Re: [jgroups-dev] Problem with FD_SOCK failure detection

On 8/21/13 8:16 AM, mayank.agarwal wrote:
>
> We are using FD_SOCK for failure detection. Recently lot of cluster hang
> issues coming when new member joins and co-ordinator goes down at same time.
>
> As per my understanding FD_SOCK detects the failure of right neighbor so
> when new member comes up it takes the responsibility of detecting the
> co-ordinator failure.  I think this scenario is error prone in case new
> member is not initialized((FD_SOCK pinger thread is not started)) and
> co-ordinator goes down then it's failure is undetected. New member goes in
> indefinite retry to join the crashed co-ordinator.

I don't think this should be the case: when the new member cannot 
establish its socket to the coordinator, it will throw an exception 
which will lead to the coordinator being suspected and eventually 
excluded by the next-in-line.

Are you running the recommended combo of FD and FD/FD_ALL ? If FD_SOCK 
didn't end up suspecting the coordinator, then FD or FD_ALL certainly 
would !

> We are thinking of changing the logic to detect the neighbor in left. Have
> anyone tried this?

That would certainly be simple; all you'd need to change is 
determinePingDest().

Can you reproduce this scenario ?

> Details:
> JGroups version: 3.3.1.Final
> Using JGroups with Infinispan 5.3.0.Final


-- 
Bela Ban, JGroups lead (http://www.jgroups.org)

------------------------------------------------------------------------------
Introducing Performance Central, a new site from SourceForge and 
AppDynamics. Performance Central is your source for news, insights, 
analysis and resources for efficient Application Performance Management. 
Visit us today!
http://pubads.g.doubleclick.net/gampad/clk?id=48897511&iu=/4140/ostg.clktrk
_______________________________________________
Javagroups-development mailing list