Update: data should now be recovered and everything back on track. Latest
data might take a bit of time to catch up since we have just restarted the
replication script on the analytics-slave.
All the details about the outage in
Thanks!
Luca
2018-03-06 13:55 GMT+01:00 Luca Toscano <ltoscano(a)wikimedia.org>rg>:
Hi everybody,
today, while performing maintenance to the Eventlogging Master database,
we ended up in
https://phabricator.wikimedia.org/T188991 (TL;DR: two
hours of data inserted to the slave database and not the master one). We
are working to find a feasible solution to avoid loosing data and getting
out this inconsistent state, so as precautionary measure the Eventlogging
mysql consumers have been stopped.
A couple of notes:
- The Eventlogging machinery is working as expected, except mysql
insertion of course.
- The HDFS data has not been affected by this issue.
Please check the task for more updates, or follow up with the Analytics
team on IRC (#wikimedia-analytics on freenode).
Thanks and sorry for the trouble!
Luca (on behalf of the Analytics team)