ucx-py: Unable to connect on any available transport

Any suggestions on how to diagnose/debug this?

(dev) mrocklin@dgx16:~$ python ucx-py/tests/send-recv-py-obj.py
[1556057487.297828] [dgx16:3899 :0]   ucp_listener.c:263  UCX  ERROR none of the available transports can listen for connections on 0.0.0.0:13337
[1556057487.297838] [dgx16:3899 :0]   ucp_listener.c:263  UCX  ERROR none of the available transports can listen for connections on 0.0.0.0:13338
[1556057487.297841] [dgx16:3899 :0]   ucp_listener.c:263  UCX  ERROR none of the available transports can listen for connections on 0.0.0.0:13339
[1556057487.297844] [dgx16:3899 :0]   ucp_listener.c:263  UCX  ERROR none of the available transports can listen for connections on 0.0.0.0:13340
[1556057487.297846] [dgx16:3899 :0]   ucp_listener.c:263  UCX  ERROR none of the available transports can listen for connections on 0.0.0.0:13341

About this issue

  • Original URL
  • State: closed
  • Created 5 years ago
  • Comments: 17 (14 by maintainers)

Most upvoted comments

@madsbk I believe TCP support has been added in these PRs:

@Akshay-Venkatesh has tested the work here: https://gist.github.com/Akshay-Venkatesh/2c48e3c8b682d44ad98bb40ce8e043aa#file-pytest-results-txt

Relevant ENV VARS:

UCX_TLS=tcp,cuda_copy,sockcm UCX_SOCKADDR_TLS_PRIORITY=sockcm

@Akshay-Venkatesh, what is the status of TCP support in UCX? Should it work now?

I am trying to install UCX+UCX-Py on my workstation and with UCX master I get the error:

UCX  ERROR none of the available transports can listen for connections on 0.0.0.0:13337