Page tree

Versions Compared


  • This line was added.
  • This line was removed.
  • Formatting was changed.
Comment: Published by Scroll Versions from this space and version 1.4

Background and Problem

There are several use cases where it would be nice to run a little bit of code before launching a container.

One such case is when running the BIDS Apps (

serverNeuroinformatics Research Group JIRA
). The raison d’être of the BIDS Apps is that they run on BIDS-formatted data. The user launches them by sending in a path, and it is implied that the data found at that path will be formatted in the BIDS structure. However, this is incompatible with how the container service typically works, which is by mounting data directly from the archive into a container and kicking off a process. When our data is in the archive it isn’t stored in the BIDS structure, it is stored in the XNAT archive structure.

We do have a container, xnat/dcm2bids-session (github) (docker hub), that ostensibly converts DICOM data in XNAT to BIDS format. That is correct up to a point. It generates the NIFTI files and BIDS JSON metadata, but it keeps all that data in the XNAT archive format rather than storing it all at rest in BIDS format. Even after running this container on our data we still can’t launch any BIDS Apps directly; because the data are stored in the XNAT archive format the BIDS Apps would not be able to read them. The dicom2bids container gets us most of the way to launching BIDS Apps. We need an extra push to get all the way.

Solution: Setup Commands

The way we solve this problem is by introducing a new type of command called a “setup command”. The job of a setup command will be to take data that is in the XNAT archive format and stage it / move it / convert it into whatever format is required for the “main” command. The author of a main command (for instance, a command that describes a BIDS App) who has a need for a setup command will have to first create a setup docker image, write a setup command to describe it, and reference that setup command in their main command. At launch time, the Container Service will launch a container from the setup command before starting the main container from the main command, thus giving the setup container a chance to stage the files in the right way for the main container.

How a Setup Command Changes the Container Launch Flow

When the Container Service goes to launch a container from a command, if it sees that the command makes reference to a setup command, it will create and start a setup container before starting the main container. To see where and how that happens, let us examine the launch sequence with and without a setup command.

Without Setup Command

First we will examine the Command Resolution and Container Launch sequence for a command that does not reference a setup command. Here is a (very abbreviated) version of finding the file paths for a mount.

  1. The Command has a mount named input. The mount defines the path inside the container, which we will call p.
  2. The Command Wrapper has an input which provides files for mount input.
  3. That Command Wrapper input has a runtime value which we can resolve into some XNAT object. We find the path to that object’s files in the archive, which we will call A.
  4. We will mount the path A on the host to the path p in the container.

Once the rest of the Command Resolution process is complete, the container will be created and started with a mount which takes path A outside the container to path p inside the container.

With Setup Command

Now we will examine the same process, only this time the Command Wrapper input references a Setup Command.

  1. The Command has a mount named input. The mount defines the path inside the container, which we will call p.
  2. The Command Wrapper has an input which provides files for mount input.
  3. That Command Wrapper input has a runtime value which we can resolve into some XNAT object. We find the path to that object’s files in the archive, which we will call A.
  4. The Command Wrapper input has a value for its "via-setup-command" property. We must now resolve that Setup Command.
  5. A Setup Command is created with two mounts, which have inside-the-container paths /input and /output, repsectively.
  6. We will mount the path A on the host to the the path /input in the setup container.
  7. We make a new writable directory B
  8. We will mount the path B on the host to the the path /output in the setup container.
  9. We will mount the path B on the host to the the path p in the main container.

Once the rest of the Command Resolution process is complete, we will create two containers:

  1. The setup container. This will have two mounts:
    1. The archive path A to /input
    2. A writable directory B to /output
  2. The main container, which has one mount: B to p.

In this way, whatever the setup container writes to its output mount is given to the main container through its input mount.

Though we create the setup and main containers at the same time, we only launch one: the setup container. Once the setup container is complete, the Container Service will launch the main container. If the setup container fails, the main container will not be launched, but will receive a status “Failed Setup”.

Creating a Setup Docker Image

The purpose of the setup image (and the setup containers created from it) is to take files in the XNAT archive format and copy / transform / convert them to whatever format they need to be. To fulfill this purpose, setup containers will always be given mounts at the same places: /input and /output. Inside the setup image, you can run whatever scripts you need to accomplish your task.

