Enroot Integration Strategy

[dciangot]

Configuration Extension

Add Enroot-specific fields to SlurmConfig in pkg/slurm/types.go:23-27:

  EnrootDefaultOptions []string `yaml:"EnrootDefaultOptions" default:"[\"--rw\"]"`
  EnrootPrefix         string   `yaml:"EnrootPrefix"`
  EnrootPath           string   `yaml:"EnrootPath"`
  ContainerRuntime     string   `yaml:"ContainerRuntime" default:"singularity"` // "singularity" or "enroot"

Command Structure Adaptation

• Singularity: singularity run/exec [options] [image] [command] [args]
• Enroot: enroot start [options] [container_name] [command] [args]

Key differences:

• Enroot requires container creation before execution (enroot create)
• Enroot uses container names instead of direct image paths -> How to handle images?

Implementation Changes

Abstract Container Runtime (pkg/slurm/Create.go:86-160)

Replace hardcoded Singularity logic with runtime abstraction:

• Create ContainerCommand interface
• Implement SingularityCommand and EnrootCommand structs
• Factory pattern for runtime selection

Mount Handling (pkg/slurm/prepare.go:346-347)

Enroot mount differences:

• Use --mount instead of --bind
• Different syntax: --mount type=bind,src=/host/path,dst=/container/path
• Support additional mount types: tmpfs, devpts

Environment Variables (pkg/slurm/prepare.go:196-234)

Enroot environment handling:

• use --env-file with different format requirements

Image Management Strategy

Image Import Phase

• Add pre-job image preparation:

  # Check if container exists
  enroot list | grep -q $CONTAINER_NAME || \
  # Import and create if not exists
  enroot import $IMAGE_URI && \
  enroot create --name $CONTAINER_NAME $IMAGE_SQSH

Container Lifecycle

• Import: enroot import docker://image:tag
• Create: enroot create --name pod-uid-container image.sqsh
• Start: enroot start --rw pod-uid-container command args
• Cleanup: enroot remove pod-uid-container

Script Generation Modifications

• Update produceSLURMScript() function in pkg/slurm/prepare.go:522-814:

Pre-execution Setup

  # Container preparation phase
  prepare_enroot_containers() {
      for container in $CONTAINERS; do
          if ! enroot list | grep -q $container; then
              enroot import $IMAGE_URI
              enroot create --name $container $IMAGE_SQSH
          fi
      done
  }

Command Construction

  # Enroot execution
  runCtn container-name enroot start --rw \
      --mount /host/path:/container/path \
      --env KEY=VALUE \
      container-name \
      /command args

Backward Compatibility

• Default to Singularity runtime if not specified
• Maintain existing Singularity annotations and behavior
• Gradual migration path through configuration

[dciangot]

@ffais I can help with the overall implementation and how to test it. Not an expert in enroot though, so I'd expect you to help in that, let me know if the assumption is wrong.

[ffais]

I've created a draft PR #79 to work on this. The first version is a POC in which the basic container functionality (creation and deletion) works, at least in our environment. I need to work on a better abstraction to better integrate the two runtimes.