@DAbad: Would T288840: Migrate WikiWho service to VPS be a parent or subtask of this task? If it is, feel free to set that semantic connection via Edit Related Tasks.... Thanks!

API Questions:

• Does the API have any reference documentation? (e.g. Wikimedia API description)
  • WikiWho Documentation
  • WikiWho API Reference Documentation
• Is this a REST API? If not does it follow any standard protocol? Yes, it is RESTful
• Are there any known set of consumers?
• Do we have any SLO/SLAs defined for the service?
• Where is the API currently hosted?

2021/10/14 - Platform PMs met to discuss the request and break down work required. next steps include:

• Focusing on a data persistence (storage) solution and evaluating if we can potentially leverage data platform in the short-term (Luke)
• evaluating api requirements (Seve)
• all open questions will be posted to this ticket

Data Platform Notes:

• Based on storage estimates we should be ok to onboard the datasets with growth estimates in our Cassandra DB
• Data persistence should be simple enough but we don't have anything implemented yet for event streaming and triggering services

First off, let me thank you all for the quick assistance!

In T293386#7429031, @sdkim wrote:

• Does the API have any reference documentation?

There is also more detailed internal docs on setting up the API server from scratch. It is on a private GitHub repo. I can try to get these docs prepared for you, but I imagine most of it won't apply to our stack. https://github.com/wikiwho/WikiWho is the actual algorithm. How we store and serve that output will probably differ greatly from how Gesis (the current maintainer) does it. See F34639572 for a high-level overview of their stack.

There is one other component that we will need: https://github.com/wikiwho/WhoColor/. This is basically an extension to the WikiWho algorithm, which adds annotated HTML so you can visually represent attribution data of an article. This is the API endpoint used by Who-Wrote-That.

• Are there any known set of consumers?

As for Wikimedia specifically, we know of three major consumers: XTools, Who-Wrote-That and Education-Program-Dashboard. Gesis has informed us they receive roughly 292,500 requests a month (~9,750 a day). XTools caches its results; uncached it would by itself make roughly 8,000 requests a day.

• Do we have any SLO/SLAs defined for the service?

No.

• Where is the API currently hosted?

On Gesis' internal servers. We do not have access to these servers.

Some additional comments on the database side after discussion with SRE and Platform team today:

• Postgres DB and Pickle files are not supported by SRE
• Longer term would need to re-design to not use these

11/1 Platform PM Triage Meeting:

• Need to look at how existing will go into new framework
• Next step is that if they were relying on putting it on cloud services, should instead be using kubernetes
• Data Persistence: putting it on Cassandra would be reasonable to do by mid-June
  • Event streams + refactoring data model
• DA to follow-up on timeline

2021-12-08 Technology Steering Committee Discussion
Notes:

• Enterprise has been interested in how to potential leverage Wikiwho as well as how we would redesign
• Current timeline is not feasible to meet for data platform, however we will keep as use case

Action Items:

• Review how this could fit into future Data Platform, API Platform, etc. (include Enterprise)

An update: our VPS installation has been up and running for some time now, and performing very well. It appears we may even be able to acquire more storage capacity to add more languages. Re-engineering the stack to work in production will be a huge effort that I don't think Community-Tech has time for in the near-term. That said, we are still interested in bringing this to production, but are considering it low-priority for now.

Thanks to everyone who has investigated and helped with this effort so far! We don't want to forget about it, but I believe for the time being there's no need to dedicate more time to this unless you feel compelled to do so :) I will let someone else adjust the priority as they see fit. I will remove T288840 as a parent task, as that was repurposed to be about only the VPS installation, which is now resolved.

Thanks again!

One more update...

Today @Andrew took the system down to do some hypervisor maintenance, and he noted that the system is using server-local storage rather than ceph, that this is probably just a mistake, and that we can avoid future downtime if we rebuild the server and use distributed storage instead. (We can keep the Cinder volume and attach it to the new server.) We'll do that if/when we rebuilt the server.

Also, we still have one element of the system that doesn't come back up automatically after a reboot: the init.d celery daemon (sudo /etc/init.d/celeryd start). When we rebuild the system, we should also switch that init.d celery config out and use a systemd service instead.

After receiving a report of the service not working, I logged in and found celery down again. I've restarted it.

@lbowmaker: Any feedback to the previous comment? Thanks.