Skip to content

CABINET is a utility that standardizes subsequent container execution. Emphasizing reproducibility, users can share JSONs to exactly replicate how containerized processes follow each other.

Notifications You must be signed in to change notification settings

DCAN-Labs/CABINET

Repository files navigation

CABINET

CABINET is a way to bind multiple containers together and run them with a single command. This is done by composing a parameters json that contains any necessary inputs for whichever containers your heart desires. With one command, CABINET will then run these containers one after the other to produce the desired outputs.

Looking for BIBSnet?

BIBSnet is a segmentation pipeline that historically had its development tied to the development of CABINET. If you used the DCAN Labs codebase named CABINET prior to October 2023 you were using the codebase we now call BIBSnet. Find BIBSnet here.

Arguments

Required positional arguments

parameter-json: A filepath to the parmeter JSON to run. See Setting up the parameters.json section for more information.

Optional flag arguments

--dryrun: If included run.py will validate the parameter json file without running any containers.

Setting up the parameters.json

To use CABINET, users will need to create a parameters json that contains the information about which containers will be run. Any amount of containers can be added to the json. The json will have any necessary binds, arguments, and flags to run the container. There are two required keys: cabinet, which contains options for how CABINET will run and stages which contains options for how each container in CABINET will run. If you would like to make sure your json is valid before trying to run all of the containers, run run.py with the --dryrun flag. Before trying to run all of the containers, the main run.py script will also validate the json. If the json is invalid, it will not try to run the containers.

The parameter-jsons folder contains examples of parameter jsons that string together multiple containers. If you use CABINET for any containers that aren't in the parameter-jsons folder, please add your json as a resource for others to use.

cabinet variables

container_type: Required; the type of containers you are running (singularity or docker).

stages: Required; a list of stages to run in the order they should be run in. All stages listed here must have an entry in stages but not all entries in stages need to be listed here. This allows you to set up run commands for containers but not run them every time you run CABINET.

verbose: Optional; value of true or false; if not specified, defaults to false.

handle_missing_host_paths: Optional; how you want the container to behave if any given paths don't exist; value of stop, make_directories, or allow; defaults to stop.

  • stop will cause the json to be invalidated if paths don't exist.

  • make_directories will create the directory if it doesn't exist. NOTE: If you're binding a file (instead of a directory) and it doesn't find the file, it will create a directory instead (not recommended).

  • allow will not do anything about the missing directories.

log_directory: Optional; a valid filepath. If included stdout will be routed to .log files that will be stored in this directory, one log file for each stage. Overwrites any existing files with the same name. If left blank stdout will behave normally.

job_id: Optional; string to prepend to log file name. Only used if log_directory is specified. Defaults to the timestamp at the time of JSON validation.

stop_on_stage_fail: Optional; boolean. Whether to stop when one stage fails (true) or continue executing other stages (false). Defaults to true.

stages variables

action: Optional; value of exec or run; defaults to run

container_args: Optional; arguments for singularity or docker run options

positional_args: Optional; any necessary arguments for the container

flags: Optional; any necessary flags to run the container

Singularity stage variables

container_filepath: Required; path to sif file of container

binds: Optional; any necessary binds for the containers; if specified, need both the host_path and container_path

Docker stage variables

image_name: Required; name of the Docker image to run.

mounts: Optional; any necessary mounts for the containers; if specified, need both the host_path and container_path

All optional variables without an explicit default will appear in the validated JSON as an empty data structure (when using the verbose output).

Example JSON

Below is an example for how BIBSNet would be run.

    {
        "cabinet": {
            "container_type": "singularity",
            "stages": ['bibsnet']
            "verbose": true,
            "handle_missing_host_paths": "make_directories"
        },

        "stages": {
            "bibsnet": {
                "container_filepath": "/path/to/container/bibsnet.sif",
                "action": "run",
                "container_args": {
                    "--cleanenv": true,
                    "--nv": true
                },
                "binds": [
                    {
                        "host_path":"/path/to/input/",
                        "container_path":"/input"
                    },
                    {
                        "host_path":"/path/to/output/derivatives/",
                        "container_path":"/output"
                    },
                    {
                        "host_path":"/path/to/workdir/",
                        "container_path":"/work"
                    }
                ],
                "positional_args": [
                    "/input", "/output", "participant"
                ],
                "flags": {
                    "--parameter-json":"/home/cabinet/parameter-file-container.json",
                    "-start":"prebibsnet",
                    "-end":"postbibsnet",
                    "-v": true,
                    "--participant-label": SUBJECT-ID,
                    "-w":"/work"
                }
            }
        }
    }

Running

Once you have created your parameter.json, run run.py with your json as the single argument.

