[infinispan-dev] [infinispan-internal] Unstable Cluster

[Dan Berindei]

On Mon, Mar 4, 2013 at 10:28 AM, Bela Ban <bban at redhat.com> wrote:

> Another node: in general, would it make sense to use shorter names ?
> E.g. instead of
>
> ** New view: [jdg-perf-01-60164|9] [jdg-perf-01-60164,
> | jdg-perf-01-24167, jdg-perf-01-53841, jdg-perf-01-39558,
> | jdg-perf-01-8977, jdg-perf-01-49115, jdg-perf-01-24774,
> | jdg-perf-01-5758, jdg-perf-01-37137, jdg-perf-01-45330,
> | jdg-perf-01-24793, jdg-perf-01-35602, jdg-perf-02-7751,
> | jdg-perf-02-37056, jdg-perf-02-50381, jdg-perf-02-53449,
> | jdg-perf-02-64954, jdg-perf-02-34066, jdg-perf-02-61515,
> | jdg-perf-02-65045 ...]
>
>
> we could have
> ** New view: [1|9] [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15,
> 16, 17, 18, 19, 20, ...]
>
> This makes reading logs *much* easier than having those long names.
>
>
Yes and no... I sometimes find it useful to have a somehow longer name, as
searching/filtering for a node in the log is pretty much impossible with a
name like "1".

I also think we need the random number at the end so that we can debug
problems with node restarts. In JDG/AS7 they don't add a random number, and
it was very difficult to see what was happening when a node name appeared
twice in the consistent hash. But we could make the random number shorter.



> If we wanted the host name to be part of a cluster name, we could use
> the alphabet, e.g. A=jdk-perf-01, B=jdg-perf-02:
>
> ** New view: [A1|9][A1, A2, A3, B4, B6, C2, C3, ...]
>
> This is of course tied to a given host naming scheme. But oftentimes,
> host names include numbers, so perhaps we could use a regexp to extract
> that number and use it as a prefix to the name, e.g.
> cluster-01 first instance: 1-1
> cluster-02 2nd instance: 1-2
> etc.
>
> Thoughts ?
>
>
Are you thinking of an automatic way of assigning a letter+digit
combination to a node on startup? We also use the node name for some other
stuff (e.g. thread names), so I'm not sure if it's feasible to wait until
we have connected to the JGroups cluster to set the node name dynamically.

For RadarGun we could use a static system where we configure a node name
for each slave and then RadarGun passes the node name to Infinispan via a
system property. No Infinispan changes required (except perhaps making the
random number in the node name optional.)


Cheers
Dan



>
> On 3/4/13 8:43 AM, Radim Vansa wrote:
> > Just a small sidenote: if you want to print full view (not just first 20
> nodes and ellipsis after that), use -Dmax.list.print_size=cluster_size
> >
> > Radim
> >
> > ----- Original Message -----
> > | From: "Shane Johnson" <shjohnso at redhat.com>
> > | To: infinispan-internal at redhat.com
> > | Sent: Friday, March 1, 2013 5:52:17 PM
> > | Subject: Re: [infinispan-internal] Unstable Cluster
> > |
> > | The JGroups cluster appeared stable, at first. However, I did notice
> > | that the logs looked a little bit different on one machine /
> > | instance. I'm not sure if that means anything or not.
> > |
> > | Machine 1 / Instance 1-12
> > |
> > | ** New view: [jdg-perf-01-60164|9] [jdg-perf-01-60164,
> > | jdg-perf-01-24167, jdg-perf-01-53841, jdg-perf-01-39558,
> > | jdg-perf-01-8977, jdg-perf-01-49115, jdg-perf-01-24774,
> > | jdg-perf-01-5758, jdg-perf-01-37137, jdg-perf-01-45330,
> > | jdg-perf-01-24793, jdg-perf-01-35602, jdg-perf-02-7751,
> > | jdg-perf-02-37056, jdg-perf-02-50381, jdg-perf-02-53449,
> > | jdg-perf-02-64954, jdg-perf-02-34066, jdg-perf-02-61515,
> > | jdg-perf-02-65045 ...]
>
>
> --
> Bela Ban, JGroups lead (http://www.jgroups.org)
> _______________________________________________
> infinispan-dev mailing list
> infinispan-dev at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/infinispan-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.jboss.org/pipermail/infinispan-dev/attachments/20130304/ac20e4b3/attachment.html