Skip to content

[Bug]: mpirun doesn't work on non-GPU nodes #3434

@un-def

Description

@un-def

Steps to reproduce

type: task

nodes: 2
startup_order: workers-first
stop_criteria: master-done

commands:
  - |
    if [ $DSTACK_NODE_RANK -eq 0 ]; then
      mpirun \
        --allow-run-as-root \
        --hostfile $DSTACK_MPI_HOSTFILE \
        -n 2 \
        -N 1 \
        --bind-to none \
        hostname
    else
      sleep infinity
    fi

resources:
  gpu: 0

Actual behaviour

Warning: Permanently added '[192.168.0.43]:10022' (ED25519) to the list of known hosts.
--------------------------------------------------------------------------
All nodes which are allocated for this job are already filled.
--------------------------------------------------------------------------

Expected behaviour

No response

dstack version

0.20.1

Server logs

Additional information

# cat $DSTACK_MPI_HOSTFILE
192.168.0.94 slots=0
192.168.0.43 slots=0

Note slots=0

Metadata

Metadata

Assignees

Labels

bugSomething isn't working

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions