Firewall settings? #251
-
Hi, I'm trying out clustermq on a Slurm cluster. Using a simple example, the job launches, but then nothing happens. AFAIU, it is ZeroMQ that fails to set up a working communication with the main R session and the worker(s) launched on the cluster's compute node. BTW, I've verified that ZeroMQ works on the localhost using the rzmq main-worker example from https://cran.r-project.org/web/packages/rzmq/readme/README.html. As soon as I attempt to use the same example running across two hosts, it blocks. I suspect I need to request to have the ZeroMQ protocol opened up on the cluster for clustermq to be able to run on this cluster. I have near-zero experience with ZeroMQ - is this a matter of opening up TCP ports in the firewall? If so, is there a standard range that ZeroMQ uses? |
Beta Was this translation helpful? Give feedback.
Replies: 3 comments
-
As far as network connections are concerned, ZeroMQ behaves just like regular sockets. The node from where you call I've seen in the past that a certain network interface blocks these connections, so it may be possible to solve this by setting the |
Beta Was this translation helpful? Give feedback.
-
Can you please explain this a bit further? I think I am seeing a similar problem, but setting the (Worker log without setting
and the console freezes at |
Beta Was this translation helpful? Give feedback.
-
I assume this resolves the initial question @HenrikBengtsson, so I'm closing this @mhesselbarth If your problem persists, please open a separate issue with a more detailed description of what you are trying to do + where it fails |
Beta Was this translation helpful? Give feedback.
As far as network connections are concerned, ZeroMQ behaves just like regular sockets. The node from where you call
Q
will need to be able to accept incoming connections onSys.info()["nodename"]
on the same port as the clustermqID
(printed when starting this job, between 6000 and 9999).I've seen in the past that a certain network interface blocks these connections, so it may be possible to solve this by setting the
clustermq.host=<network interface>
option.