As an example, the setup image xnat/xnat2bids runs a script that takes files with BIDS metadata, but stored in the XNAT archive format, and moves them into the BIDS format. To do this it runs a python script, Here is this script (as of 2017-12-20; the current version may be different).

Code Block
#!/usr/bin/env python

Turn files in XNAT archive format into BIDS format.

Usage: <inputDir> <outputDir> (-h | --help) --version

    -h --help           Show the usage
    --version           Show the version
    <inputDir>          Directory with XNAT-archive-formatted files.
                        There should be scan directories, each having a NIFTI resource with NIFTI files, and
                        BIDS resources with BIDS sidecar JSON files.
    <outputDir>         Directory in which BIDS formatted files should be written.

import os
import sys
import json
import shutil
from glob import glob
from docopt import docopt

bidsAnatModalities = [T1w, T2w, T1rho, T1map, T2map, T2star, FLAIR, FLASH, PD, PDmap, PDT2, inplaneT1, inplaneT2, angio, defacemask, SWImagandphase]
bidsFuncModalities = [bold, physio, stim, sbref]
bidsDwiModalities = [dwi, dti]
bidsBehavioralModalities = [beh]
bidsFieldmapModalities = [phasemap, magnitude1]

class BidsScan(object):
    def __init__(self, scanId, bidsNameMap, *args):
        self.scanId = scanId
        self.bidsNameMap = bidsNameMap
        self.subject = bidsNameMap.get(sub)
        self.modality = bidsNameMap.get(modality)
        self.subDir = anat if self.modality in bidsAnatModalities else \
                      func if self.modality in bidsFuncModalities else \
                      dwi if self.modality in bidsDwiModalities else \
                      beh if self.modality in bidsBehavioralModalities else \
                      fmap if self.modality in bidsFieldmapModalities else \
        self.sourceFiles = list(args)

class BidsSession(object):
    def __init__(self, sessionLabel, bidsScans=[]):
        self.sessionLabel = sessionLabel
        self.bidsScans = bidsScans

class BidsSubject(object):
    def __init__(self, subjectLabel, bidsSession=None, bidsScans=[]):
        self.subjectLabel = subjectLabel
        if bidsSession:
            self.bidsSessions = [bidsSession]
            self.bidsScans = None
        if bidsScans:
            self.bidsScans = bidsScans
            self.bidsSessions = None

    def addBidsSession(self, bidsSession):
        if self.bidsScans:
            raise ValueError("Cannot add a BidsSession when the subject already has a list of BidsScans.")
        if not self.bidsSessions:
            self.bidsSessions = []

    def hasSessions(self):
        return bool(self.bidsSessions is not None and self.bidsSessions is not [])

    def hasScans(self):
        return bool(self.bidsScans is not None and self.bidsScans is not [])

def generateBidsNameMap(bidsFileName):

    # The BIDS file names will look like
    # sub-<participant_label>[_ses-<session_label>][_acq-<label>][_ce-<label>][_rec-<label>][_run-<index>][_mod-<label>]_<modality_label>
    # (that example is for anat. There may be other fields and labels in the other file types.)
    # So we split by underscores to get the individual field values.
    # However, some of the values may contain underscores themselves, so we have to check that each entry (save the last)
    #   contains a -.
    underscoreSplitListRaw = bidsFileName.split(_)
    underscoreSplitList = []

    for splitListEntryRaw in underscoreSplitListRaw[:-1]:
        if - not in splitListEntryRaw:
            underscoreSplitList[-1] = underscoreSplitList[-1] + splitListEntryRaw

    bidsNameMap = dict(splitListEntry.split(-) for splitListEntry in underscoreSplitList)
    bidsNameMap[modality] = underscoreSplitListRaw[-1]

    return bidsNameMap

def bidsifySession(sessionDir):
    print("Checking for session structure in " + sessionDir)

    sessionBidsJsonPath = os.path.join(sessionDir, RESOURCES, BIDS, dataset_description.json)

    scansDir = os.path.join(sessionDir, SCANS)
    if not os.path.exists(scansDir):
        # I guess we dont have any scans with BIDS data in this session
        print("STOPPING. Could not find SCANS directory.")

    print("Found SCANS directory. Checking scans for BIDS data.")

