As above. It'll remove numerous config hacks and things we have.
Some of the PHP config will need updating, as will the Apache config... etc
Version: unspecified
Severity: enhancement
As above. It'll remove numerous config hacks and things we have.
Some of the PHP config will need updating, as will the Apache config... etc
Version: unspecified
Severity: enhancement
No, isn't.
bug 44259 was that testwiki was completely unuseable due to it trying to use a mysql "master" in PMTPA that was readonly. Tim fixed this by adding in the EQIAD master which isn't readonly.
srv193 doesn't serve testwiki from NFS now.
I thought the original intention was completely to remove testwiki/srv193 as an exception, so it would work from the normal pool of apaches, not one specific hosts in PMTPA?
That's the intent of this bug and I believe has been the plan all along. But I think we'd need to communicate this out effectively before hand to all of Engineering with a transition window. I know some where using it in special ways before. I'm confirming with E2/E3 right now.
(In reply to comment #5 by Greg)
I'm confirming with E2/E3 right now.
Greg: Any news?
The switch to a different server was done a long time ago - now it's at mw1017. The other part is icky - it is really that much fun to risk breaking whole cluster to test something? We at mobile took advantage of testwiki being separated from the rest of the cluster countless times, and it was really convenient.
Max: Since you all are riding the train now, are you still using test in that way? (cc'ing Michelle for her thoughts as well)
So, after thinking about this (and talking with Chad): are you sure it's doing what you think it's doing?
testwiki no longer is served from NFS, so you actually have to scap to update the code on testwiki.
That is intentional as testing should happen either in development or while it's on the Beta Cluster. Then on the production cluster when it hits all of the test wikis.
While we removed the NFS hack, we do still need to remove the 'serve from one specific machine' hack.
As the matter of fact, staged deployment is one of requirements for new deployment system being developed. Also, some things can't be really tested anywhere else but production like DB config, etc.
mw1017 is a part of the special group of app servers that serve requests with the X-Wikimedia-Debug header set. See also https://wikitech.wikimedia.org/wiki/Debugging_in_production
This was actually done in rOPUP185a40c06094: Undo special-casing of testwiki in Varnish.