[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [HTCondor-users] [EXTERNAL] Re: Docker networking



Hi again, Greg

 

So, we were able to move to HTCondor 9.10.0 and, as expected, adding docker_network_type = host to my submit file got my docker container using the network from the host. One step forward!

 

But, two steps back, … docker is also running MPI in the container and, when I use host network, I get an error like this:

MPID_nem_init(320).................:

MPID_nem_tcp_init(173).............:

MPID_nem_tcp_get_business_card(420):

MPID_nem_tcp_init(379).............: gethostbyname failed, <<address>> (errno 1)

 

Have you seen something like this? Do I somehow need to pass the hostname from the host into the container? Based on some similar errors and searching, that’s what it seems like:
Fatal error in MPI_Init_thread: Other MPI error, error stack: · Issue #44 · conda-forge/fenics-feedstock

 

But I wanted to see if I’m missing something obvious.

 

Many thanks

Mike

 

--

Michael Fienen, Ph. D.
Research Hydrologist
United States Geological Survey

Upper Midwest Water Science Center
1 Gifford Pinchot Drive

Madison, Wisconsin 53726
phone:  608.821.3894
https://www.usgs.gov/staff-profiles/michael-n-fienen

 

 

 

From: HTCondor-users <htcondor-users-bounces@xxxxxxxxxxx> on behalf of Greg Thain <gthain@xxxxxxxxxxx>
Date: Tuesday, July 26, 2022 at 10:02 AM
To: htcondor-users@xxxxxxxxxxx <htcondor-users@xxxxxxxxxxx>
Subject: [EXTERNAL] Re: [HTCondor-users] Docker networking

 

 This email has been received from outside of DOI - Use caution before clicking on links, opening attachments, or responding.  

 

 

On 7/25/22 5:30 PM, Fienen, Michael N via HTCondor-users wrote:

Is it possible to enable host-based networking in HTCondor <9.0?

 

I tried adding DOCKER_EXTRA_ARGUMENTS = "--network=host" to my config files but that didn’t work.

 

This feature was added in HTCondor 8.9.9, so it won't be in a stable version before that.  Given that, maybe the best way to get this option added is to redefine the HTCondor knob "DOCKER" to point to a script you write, which inserts --network=host to the arguments, if we are running docker run ...  This is a bit of work, but if you can't upgrade, I don't think there's a better way.

-greg