Confluence Cluster disadvantages or how to handle cluster panics?

Question

Hi,

We are using a clustered installation of Confluence (2 nodes) but see more disadvantages than advantages of it.

- it isn't possible to configure it in a master/slave configuration, i.e. when one confluence receives a heartbeat, when cluster panic occurs, both nodes disconnect and stop working - they need to be restarted manually. These cluster panics can occur very often so the availability of Confluence is not the highest.

Questions:

- how do you handle cluster panics, failovers? do you have some linux scripts to detect them and restart the instances? Are there other better ways to handle them?

- why shouldn't we move from a clustered standalone installation with 2 nodes to a 1 standalone non-clustered installation?

Adam Laskowski · Accepted Answer

Cluster panics have a number of root causes, so narrowing down which particular scenario you're encountering will determine what you can do to prevent or prepare for the fallout from failures.  In most cases, the panic occurs when one node becomes unresponsive and the cluster safety check can't be completed properly.

The major benefit for clustering your instance is balancing the load for requests against your instance across more than one server.  If your instance as a single node would slow considerably during peak hours of the day, splitting the load across more hardware usually reluts in better performance.  There are some other considerations that you should look over, as well, before making the decision to go to a single node environment, but it's normally possible to performance tune a single node to perform well under load such that a clustered instance might not be necessary.

Try running a single node in production for one day and monitor performance during your peak hours.  If the performance hit is quite large (an additional 5-10+ seconds load time), you might still want to keep clustering.  If it's a minimal change to performance, you should be able to make a few performance modifications and run a single node long-term.