top-level Client function to shutdown workers

[jreback]

If I have remotely launched workers with a nanny process I can simply use client.restart() to dump the cluster. However a Worker (say launched via a resource manager or some such), can right now be cleanly terminated via:

from distributed.utils import sync
sync(client.loop, client.scheduler.broadcast, msg={'op': 'terminate'})

Client.shutdown() does not do this (despite it seems that the doc-string indicates it.

So maybe add a method

Client.shutdown_workers()?

[adamklein]

+1. I'd like to cleanly shutdown my cluster so the Nanny process doesn't respawn the worker.

[adamklein]

Also, can't seem to close scheduler so it exits the CLI-launched scheduler gracefully. I tried

        sync(client.loop, client.scheduler.terminate, fast=False)

Which seems to close things, but the process hangs around.

[adamklein]

@mrocklin Confirming there's still no way to do this except

sync(client.loop, client.scheduler.broadcast, msg={'op': 'terminate'}, nanny=True)

?

[mrocklin]

Ah no, this might suffice: #1876

[GenevieveBuckley]

@adamklein does the retire_workers functionality added in #1876 solve your issue?

[jrbourbeau]

Closing due to lack of activity. Feel free to re-open if folks would like to continue