I have set "btl_tcp_if_include = ib0" in ~/.openmpi/mca-params.conf. When I run mpirun without podman, it runs right. When I run mpirun with podman, the "btl_tcp_if_include = ib0" is invalid. But when I add "--mca btl_tcp_if_include ib0" to command line, it is valid. $ mpirun -np 128 --hostfile mfile --mca btl_vader_single_copy_mechanism 3 --mca orte_tmpdir_base /tmp/podman-mpirun podman run --env-host --env-file envfile --userns=keep-id --network=host --pid=host --ipc=host -v /tmp/podman-mpirun:/tmp/podman-mpirun:z -v ./bulk:/bulk -w /bulk 4756bc6d2be0 vasp_std running on 128 total cores [ga2211][[19883,1],64][btl_tcp_endpoint.c:626:mca_btl_tcp_endpoint_recv_connect_ack] received unexpected process identifier [[19883,1],88] ... ... Here is hang. ... ... VASP version: vasp.5.4.4.18Apr17-6-g9f103f2a3. OpenMPI version: 4.1.1 $ podman version Client: Podman Engine Version: 4.0.2 API Version: 4.0.2 Go Version: go1.16.13 Git Commit: c99f9f1b6960b98158b7f5fc4e6b5ac1a10b3542 Built: Wed Mar 23 18:33:41 2022 OS/Arch: linux/amd64
On Wed, Jun 22, 2022 at 10:18:46AM -0000, None via Podman wrote:
I have set "btl_tcp_if_include = ib0" in ~/.openmpi/mca-params.conf. When I run mpirun without podman, it runs right. When I run mpirun with podman, the "btl_tcp_if_include = ib0" is invalid. But when I add "--mca btl_tcp_if_include ib0" to command line, it is valid.
It looks like your home directory is not mounted into the container and maybe therefore Open MPI (inside of the container) cannot see the setting. Adrian
Thank you. I have set home directory in container, and set .openmpi directory. It runs success.
participants (2)
-
Adrian Reber -
ugiwgh@qq.com