    bidsScans = []
    for scanId in os.listdir(scansDir):
        print("Checking scan {}.".format(scanId))

        scanDir = os.path.join(scansDir, scanId)
        scanBidsDir = os.path.join(scanDir, BIDS)
        scanNiftiDir = os.path.join(scanDir, NIFTI)

        if not os.path.exists(scanBidsDir):
            # This scan does not have BIDS data
            print("SKIPPING. Scan {} does not have a BIDS directory.".format(scanId))

        scanBidsJsonGlobList = glob(scanBidsDir + /*.json)
        if len(scanBidsJsonGlobList) != 1:
            # Something went wrong here. We should only have one JSON file in this directory.
            print("SKIPPING. Scan {} has {} JSON files in its BIDS directory. I expected to see one.".format(scanId, len(scanBidsJsonGlobList)))
            for jsonFile in scanBidsJsonGlobList:
        scanBidsJsonFilePath = scanBidsJsonGlobList[0]
        scanBidsJsonFileName = os.path.basename(scanBidsJsonFilePath)
        scanBidsFileName = scanBidsJsonFileName.rstrip(.json)
        scanBidsNameMap = generateBidsNameMap(scanBidsFileName)

        print("BIDS JSON file name: {}".format(scanBidsJsonFileName))
        print("Name map: {}".format(scanBidsNameMap))

        if not scanBidsNameMap.get(sub) or not scanBidsNameMap.get(modality):
            # Either sub or modality or both werent found. Something is wrong. Lets find out what.
            if not scanBidsNameMap.get(sub) and not scanBidsNameMap.get(modality):
                print("SKIPPING. Neither sub nor modality could be parsed from the BIDS JSON file name.")
            elif not scanBidsNameMap.get(sub):
                print("SKIPPING. Could not parse sub from the BIDS JSON file name.")
                print("SKIPPING. Could not parse modality from the BIDS JSON file name.")

        scanBidsDirFilePaths = glob(os.path.join(scanBidsDir, scanBidsFileName) + .*)
        scanNiftiDirFilePaths = glob(os.path.join(scanNiftiDir, scanBidsFileName) + .*)
        allFilePaths = scanBidsDirFilePaths + scanNiftiDirFilePaths

        bidsScan = BidsScan(scanId, scanBidsNameMap, *allFilePaths)
        if not bidsScan.subDir:
            print("SKIPPING. Could not determine subdirectory for modality {}.".format(bidsScan.modality))

        print("Done checking scan {}.".format(scanId))

    print("Done checking all scans.")
    return bidsScans

def getSubjectForBidsScans(bidsScanList):
    print("Finding subject for list of BIDS scans.")
    subjects = list({bidsScan.subject for bidsScan in bidsScanList if bidsScan.subject})

    if len(subjects) == 1:
        print("Found subject {}.".format(subjects[0]))
        return subjects[0]
    elif len(subjects) > 1:
        print("ERROR: Found more than one subject: {}.".format(", ".join(subjects)))
        print("ERROR: Found no subjects.")

    return None

def copyScanBidsFiles(destDirBase, bidsScanList):
    # First make all the "anat", "func", etc. subdirectories that we will need
    for subDir in {scan.subDir for scan in bidsScanList}:
        os.mkdir(os.path.join(destDirBase, subDir))

    # Now go through all the scans and copy their files into the correct subdirectory
    for scan in bidsScanList:
        destDir = os.path.join(destDirBase, scan.subDir)
        for f in scan.sourceFiles:
            shutil.copy(f, destDir)

version = "1.0"
args = docopt(__doc__, version=version)

inputDir = args[<inputDir>]
outputDir = args[<outputDir>]

print("Input dir: {}".format(inputDir))
print("Output dir: {}".format(outputDir))

