> I have prolonged the post_join_delay to 120 seconds, but even when 
>> both nodes are trying to join in the nearly same time (~30 sec 
>> difference), they are unsuccessful - the log shows
>>
>>     "not a cluster member after 120 sec post_join_delay"
>>
>> and the other node is fenced.
>>
>> I am running the cluster in following steps:
>>
>> cman_tool -t 120 -w join -n node1 -c cluster
>> groupd
>> fenced
>> dlm_controld
>> gfs_controld
>> fence_tool -w -t 300 -m 20 join
>>
>> how can I make the nodes less sensitive to boot synchronization?
>>
>> thanks for your advices.
>> stepan
>>
>> -- 
>> Linux-cluster mailing list
>> Linux-cluster at redhat.com
>> https://www.redhat.com/mailman/listinfo/linux-cluster
>>
> 

-- 
Eurosoftware s.r.o.
skadlec at gk-software.com
+420 379 307 379
+420 724 554 104




More information about the Linux-cluster mailing list