Suggestion: should we do this in parallel instead?

What's the argument for doing it in parallel? My gut feeling is I want to avoid spamming coderd. If this is something we want, we should allow deleting multiple agents in one request instead. // cc@DanielleMaywood

Main argument is to reduce time to shutdown. Currently it is worst case maximumlen(subAgentIDs) * deleteCtx timeout. Not a blocker though.

Gotcha 👍🏻. It's btw just1 * deleteCtx timeout, we give up on deleting the rest if it takes too long.

Gotcha. That may be another argument for doing multiple at once - I'd pefer it to be all or nothing.

This name sounds confusing to me 🤔. From my understanding, we have two agents, WorkspaceAgent, and Devcontainer Agent, is that correct? If yes, why not call DevcontainerAgent to simplify the naming?

I think prefixing the resource names, when not necessary, a bad practice, but I see we do that in a lot of places in coderd, so it is not a blocker, but I would like to bring attention to it.

I do see your point. Technically I suppose "Workspace" in "WorkspaceAgent" is also somehwhat useless. The main idea with this is to follow existing naming conventions but also to symbolize the dependency tree. A devcontainer agent (a workspace agents subagent) cannot exist without a (parent) workspace agent. Hence it is nested in the naming. Another existing type likeWorkspaceAgentContainer also already exists so I don't think it's worth diverging right now without a plan for how to tackle existing names.

In the FE, we handle requests in two ways: queries or mutations. In this case, you should wrap this request into a mutation.Here is the docs.

When using a mutation, you can invalidate other queries or update their data to reflect the most recent status, instead of doing it manually, so you wouldn't need the isRebuilding or subAgentRemoved statuses.Here is how you can do it. If you need more examples, you can search in the codebase for "invalidateQueries".