ERROR 'NoneType' object has no attribute 'get_mount_point'

[howff]

DEBUG    EXECUTING command test -e '/tmp/streamflow' && readlink -f '/tmp/streamflow' 2>&1 on singularity-wraps-slurm/752445a6-876d-403c-9170-603d07886508
DEBUG    EXECUTING command singularity exec   instance://752445a6-876d-403c-9170-603d07886508 sh -c 'echo dGVzdCAtZSAnL3RtcC9zdHJlYW1mbG93JyAmJiByZWFkbGluayAtZiAnL3RtcC9zdHJlYW1mbG93JyAyPiYx | base64 -d | sh' 2>&1 on __LOCAL__/__LOCAL__
ERROR    'NoneType' object has no attribute 'get_mount_point'

Traceback

  File "streamflow/core/recovery.py", line 44, in wrapper
    await func(*args, **kwargs)
  File "streamflow/workflow/step.py", line 1432, in _schedule
    await self.workflow.context.scheduler.schedule(
  File "streamflow/scheduling/scheduler.py", line 554, in schedule
    task.result()
  File "streamflow/scheduling/scheduler.py", line 339, in _process_target
    for requirements in await asyncio.gather(
  File "streamflow/scheduling/scheduler.py", line 474, in _resolve_hardware_requirement
    hardware_requirement = await utils.bind_mount_point(
  File "streamflow/data/utils.py", line 38, in bind_mount_point
    mount_point = await get_mount_point(context, location, disk.bind)
  File "streamflow/data/utils.py", line 70, in get_mount_point
    return location.hardware.get_mount_point(path)
AttributeError: 'NoneType' object has no attribute 'get_mount_point'

This is when I try to use a streamflow like this:

version: v1.0
workflows:
  extract-and-compile:
    type: cwl
    config:
      file: workflow.cwl
      settings: config.yml
    bindings:
      - step: /compile
        target:
          deployment: singularity-wraps-slurm
          service: localslurm

deployments:
  slurm:
    type: slurm
    config:
      services:
        localslurm:
          nodes: "1"
          mem: 1gb
          file: my_sbatch.sh

  singularity-wraps-slurm:
    type: singularity
    config:
      image: jdk17.sif
    wraps: slurm

[howff]

location.hardware is None because

location = slurm/__LOCAL__/slurmctld
location.hardware = None

In streamflow/data/utils.py if I adjust the try block to except AttributeError and return None then everything seems to work...

[LanderOtto]

Your approach of wrapping the Singularity deployment over the Slurm deployment is correct. However, fully integrating Singularity (and Docker) with the Slurm connector is still a work in progress, and we are aware that some issues may arise. We are actively working on a more permanent fix.

In the meantime, we have been bypassing these issues in our own experiments by leveraging the file attribute of the SlurmService (documentation).

That said, if your proposed solution fixes your specific use case, we are happy to hear it. Please let us know if you encounter any further problems

[howff]

Thank you for commenting on the singularity-wraps-slurm aspect. The problem that I have here is that when I specify service: localslurm in bindings it is passed onto the singularity-wraps-slurm deployment, but then it gets lost and not passed onto the slurm deployment which means that the localslurm service config is ignored. Is there a different syntax I should be using?

[LanderOtto]

When using the wraps directive, you can explicitly specify both the deployment and the service (documentation).

For example, your streamflow.yml should be structured like this:

deployments:
  slurm:
    type: slurm
    config:
      services:
        localslurm:
          nodes: "1"
          mem: 1gb
          file: my_sbatch.sh

  singularity-wraps-slurm:
    type: singularity
    config:
      image: jdk17.sif
    wraps: 
      deployment: slurm
      service: localslurm