[Building Sakai] tomcat6/7

[Adams, David]

On Wed, May 9, 2012 at 7:14 PM, John Bush <john.bush at rsmart.com> wrote:
> Is anyone running tomcat 6 or 7 in production ?  And if so has anyone
> seen any performance improvement or other compelling reasons to move ?

Virginia Tech just upgraded to 2.8.1 on May 12, and we decided to move to Tomcat 7 at that time. Other than a change to catalina.properties, a fix to dependencies on the DAV project, a couple of libraries that had to be relocated, and a number of (very easy-to-fix) JSP syntax problems triggered by the JSP 2.2 spec, we had no problems. There is one other new error we're seeing related to the site info tool trying to send data to the client after the response object has already been closed after users edit site information. The users don't see the error, and it doesn't seem to have any negative affect on data consistency or server stability. I believe we were seeing a warning for this before, but as of our upgrade, whether it's due to 2.8.1 or Tomcat 7, those errors are now triggering the mailed bug reports, and so are showing up in our bug report stats even though the users never see the bug report page. I plan to write up a summary of our changes and findings to share with this list once we feel like we've shaken out most of the issues.

As for benefits, the ones I've noticed are mostly minor. It's possible to define thread pools independently of connectors and then share thread pools across connectors, which could lead to some efficiencies. It's possible to give the listener threads reasonable names, which is nice for log correlation and general sensibility--so now instead of http%2F-localhost-127.0.0.1-Processor123, now the thread names are just "tomcat-123" (or you can set whatever prefix you want). Those are small changes, and maybe they were possible in 5.5. Otherwise, Tomcat 7 feels very similar to Tomcat 5.5. I think what changes there have been to the layout have been towards simplification, and I look forward to the Sakai maven plugin taking advantage of the new setup for deployment.

Performance-wise, it's hard to tell at this point, since we upgraded at the end of the semester. We've never had any real issues with Tomcat itself as a performance bottleneck.

As for compelling reasons to move, the fact that Tomcat 5.5 is hitting end of life in October is compelling enough for me. The move was a lot easier than I had feared, and my general philosophy is that it's better to embrace moves like this that are inevitable sooner rather than later.

David Adams
Director, Systems Integration and Support
Virginia Tech Learning Technologies