Suggestions on error of 6 | Connection refused

[psychelzh]

Originally posted in wlandau/crew.cluster#38

Thanks for this great package, which is especially useful for crew package. The problem comes from using it with sge system, and these days I figured out that the problem might come from mirai, given these logged errors:

Error in dial(sock, url = url, autostart = asyncdial || NA, tls = tls,  : 
  6 | Connection refused
Calls: <Anonymous> ... do.call -> <Anonymous> -> dial_and_sync_socket -> dial
Execution halted

Thank you very much for any suggestions.

[shikokuchuo]

This error is simply that the daemons (workers) cannot connect back to the host. Two likely possibilities:

1. TLS is misconfigured - e.g. you are using your own certificate, but it is for a domain name which doesn't match that of the node you are connecting back to. Best way to rule out is by simply disabling TLS in crew.
2. There is a firewall around your login node that prevents inbound connections from the worker nodes.

Does the above help? For 2, I am guessing this may require your sysadmins to resolve.

[psychelzh]

Thanks! I will do troubleshooting first, and post whatever I found in a proper place.

[shikokuchuo]

Actually I have a theory as to what this is. Most likely the versions of mirai on your login node and worker nodes have diverged. This is quite possible as mirai 0.13.2 was just released to CRAN a couple of days ago. If some have updated but some have not then you will run into this protocol error.

[psychelzh]

I have also thought about this possibility. However, given I am using renv to manage the versions of the packages, I found it is not an issue. Actually, I am still using 0.12.1 version of mirai, which is configured by the renv lock file. Will this rule out that issue? Or will renv mess things up?

[shikokuchuo]

Is it possible for you to upgrade to mirai 0.13.2 everywhere? Version 0.12.1 suffers from an issue that is somewhat tangentially related and was fixed in 0.13.1. I don't think it will be possible to cleanly diagnose this current issue using that version.

Use of renv itself should be fine and I know there are crew users who use renv.

[psychelzh]

Great. That is an option. I will have a try on this.