Proposed additions to `dask-ssh` CLI
See original GitHub issueI’ve been using dask-ssh
the last few days and it has been super helpful. Thanks!
I’ve run into two small things that I think might be of benefit to others to add but wanted to check in (especially given what looks like an upcoming switchover to asyncssh
).
-
dask-ssh
keeps an SSH connection open while the scheduler and clients are running – killing the process also shuts down the remote workers (this seems good). However, SSH timeouts also cause this to break and fall over.paramiko
(which is currently driving the show here) can send some noise over the wire to keep the connection “active”. This could be an optional way (given some--keepalive
flag) to work around timeouts. Might also be a bit of scope creep, but I thought I’d raise it. -
Add an option to disable
StrictHostKeyChecking
– shouldn’t be the default, but if you’ve just spun up resources to add to a cluster, you might not worry so much about having those machines inknown_hosts
.
Let me know if either/both of these are of interest and I’m happy to take them on.
Issue Analytics
- State:
- Created 4 years ago
- Comments:9 (9 by maintainers)
Top GitHub Comments
Hi @gforsyth ! I’m very glad to see that you’re engaging with Dask and SSH.
One of the biggest surprises of our user survey was that SSH is the most common deployment strategy, so improvements here likely have high impact.
Have you tried the solution in distributed.deploy.ssh2 ? This new implementation isn’t as battle-tested as the one that you’re currently using, but I think that it’s probably easier to manipulate. It just passes most things through to asyncssh and the workers.
https://github.com/dask/distributed/blob/f7f6bd77ab8d28d8811b94aa84528d3a470f1ff1/distributed/deploy/ssh2.py#L193-L213
Thanks Gil!
On Wed, Oct 16, 2019, 1:18 PM Gil Forsyth notifications@github.com wrote: