17 Oct 2012
What’s going on with Alfresco clustering?
I am way past due getting this blog post out. We’ve had multiple discussions on the topic in #alfresco on IRC over the last several weeks, but I want to make sure those of you who aren’t yet hanging out in IRC regularly are aware of some changes related to Alfresco clustering that have already happened with the 4.2.a release. In general, I would prefer to share this stuff way ahead of any release in which it takes effect, but that didn’t happen this time and I’m sorry about that. I promise to try to do better!
With that out of the way, let me shed some light on some recent changes that may affect some of you running Community Edition…
Until recently, clustering in Alfresco has been implemented using a combination of three main technologies: Ehcache, Hazelcast, and JGroups. If you’ve looked at it lately, you may have noticed that JGroups has been removed from the repository source code. What’s going on is that we are consolidating our clustering implementation on Hazelcast because it can handle everything we need to make clustering work. (On a side note, I believe this is one of those improvements that we’ve made to our on-premise software as a result of lessons learned running our own large-scale Alfresco implementation in the cloud).
So that explains what’s going on with clustering in the Enterprise Edition. But if you looked closely at the Community Edition source code, you may have noticed that Hazelcast is no longer included at all. In fact, all clustering related code (org.alfresco.repo.cluster.*) has been removed, including configuration files like cluster-context.xml and hazelcast/* and some changes to existing Spring configuration.
Now, at this point, some of you are probably thinking, “There was clustering code in Community Edition? I didn’t think clustering was supported in Community Edition.” You are correct. Clustering has never been supported in Community Edition. Community Edition is not commercially supported by Alfresco at all. But a lot of the pieces you need to make clustering work in Community Edition have been available until now, and there may be people out there who chose to get it working themselves rather than pay for an Enterprise subscription that includes support for clustering. These recent changes make it much harder to do that.
What I want to make sure everyone is clear on is that the removal of the ability to cluster Community Edition does not represent a shift in our philosophy on what should be in Community Edition versus what should be in Enterprise Edition. The principles John Newton outlined in his blog post, “Building a stronger open source product” back in 2009 still apply today. In short, functionality that supports large-scale rollouts (like clustering) or that depend on paid “Enterprise” software (like Oracle and WebSphere) should be Enterprise-only while everything else should be available to the community.
So rather than a change in philosophy, the removal of the clustering code from Community Edition simply implements the existing philosophy more explicitly. To be blunt, if your implementation is critical enough to require the 7×24 up-time an active-active, multi-node cluster provides, you should be able to justify an Enterprise Edition subscription. If you have high availability requirements but your rollout is relatively small or cost is an issue, perhaps Alfresco in the Cloud will be a fit.
Of the estimated 100,000+ installations of Community Edition currently up-and-running, my hunch is that this change affects only an infinitesimal fraction. But if you have feedback on this change, please do let your voice be heard, either here or by sending me email directly at jeff dot potts at alfresco dot com.