Moving parts of Wikimetrics onto production systems
Benefits:
- The Data Warehouse is on the production DBs. Wikimetrics will be able to run its Vital Signs queries against this DW.
- Labs DB (which wikimetrics currently uses) has had many outages and issues lately
- Production DBs perform better
Drawbacks:
- Wikimetrics was originally envisionned to run on public data (data in labs db)
- Will buy us two instances. One in labs, one in production.
- Pressure to move data to public cannot be made that easy any longer
Scope:
- Only productionizing backend (queue, scheduler for recurrent reports. 3 instances (dev, staging, production))
- Not the web frontend.
Reasons why wikimetrics has not been pushed into production:
- python dependencies:
Yuvi in the meantime debianized some and also offered to help with the remaining ones
Tasks:
- T76769 [Dev - 8 pts] upgrade wikimetrics to trusty (using same install process as we have now. Probably just setting up new instances, not uprgrading them)
- T76789 (Yuvi + Ops) debianizing any (scheduler/queue) wikimetrics dependencies that are not in trusty / our apt
- T76790 changing puppetization for scheduler/queue (Kevin -> assign to Andrew and provide context, add blocking task)
- T76791 Puppet Production role class for wikimetrics scheduler/queue (Kevin -> merge into 76790)
- T76770 [Dev 13 pts] change wikimetrics in labs to use debian packages, try in staging 1st, make sure this works alongside of pip install with webserver part too
- T76779 [Dev 13 pts] Fix oauth and do a quick pre-security review.
- T76780 [Dev 8 pts] Change admin script so it doesn't have to make symlinks, configure production instance to write files to /a/limn-public-data
- T76792 [Ops] new group that would allow you to sudo to the wikimetrics user (to enable dev to run admin script)
- T76782 [Security] have security review