[Building Sakai] Sakai 2.9.2 and Kernel 1.3.2 problems during load testing

[Charles Severance]

On Aug 6, 2013, at 9:55 AM, "Kusnetz, Jeremy" <JKusnetz at APUS.EDU> wrote:

> The graphs are a little misleading.  We upgraded to 2.9.x on an off week so we had lower usage anyway.
>  
> Yesterday was a semester start for us, we logged 62,088 site.upd events for the day.  Our last semester start last month which has very similar number of users and usage we saw 8,452 site.upd events.  So a bit more than 7x the number of site.upd events in 2.9.x.

Thanks - that makes more sense.   

With the new iframe xframe processing we should expect more site.upd events.   But not a massive number because the results are cached for two hours by default and it is a value you can set.   So for every active web content tool,during a day - you can expect one *additional* site.upd event caused by a view every two hours.

Now because I missed a detail (not exactly a bug :) ) - A default-configured "Home" tool is running this xframe check when it should not be running the check.  Everything works and the Home tool caches its xframe results properly for two hours - but for a Home tool with a default configuration it should not be doing an xframe check at all.

So when I fix

https://jira.sakaiproject.org/browse/SAK-23841

It should eliminate (a) all the site.upd events from unconfigured Home tools and eliminated site.upd events for web content tools from the same host as Sakai is running on.   So they should drop - depending on whether you are running a load test (all home tools unconfigured) or real user load with lots of home tools reconfigured - the expected drop in site.upd events can range anywhere from eliminating 3/4 of the new events to about 1/4 of the new events.   The events will not go away after SAK-23841 - they will will not be so many un-needed events.

And you canfurther reduce the number of events by setting the cache time to be something like 6 hours.   How often does a site change its xframe policy ?  We could even decude that the right default should be six hours with a little more experience.   I would like ot fist fix SAK-23841 and see the effect that has both on load test scenarios (which I expect to be very significant) and normal usage scenarios.

Thanks for the investigation on this.   It is good to figure out the subtle tweaks needed to this feature.  The feature is increasingly important for end-user usability - but we need to get it right.

I expect to have a patch/fix by the end of the week. 

Make sense?

/Chuck
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://collab.sakaiproject.org/pipermail/sakai-dev/attachments/20130807/b68ff36e/attachment.html