On Thu, Mar 21, 2013 at 11:46 AM, Leslie Carr <lcarr(a)wikimedia.org> wrote:
https://bugzilla.wikimedia.org/show_bug.cgi?id=46428
If one of the php-knowledgable peeps can take a look at this (sadly my
php-foo is quite weak).
Hmm... ok this explains why I couldn't find Wikipedia Zero-related
memcache entries when testing from fenari last night. :(
It sounds like maintenance scripts could have a serious split-brain
problem if they're being run in pmtpa while the web servers are in
eqiad, if each cluster uses its own local memcached servers.
Either updates need to be proxied to both clusters (race conditions?)
or it should only be possible to run maintenance scripts in the same
cluster that's running the web servers. The former sounds scary and
latency-intense, while the latter requires more configuration on the
eqiad bastion hosts so MW scripts can be run. :(
-- brion