More (non-live) statistics

CLOSED This issue is closed by Nemo_bis

There is no interest in statistics from TWN shell users. The only option is to self-produce and use XML dumps. Currently tracked at phabricator:T86466.

I'd like to have some more detailed multi-years graphs and repos to stare at, as on m:stats:. Those scripts are quite a hassle to adapt and they require dumps, so double effort; however there are some nice solutions which should only require install and connect to the database, like m:StatMediaWiki. I've no idea how much resources this can take but hopefully it can be adjusted to run updates only as often and as quickly as permitted by our capacity, maybe taking advantage of some low-traffic hours.

More direct links: manual, checkout. Can we just install it (even just on devwiki I'm told that doesn't help) and see how it goes? If yes I can submit patches for the config if needed, but I have no shell access so someone would need to do the checkout I guess (or should that be done by puppet too?).

We don't really have low-traffic hours. The difference between highest and lowest is less than 100%.

If you mean less than a factor 2, half standard capacity free sounds quite a lot. :)

Update: the work needed from outside the server is disproportionately higher, but I'm working on it nevertheless. Nike also wrote a blog post, so I'm gradually expanding a FAQ item at FAQ#Are_there_statistics_on_translatewiki.net?.

• m:StatMediaWiki: I've tried it (see docs on Meta) and it's extremely easy to set up once you have a MediaWiki install with the database (about ten minutes). However, building a MediaWiki instance and importing the twn data is extremely tedious: I spent several hours on this and now the import is running, it will take a few weeks. Then I expect it will take up to several days to run (in one thread).
• m:Wikistats. This is designed for huge datasets, it's ridiculously fast for a wiki like ours. I spent most of today fighting with its weirdnesses and hardcoded configs but then it took 70 min to run plus 6 min to generate the reports. See links in FAQ. Now I'm trying to run the more complete version to fill all the blanks in tables and graphs, but the most important stats are already there!