Page MenuHomePhabricator

TTS server deployment strategy
Closed, InvalidPublic

Description

  • The TTS service is getting its own vagrant role as part of T151877: Set up puppet role for TTS server to make testing easier.
    • Currently this is done using the provided docker-compose. @Reedy if using a docker-compose (pointing to wmf-hosted images/repos etc) is a no-no for final deployment then give me a shout as we would then have to investigate some other deployment strategy.

So, SRE can you tell @Lokal_Profil whether docker-compose is a valid deployment strategy? Or if they need to do so something else...

Event Timeline

So, Operations can you tell @Lokal_Profil whether docker-compose is a valid deployment strategy? Or if they need to do so something else...

A valid deployment strategy for production ? No, it is not. It's fine for Mediawiki-vagrant of course, but it can't really lead to work that is reusable in production (which might or might not be the end goal, I am not sure yet, but I guess it is per T180015)

RobH triaged this task as Medium priority.May 1 2018, 2:36 PM
Theklan changed the task status from Open to Stalled.May 19 2018, 8:24 AM
Theklan subscribed.

This is a blocker for T183780

Vvjjkkii renamed this task from TTS server deployment strategy to 77daaaaaaa.Jul 1 2018, 1:13 AM
Vvjjkkii changed the task status from Stalled to Open.
Vvjjkkii raised the priority of this task from Medium to High.
Vvjjkkii updated the task description. (Show Details)
Mainframe98 renamed this task from 77daaaaaaa to TTS server deployment strategy.Jul 1 2018, 9:56 AM
Mainframe98 changed the task status from Open to Stalled.
Mainframe98 lowered the priority of this task from High to Medium.
Mainframe98 updated the task description. (Show Details)

Deployment via containers to our production Kubernetes cluster is probably an option. The containers themselves will need to be built using the tooling that SRE and Release-Engineering-Team are creating for the Release Pipeline project. As far as I understand it, Mathoid is now using this system (Release Pipeline (Blubber) + helm) for deployment. I can't find any comprehensive overview of setting up a new project this way, but that seems kind of reasonable since I think there is only one thing that has actually been made to use the pipeline so far.

I think the goal here would be to write a blubber file and have the service then deployed ot labs in a similar way to the wikidata termbox service.

@Tarrow tells me that right now the details of how we do that are in https://github.com/wikimedia/wikibase-termbox/tree/master/infrastructure (this is a mirror, actual code on gerrit)

The Speechoid service has been changed to use Blubber. A separate task will be set up to track deployment of the service/sub-services.