Rethink how Codewind is deployed on Che?

[johnmcollier]

Currently when a user starts up Codewind on Che, the following happens: 1) The Codewind Theia extension gets downloaded and started 2) The Codewind Che sidecar automatically deploys Codewind and sets up a proxy for the Codewind Theia extension to communicate with it.

This current approach (specifically the Che sidecar) has a number of disadvantages:

• It requires modification of the Che install by the user before hand, to patch the Che roles to allow the sidecar to deploy Codewind
• Can’t run on any vanilla Che install, which hurts our user base and means we can’t ever onboard to the Che devfile registry.
• Requires us to run all of our workspaces in one namespace
• Hard to configure specifics of the Codewind install (volume size, custom ingress, tls, etc)

I've been wracking my brain for a while about how to address this in the current framework (che sidecar auto deploying Codewind), and I couldn't come up with anything that didn't require modification of the Che install/roles at the very least

Additionally, with the creation of the Codewind operator (and with the changes to cwctl in https://github.com/eclipse/codewind/issues/1735), it will be possible to have a Codewind Theia extension on Che that connects to an already running Codewind instance, via cwctl (rather than auto deploying Codewind).

So, I've been thinking, once we have the Codewind operator, is there much added value in having Codewind on Che automatically deploy Codewind for the user, given the downsides listed above?

What if we moved to a model where Codewind on Che just deployed the Codewind Theia/VS Code extension in a Che workspace, but Codewind would be configured separately (via it’s operator)? It would allow us to:

• Run on just about any Che install (no changes necessary to Che configurations or roles)
• Potentially onboard to the Che devfile registry, because we have no restrictions of where Codewind can run
• Make it easier for the user to configure the Codewind instance

But, we'd lose the niceties of having Codewind autodeployed for the user

[jopit]

Seems like we're trading ease of use for the user vs ease of use for admin

Plus is that Hybrid and Che are more consistent

No real strong consensus one way or the other. Need input from Elson/TimD. Table this and come back to this in a future tech-topic.

[johnmcollier]

Discussed more on today's dev stat. We acknowledged the advantages and disadvantages of both approaches and rather than deciding on one approach over another, decided to support a hybrid model:

• Support a Che plugin that can be installed into any Che. Only sets up the Codewind VS Code / Theia extension. Up to user (or cluster admin, etc) to set up and configure Codewind as needed
• Also support the existing "Codewind and Che" model where users can set up Che to work with Codewind and have an "automatic setup" of Codewind, for the users who wish to have it.

We decided to leave the techtopics label on for now, and we'll discuss more / finalize when everyone is back in two weeks.

[johnmcollier]

Removing techtopics label. Discussed with Elson and the squad and decided to go with the hybrid approach described above.