Re: [jgroups-dev] Message explosion in FD_ALL

Bela Ban <[email protected]>
Newsgroups gmane.comp.java.javagroups.devel,gmane.comp.java.javagroups.general
Message-ID <[email protected]>

Mike Jensen wrote:
> Maybe this is a stupid question....(darn, and I missed stupid question
> day: http://en.wikipedia.org/wiki/Ask_a_Stupid_Question_Day)
>
> But why do you even send an "ARE_YOU_ALIVE" to the node? If your not
> receiving heartbeats from the node, what situation would occur where you
> would receive a response "I_AM_NOT_DEAD"?

The double check is optional, and is done in VERIFY_SUSPECT, so if you 
don't want it, remove VERIFY_CONFIG from your stack.

The reason for its existence is to reduce false suspicions. Say we have 
{A,B,C,D,E}. D suspects E. The double check is done by A. So, in the 
case of different network paths, A can possibly have better connectivity 
to E than D, so A would drop the suspect and not exclude E from the cluster.

Also, if E is temporarily overloaded, D might generate a spurious 
SUSPECT(E). When A double-checks later, the temp spike might be gone.

> I just don't see a case where a reliable heartbeat would get lost,

What do you mean ? The heartbeat is not reliable, it is always under the 
UNICAST and NAKACK protocols, so they can indeed get lost.

> but this query response would keep
> the node alive. Have you seen such a situation in the past?

Yes

> If so, do you think this is something I need to attempt in my tree 
> mesh FD protocol?

Reducing false suspicions is always a good idea, however, it might 
become too costly for a system like treemesh, which is designed for 
large scale. I guess nodes leaving and joining is frequent in your 
system, so a false suspicion making a node leave and later rejoin is 
probably ok.


-- 
Bela Ban
Lead JGroups / Clustering Team
JBoss

------------------------------------------------------------------------------
Start uncovering the many advantages of virtual appliances
and start using them to simplify application deployment and
accelerate your shift to cloud computing.
http://p.sf.net/sfu/novell-sfdev2dev
_______________________________________________
Javagroups-development mailing list
lmpx.com only provides a reader for public news (NNTP) servers. It is not affiliated with the servers or forums shown here and is not responsible for the content of articles, which is written by their respective authors.