zookeeper-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Patrick Hunt <ph...@apache.org>
Subject Re: monitoring cluster health via zk mbean data
Date Fri, 25 Feb 2011 18:10:02 GMT
I've always considered that code (mbean naming) to be a mess, it
really wasn't done very well. It works fine for jconsole, but has
issues as you describe. If you have any thoughts please do discuss
them here and contribute some changes, I'd love to address this.
Prolly we'd need to ensure b/w compatibility, but mapping the
existing/new beans into a new namespace that facilitates what you are
asking for would be great!

Do you have to use JMX? Perhaps you could use the 4letter words instead?

Andrei might also have some insight, as he worked on the monitoring
support we currently have in contrib. (cacti/ganglia/...)


On Thu, Feb 24, 2011 at 11:56 AM, Andrew Ebaugh <aebaugh@real.com> wrote:
> We make extensive use of Zabbix for monitoring our systems, and zookeeper is being used
in conjunction with cassandra and tomcat webapps. These other systems are relatively easy
to stamp out generic monitoring templates for.
> Recently we've tried to extend the monitoring to Zookeeper's JMX mbeans, but have found
this difficult considering mbeans are all dynamically named, depending on standalone vs. cluster,
 and member to member within a cluster, and which member is elected leader.
> Wondering what others are doing for monitoring this data (with any framework). For Zabbix
we're looking at tailoring templates to individual members. Another approach might be to re-export
the dynamic-named mbeans with a common naming.
> The main thing we want to monitor is whether the service is alive, request latency, number
of connections, which node is leader, etc.

View raw message