User:Lucas Werkmeister (WMDE)/Dispatch queue debugging

Just a rough collection of some Quarry queries to get a better idea of the state of the dispatch queue. Note that just opening the links will show you old results – to get current data, fork the queries and then run them (using your own login).

Most of these queries are intended for the case where there’s one very lagged wiki while other wikis are fine. If that’s not English Wikipedia, replace enwiki with the lagged wiki’s code in the forked queries before running them. (And then after starting the queries, while you’re waiting for them to complete, perhaps update their descriptions as well, to avoid confusion later on.)

Who’s doing stuff? edit

This is essentially a variant of Wikiscan specifically for edits that end up in the dispatch queue. If a handful of users dominate the results, perhaps ask them to slow down.

If you think those queries will take too long (on a dispatch queue lagged by 6 hours, the first one took about 30 seconds), try one of these instead, which only look at 10000 entries:

oldest newest
enwiki query query
all wikis query query

Comparing the “oldest” and “newest” lists might also inform you which users already stopped or throttled editing and aren’t pushing as many new edits into the dispatch queue, or which ones have just started editing but at a high enough rate that they would cause problems if not throttled.

What’s going on? edit

Oldest changes not dispatched to English Wikipedia

You can browse through the list and look at what the changes are actually doing (take the change_revision_id and plug it into Special:Diff).

How will it evolve? edit

Number of edits to be dispatched to English Wikipedia per ten-minute interval

Take those results and chart them in your favorite spreadsheet or other number-crunching application. If you’re lucky, there’s a large peak of old edits, and after that’s been processed, the lag could catch up very fast. If you’re unlucky, it might be the other way around – the current lag is just a long tail of relatively few edits, and once dispatching reaches the mountain of newer edits it will slow down even more.