# First check if the input directory is a session directory
sessionBidsScans = bidsifySession(inputDir)

bidsSubjectMap = {}
if sessionBidsScans:
    subject = getSubjectForBidsScans(sessionBidsScans)
    if not subject:
        # We would have already printed an error message, so no need to print anything here
    bidsSubjectMap = {subject: BidsSubject(subject, bidsScans=sessionBidsScans)}
    # Ok, we didnt find any BIDS scan directories in inputDir. We may be looking at a collection of session directories.
    print("Checking subdirectories of {}.".format(inputDir))

    for subSessionDir in os.listdir(inputDir):
        subSessionBidsScans = bidsifySession(os.path.join(inputDir, subSessionDir))
        if subSessionBidsScans:
            subject = getSubjectForBidsScans(subSessionBidsScans)
            if not subject:
                print("SKIPPING. Could not determine subject for session {}.".format(subSessionDir))

            print("Adding BIDS session {} to list for subject {}.".format(subSessionDir, subject))
            bidsSession = BidsSession(subSessionDir, subSessionBidsScans)
            if subject not in bidsSubjectMap:
                bidsSubjectMap[subject] = BidsSubject(subject, bidsSession=bidsSession)

            print("No BIDS data found in session {}.".format(subSessionDir))


if not bidsSubjectMap:
    print("No BIDS data found anywhere in inputDir {}.".format(inputDir))

allHaveSessions = True
allHaveScans = True
for bidsSubject in bidsSubjectMap.itervalues():
    allHaveSessions = allHaveSessions and bidsSubject.hasSessions()
    allHaveScans = allHaveScans and bidsSubject.hasScans()

if not (allHaveSessions ^ allHaveScans):
    print("ERROR: Somehow we have a mix of subjects with explicit sessions and subjects without explicit sessions. We must have either all subjects with sessions, or all subjects without. They cannot be mixed.")

print("Copying BIDS data.")
for bidsSubject in bidsSubjectMap.itervalues():
    subjectDir = os.path.join(outputDir, "sub-" + bidsSubject.subjectLabel)

    if allHaveSessions:
        for bidsSession in bidsSubject.bidsSessions:
            sessionDir = os.path.join(subjectDir, "ses-" + bidsSession.sessionLabel)
            copyScanBidsFiles(sessionDir, bidsSession.bidsScans)
        copyScanBidsFiles(subjectDir, bidsSubject.bidsScans)


Writing a Setup Command

The setup command is a command, and as such it must follow the command definition. However, setup commands cannot use the full set of features that most commands can. Setup commands cannot define any inputs, outputs, mounts, or wrappers. The mounts are always the same, /input and /output, so they need not be specified.

For a command to be recognized as a setup command, it must have the property

Code Block
"type": "docker-setup"

This is in contrast to standard commands, which have "type": "docker" (usually implicitly, since that is the default value).

The only properties that can be set in a setup command are

  • name (required)
  • description (optional, but recommended)
  • version (optional, but recommended)
  • type (required - must always have the value "docker-setup")
  • command-line (required)
  • working-directory (optional)

Here is the command JSON for the xnat/xnat2bids setup command (as of 2017-12-20; the current version may be different).

Code Block
    "name": "xnat2bids",
    "description": "xnat2bids setup command. Transforms an XNAT session with BIDS and NIFTI resources into BIDS format.",
    "version": "1.0",
    "type": "docker-setup",
    "command-line": " /input /output"

Referencing a Setup Command in a Main Command

This is a simple matter of setting one property. On any Command Wrapper Input (external or derived) that has a value for the property "provides-files-for-command-mount", you can set a value for the additional property "via-setup-command". This will tell the container service to take the files from the input, run them through the indicated setup command, and then give the resulting files to the container.

The value for the "via-setup-command" property should be in the docker image format: repo/image:version. So to reference the xnat2bids setup command, we could use the value xnat/xnat2bids:1.0 or xnat/xnat2bids:latest.

Additionally, we support one additional property in this value: the command name. This allows you to create one setup image which can have multiple commands. The full format of the "via-setup-command" property is repo/image:version:commandname.

For example, here is a snippet from the xnat/bids-mriqc command (as of 2017-12-20; the current version may be different). Here we can see how the command references the xnat2bids setup command in the xnat/xnat2bids-setup image.

Code Block
    "name": "bids-mriqc",
    "mounts": [
            "name": "in",
            "writable": "false",
            "path": "/input"
    "xnat": [
            "name": "bids-mriqc-session",
            "description": "Run the MRIQC BIDS App with a session mounted",
            "contexts": ["xnat:imageSessionData"],
            "external-inputs": [
                    "name": "session",
                    "description": "Input session",
                    "type": "Session",
                    "required": true,
                    "provides-files-for-command-mount": "in",
                    "via-setup-command": "xnat/xnat2bids-setup:1.0:xnat2bids"