Hi devs,
As you know our goal is to use
myxwiki.org as a real life test platform to validate
releases of XWiki.
Current Situation
==================
However this is currently not working very well for 2 reasons:
1) We’re always lagging behind on the version installed on
myxwiki.org. Right now it’s
7.1.2 and our last released version is 7.3M2. Thus if we notice a problem on
myxwiki.org,
it’ll be fixed only in much later versions and
myxwiki.org is not playing its role of
helping validate releases before we release final versions.
2) We don’t really monitor the performance of
myxwiki.org.
3) We don’t really analyse issues that can happen on it because we don’t check the logs.
Thus I think we need to find a better process for benefitting from
myxwiki.org.
Question
=========
Are we still interested in benefitting from
myxwiki.org for testing our releases?
If not, then stop reading at this point :)
If we are, then I’m making some proposals below.
Proposal
=========
For 1):
I’d like to propose to add a step in our ReleasePlan template as the last step:
- Check the
myxwiki.org upgrade roster and ping the next person to update
myxwiki.org
So the idea would be to not make the RM do the upgrade since he/she already has a lot to
do to release XWiki but to make him/her responsible for pinging someone to do it. Then we
would take turn to upgrade it (in a similar fashion as we do for releasing XWiki).
Note that I believe this would also make us work on making it simpler to perform XWiki
upgrades and this would benefit our users. We would eat our own dog food basically :)
For 2):
Here are ideas of what we could monitor and receive alerts when they go beyond a given
threshold:
- the average response times users get on it,
- when a specific requests takes more than N seconds (this would also allow us to find
wiki pages written by our users and which take too much CPU thus making the farm slower
than it should be),
- its uptime
- the memory used
For 3):
I think we could set up some elastic search/kibana solution as we had set up at some point
(this makes it nice to browse and search for logs) and then send automatic mails to the
devs list or IRC when exception happen. This would have the nice benefit of making us work
on fixing the code and not generating exceptions when we have only warnings that don’t
impact the stability of the platform.
WDYT?
If we agree, then we’ll need to discuss how to setup 2 and 3.
Thanks
-Vincent