At the beginning of each stage, CABINET will print the name of the stage and at which time it started. At the end of the stage, it will print out how long that stage took to run and if it completed successfully or failed. Right now, even if a stage fails, it will continue to try to run the remaining containers. If you have the verbose option set to true, at the very beginning it will print the validated JSON and during each container it will print the singularity commands it is running.

After it runs all of the containers, it will print out how long the entire process took to run and if every stage completed successfully or if some stages failed. Right now, it will not tell you which stages failed at the end, you will have to look at the error and output logs to determine why and where it failed.

Below is an example output from running bibsnet and lolcow.

INFO 2023-09-08 16:22:18,049: Getting Arguments from arg file: /panfs/jay/groups/6/faird/tikal004/CABINET/parameter-jsons/bibsnet-nibabies-xcpd.json

INFO 2023-09-08 16:22:18,052: Validating parameter JSON

INFO 2023-09-08 16:22:18,082: Made directory /home/feczk001/shared/projects/segpipeline_testing/Barry_test/CABINET-test/derivatives/bibsnet/

INFO 2023-09-08 16:22:18,092: Parameter JSON /panfs/jay/groups/6/faird/tikal004/CABINET/parameter-jsons/bibsnet-nibabies-xcpd.json is valid.
Validated JSON: {'cabinet': {'container_type': 'singularity', 'verbose': True, 'handle_missing_host_paths': 'make_directories'}, 
'stages': [{'name': 'bibsnet', 'container_filepath': '/home/faird/shared/code/internal/pipelines/cabinet_container/cabinet_v2.4.3.sif', 'container_args': {'--cleanenv': True, '--nv': True}, 
'binds': [{'host_path': '/home/faird/shared/data/BCP-MVP_BIBSNet/input/', 'container_path': '/input'}, {'host_path': '/home/feczk001/shared/projects/segpipeline_testing/Barry_test/cabinet-test/derivatives/', 'container_path': '/output'}, 
{'host_path': '/home/feczk001/shared/projects/segpipeline_testing/Barry_test/cabinet-test/work/', 'container_path': '/work'}], 'positional_args': ['/input', '/output', 'participant'], 
'flags': {'--parameter-json': '/home/cabinet/parameter-file-container.json', '-start': 'prebibsnet', '-end': 'postbibsnet', '-v': True, '--participant-label': '123456', '-w': '/work'}, 'action': 'run'}, 
{'name': 'lolcow', 'container_filepath': '/home/faird/tikal004/lolcow.sif', 'action': 'exec', 'positional_args': ['cowsay', '"Thank you for testing CABINET"'], 'container_args': {}, 'binds': [], 'flags': {}}]}

INFO 2023-09-08 16:22:18,092: Identified stages to be run: ['bibsnet', 'nibabies', 'xcpd', 'lolcow']

INFO 2023-09-08 16:22:18,092: Now running stage: bibsnet

INFO 2023-09-08 16:22:18,093: run command for bibsnet:
singularity run -B /home/faird/shared/data/BCP-MVP_BIBSNet/input/:/input -B /home/feczk001/shared/projects/segpipeline_testing/Barry_test/cabinet-test/derivatives/:/output 
-B /home/feczk001/shared/projects/segpipeline_testing/Barry_test/cabinet-test/work/:/work --cleanenv --nv 
/home/faird/shared/code/internal/pipelines/cabinet_container/cabinet_v2.4.3.sif /input /output participant --parameter-json /home/cabinet/parameter-file-container.json -start prebibsnet -end postbibsnet -v --participant-label 123456 -w /work

INFO 2023-09-08 22:00:30,762: bibsnet failed. Time elapsed since bibsnet started: 1:05:52.760840

INFO 2023-09-08 22:00:30,762: Now running stage: lolcow

INFO 2023-09-08 22:00:30,762: run command for lolcow:
singularity exec /home/faird/tikal004/lolcow.sif cowsay "Thank you for testing CABINET"

_________________________________
< "Thank you for testing CABINET" >
---------------------------------
        \   ^__^
         \  (oo)\_______
            (__)\       )\/\
                ||----w |
                ||     ||

INFO 2023-09-08 22:00:31,781: lolcow finished. Time elapsed since lolcow started: 0:00:01.018425
CABINET ran for this long but some stages were not successful: 5:38:13.735623

About

CABINET is a utility that standardizes subsequent container execution. Emphasizing reproducibility, users can share JSONs to exactly replicate how containerized processes follow each other.

Resources

Stars

Watchers

Forks

Packages

No packages published