We'd like to stop refining the mediawiki.job.* queue streams into the Hive
event database. This means that the event.mediawiki_job_* tables will be
I don't expect that anyone actually uses them, but if there are objections,
please let us know at https://phabricator.wikimedia.org/T281605.
If no objections, I will stop refining and remove these the week of May 10.
We are going to change the Yarn scheduler in a bit, moving from Fair to
Capacity. More info in
It shouldn't impact ongoing jobs, but there will probably be some
tuning/adjustment to do during the next few days. If you see any weird
behavior please ping us or contact us in T277062.
A gift for your patience: this change will allow users to select hadoop
nodes with GPUs when launching jobs :) (we have 6, running the same GPUs as
Tomorrow EU morning (9:00 CET) I'll upgrade the Debian OS on an-coord1001,
and several services will need to be stopped: Hive, Oozie, Presto (may also
affect Druid, so Turnilo and Superset). The total downtime should be around
one hour (best case scenario).
Please let me know if this impacts your in
https://phabricator.wikimedia.org/T278424 (in case we can find another time
window for the downtime).
Luca (on behalf of the Analytics / Data Engineering team)
we are going to make hue-next.wikimedia.org (freshly updated to 4.9, latest
upstream) the new hue.wikimedia.org. This will mean that:
1) We'll finally remove the last dependency for the Cloudera CDH packages
that we still have on our APT repositories (not good after
hue.wikimedia.org's backend is still running a very old version of Hue with
2) We'll move Hue's user management to be fully automated via CAS, so no
more requests to the Analytics team to create a new user etc.. Every new
user should be able to use Hue straight away without any extra ping (beside
the usual ones for wmf/nda LDAP membership of course).
The main downside is that there are some bugs (github issues already opened
to upstream) that may cause some bad UX experience, see
https://phabricator.wikimedia.org/T264896. Due to 1) we'll need to proceed
asap, but if anybody is interested in following up on those github issues
feel free to :)
If nobody opposes I'll make the switch tomorrow Apr 15th during the EU
morning (there will be a little downtime but hopefully limited to one hour).
For any questions or follow up, please ping me on IRC or add a note in the
[also sent to ops@list earlier]
The current bastion host for eqiad (bast1002.wikimedia.org) is five
years old and being replaced by a new server (bast1003.wikimedia.org).
Please adapt your SSH client configs, I've also updated wmf-laptop-sre
(0.5.1) to use the new server.
bast1002.wikimedia.org will stick around for another week.