Skip to Content.
Sympa Menu

grouper-users - Re: [grouper-users] HA-ing grouper ...

Subject: Grouper Users - Open Discussion List

List archive

Re: [grouper-users] HA-ing grouper ...


Chronological Thread 
  • From: Colin Hudler <>
  • To: "" <>
  • Cc: Adrian Lee <>
  • Subject: Re: [grouper-users] HA-ing grouper ...
  • Date: Tue, 11 Oct 2011 11:41:30 -0500

Our deployment aims to reduce chance of complete failure due to failure of a single component. You could call this HA. Shared-nothing pattern makes this easy. We did not believe that different style of cluster (jvm) would be best for us.

1) Two webservice nodes are round-robin load-balanced behind a single IP. This is easy because the webservice is stateless.
2) Two UI nodes are load-balanced and very sticky based on the cookie.
3) A single "automations" server handles sundry cronjobs, scripts, GSH, and quartz jobs. This one is not used by the users and has no buddy. It is rather important, but jobs can be moved easily to another node, if needed.
4) The database is Oracle with some Dataguard features (HA). This uses a standby database that will be enabled transparently if the primary fails. This works pretty well with grouper in my experience. There's about 30 seconds downtime (db connection error), but it recovered automatically.

This design has evolved over time to be what it is now. DB availability is key, obviously. We also worked to integrate it with our release-management and configuration management. All of the grouper nodes share configuration, libraries, and have identical on-disk structure. The only difference between the nodes is the servlets, services, and scripts that are running. This is determined once at deploy-time but can be altered at any other time.


On 10/10/11 10:56 PM, Adrian Lee wrote:
Hi,

We're currently looking at piloting Grouper and the question has been raised
about the possibility of a high availability (HA) deployment. The database
and sources (LDAP in this case) are both HA. Assuming a load-balancer is
front-ending Grouper, that just leaves tomcat. Tomcat has clustering recipes
- anyone tried/thought-about-trying such a configuration for Grouper? Any
thoughts on why this might be a bad idea? Would clustering actually be
necessary?

Thanks,
Adrian
-----------------
Adrian Lee
Information Technology Services
The University of Queensland




Archive powered by MHonArc 2.6.16.

Top of Page