OneShotSlave is an EphemeralNode by design. Would need to be able to (re)create such a node as pipeline is restarted, i.e.

1. identify an AfterRestartTask enters the Queue
2. look into associated WorkflowRun.flowExecution to know about expected node() on resume
3. create a OneShotSlave with exact same name

Doesn't seem to work.

Looks to me there's no simple option to support pipeline re-hydrate that - by nature - is incompatible with this idea of an ephemeral node allocated to a build.

One-shot executor is designed to change the build:slave paradigm, relying on plain ephemeral nodes. The way pipeline expect executor to resurrect after a reboot just is incompatible, one-shot should just not be used in this context.

Under discussion with kohsuke.

Met with ndeloof and jglick today to discuss the way forward. During the meeting, we discussed relevant part of the code and a possible way to fix this.

I think that gave Nicolas some directions and ideas and he is going to give it some more shot.

The current plan is for me to check back in with Nicolas later this week, and if it doesn't work out, put this to the OSS team's queue and ask Jesse to look into this. Jesse is more optimistic that this is fixable, just that he currently lacks the context of one-shot-executor.

I talked to ndeloof and he's making good progress, but still more work to do.

Looks like Nicolas made a commit in October 2017 to the One Shot Executor Plugin, but it is not part of any official release of that plugin.  Is issue fixed?

I cannot definitely speak for ndeloof but I believe that plugin is abandoned. You can try dockerNode from the docker-plugin as a conceptual successor.

Thanks Jesse.