Page MenuHomePhabricator

EventLogging of Echo notifications broken
Closed, ResolvedPublic

Description

According to my graph provided in https://gerrit.wikimedia.org/r/110112
EventLogging of Thanks for both mobile and desktop has been broken since 19th January.


Version: unspecified
Severity: normal
See Also:
https://bugzilla.wikimedia.org/show_bug.cgi?id=60555

Details

Reference
bz60550

Related Objects

Event Timeline

bzimport raised the priority of this task from to Needs Triage.Nov 22 2014, 2:52 AM
bzimport added a project: Notifications.
bzimport set Reference to bz60550.
bzimport added a subscriber: Unknown Object (MLST).

Echo_6081131 doesn't seem to be logging any thanks events.

This may be an issue with Echo and not the Thanks extension...

(In reply to comment #2)

This may be an issue with Echo and not the Thanks extension...

I don't see anything in Thanks that has to do with EventLogging, so moving to Echo based on https://github.com/wikimedia/mediawiki-extensions-Echo/blob/master/includes/EventLogging.php#L58

So I investigated this somewhat.
If I thank on my local instance I do see an event logged in the Event log but it doesn't show up in the EventLogging database. I had to fiddle with $wgEchoConfig to get this to work.

If I event log on the client side it does show up in the database however. Weird.

Something funny is either happening in efLogServerSideEvent or the configuration on our wikis.

Note this all started on the 9th January which was a Thursday (deployment day).
What got deployed that day?

Event being logged on server on my local instance for a thanks:
{"event":{"version":"1.5","eventId":291,"notificationType":"edit-thank","notificationGroup":"positive","sender":"78","recipientUserId":1,"recipientEditCount":953,"eventSource":"mobilediff","deliveryMethod":"web","revisionId":3823},"schema":"Echo","revision":6081131,"clientValidated":true,"wiki":"my_wiki","recvFrom":"Administrators-MacBook-Air-68.local","timestamp":1391020937,"webHost":"localhost.beta","userAgent":"Mozilla/5.0\u0020(Macintosh;\u0020Intel\u0020Mac\u0020OS\u0020X\u002010_7_5)\u0020AppleWebKit/537.36\u0020(KHTML,\u0020like\u0020Gecko)\u0020Chrome/31.0.1650.63\u0020Safari/537.36"}

bsitu wrote:

Hey Jon,

There was a similar bug filed in https://bugzilla.wikimedia.org/show_bug.cgi?id=60555. I asked Dario to look at it in more detail.

cc-ing Ori and Nuria, we've all been looking at Event Logging since December.

I generated graphs for mediawiki.org and it looks like Echo logging died there on January 11th:
http://toolserver.org/~kaldari/echo/

Since it died on en.wiki on January 17th, this definitely points to something bad being deployed that week.

Thanks for the report. This was caused by an update to EventLogging's capsule schema. I have started importing events from the raw logs into the database and expect to have the data recovered within an hour. Are there other server-side events I should restore?

Thanks Ori. I'm not aware of any others. ServerSideAccount creation doesn't seem to have been effected for example.

Ori according to this graph:
http://mobile-reportcard.wmflabs.org/graphs/thanks-daily

This is still not fixed for mobile based events.
On enwiki I just tried thanking an edit but there is no sign of it in the database...

Oh.. just saw bug 60555 which I guess would explain this!

My intention is to have all events recovered by Wednesday, February 5th. I'll follow up here and on the lists.

(In reply to comment #12)

My intention is to have all events recovered by Wednesday, February 5th. I'll
follow up here and on the lists.

I missed this deadline, sorry. I will get to it this weekend.

Sorry on my side too as I could not get this this week. Will talk with ori over the weekend or Monday(2/10) to coordinate efforts.

Is there an updated ETA on this? On top of the dashboards, this issue is blocking analyses that rely on pure EventLogging data.

The recovery script is running. It should finish within 24 hours.

Ori / Nuria: Anything left to do here, or can this ticket be closed?

(In reply to Andre Klapper from comment #17)

Ori / Nuria: Anything left to do here, or can this ticket be closed?

Yep. All data has been recovered.