Sean Brandt - 15/Dec/06 02:27 PM log file containing stack trace

jason - 16/Dec/06 10:48 AM Hi Sean, Thanks for the foresight of taking a thread dump for us. Interestingly, I see in both cases the Pulse is trying to render an RSS feed. I wonder if potentially this is the source of the performance issue with Pulse. In particular, some feed readers will hit Pulse very frequently, and in some cases not provide accurate headers about when they last checked, which gives Pulse extra work to do. We have done some work in 1.1.23 to address these issues with caching on the Pulse side. This should also be happening in Pulse 1.2.9, but the caching is a little different on 1.2, so perhaps it is not working effectively. We will investigate the caching, and in the mean time, I wonder if you could try disabling RSS feeds (in the administration section, general settings) and seeing if that improves things? This would at least validate the theory. Also, is this a new thing you have just started seeing? If so, have there been any other recent changes (new projects or anything) that may be related to this problem? When you say a large configuration, how many projects/users/builds are we talking about? Thanks in advance.

Sean Brandt - 16/Dec/06 04:19 PM Turning off rss definately speeds things up. Config: 4 projects, some with MANY tests ( 17k + ) Builds are frequent during the working hours, and currently number in the 500+ per project Users are relatively few ( < 10 most are anonymous viewing of state )

jason - 18/Dec/06 01:42 PM Sean, Thanks for the further details. It appears that it is likely a problem with the RSS caching implementation on 1.2, which we are investigating in more detail now. The size of your configuration is certainly manageable, the only large number being tests/build (good work ;) ).

Daniel Ostermeier - 19/Dec/06 12:25 AM A fix for the rss performance issues is already in place for the latest 1.1.x releases, so removing this issue from 1.1.x.

Daniel Ostermeier - 19/Dec/06 03:20 PM Reviewing of the rss and caching mechanims have highlighted a couple of issues. a) the cache invalidation time being used in 1.2 was too low. It was set to 2 mins, which, given the performance problems that have been reported, might well have been so short that the data was invalidated before it had been finalised. b) part of the query used to render the build results for the rss feed was found to be excessively slow. This has been fixed. The result is that performance of the rss feeds should be significantely improved in 1.2.10. Since the systems performance improved with rss disabled, I would expect that these mesures would visibly improve the performance of 1.2.x. Of course, this has only been tested under artificial conditions, so if the performance is still sluggish we will reopen the investigation.

Daniel Ostermeier - 20/Dec/06 02:02 PM Hi Sean, I just wanted to let you know that pulse 1.2.10 is now available for download and contains the fixes mentioned above. Cheers, -Daniel