Docker Environment Setup

Set up a Docker container environment for running Nextflow training examples. This skill handles all Docker configuration needed to match the Codespaces/Gitpod environment that learners use.

Initial Questions

Use AskUserQuestion to determine the setup type:

Which Docker setup do you need?

• Basic setup (Recommended) - For tutorials without containerized processes (e.g., hello_nextflow basics, plugin_development)
• DooD setup - For tutorials with containerized processes (e.g., genomics, essential_scripting_patterns)
• Check/restart existing - Verify or restart an existing nf-training container

---

Determine NXF_VER and training image

Before any Docker commands, read the Nextflow version and the training image reference from devcontainer.json:

NXF_VER=$(grep -o '"NXF_VER":\s*"[^"]*"' .devcontainer/devcontainer.json | cut -d'"' -f4)
TRAINING_IMAGE=$(grep -o '"image":\s*"[^"]*"' .devcontainer/devcontainer.json | cut -d'"' -f4)
echo "Using NXF_VER=${NXF_VER}, TRAINING_IMAGE=${TRAINING_IMAGE}"

Always use ${TRAINING_IMAGE} as read from the currently checked-out branch's devcontainer.json — do not hardcode ghcr.io/nextflow-io/training:latest. The tag a branch pins is the version that corresponds to what that branch's docs were written/tested against; pulling whatever the registry's latest tag happens to be right now can silently diverge from it. Pull that exact reference (docker pull "${TRAINING_IMAGE}") before starting the container, so a stale local cache doesn't ship an outdated image under the same tag name — but never substitute a different tag than what the branch declares.

---

Basic Setup

For tutorials that don't use containerized processes:

# Clean up any existing container
docker stop nf-training 2>/dev/null; docker rm nf-training 2>/dev/null

# Read the image reference pinned by this branch's devcontainer.json and pull it
NXF_VER=$(grep -o '"NXF_VER":\s*"[^"]*"' .devcontainer/devcontainer.json | cut -d'"' -f4)
TRAINING_IMAGE=$(grep -o '"image":\s*"[^"]*"' .devcontainer/devcontainer.json | cut -d'"' -f4)
docker pull "${TRAINING_IMAGE}"

# Start fresh container with UTF-8 locale support
docker run -d --name nf-training \
  -e NXF_VER=${NXF_VER} \
  -e LANG=C.UTF-8 \
  -e LC_ALL=C.UTF-8 \
  -v "${PWD}:/workspaces/training" \
  -w /workspaces/training \
  "${TRAINING_IMAGE}" \
  sleep infinity

Important: The LANG=C.UTF-8 and LC_ALL=C.UTF-8 environment variables are critical for handling non-ASCII characters (like "Holà", "Grüß Gott") in file names and content.

Running Commands

docker exec -e LANG=C.UTF-8 -e LC_ALL=C.UTF-8 \
  -w /workspaces/training/<working-dir> \
  nf-training \
  <command>

---

Docker-outside-of-Docker (DooD) Setup

For tutorials with containerized processes (FASTP, BWA, SAMTOOLS, etc.):

# Clean up any existing container
docker stop nf-training 2>/dev/null; docker rm nf-training 2>/dev/null

# Read the image reference pinned by this branch's devcontainer.json and pull it
NXF_VER=$(grep -o '"NXF_VER":\s*"[^"]*"' .devcontainer/devcontainer.json | cut -d'"' -f4)
TRAINING_IMAGE=$(grep -o '"image":\s*"[^"]*"' .devcontainer/devcontainer.json | cut -d'"' -f4)
HOST_PATH="${PWD}"
docker pull "${TRAINING_IMAGE}"

# Start container with DooD support
docker run -d --name nf-training \
  -e NXF_VER=${NXF_VER} \
  -e LANG=C.UTF-8 \
  -e LC_ALL=C.UTF-8 \
  -v /var/run/docker.sock:/var/run/docker.sock \
  -v "${HOST_PATH}:${HOST_PATH}" \
  -w "${HOST_PATH}" \
  "${TRAINING_IMAGE}" \
  sleep infinity

# Create symlink for Codespaces paths
docker exec nf-training bash -c "rm -rf /workspaces/training && mkdir -p /workspaces && ln -sf ${HOST_PATH} /workspaces/training"

Critical differences from basic setup:

1. Docker socket mount (-v /var/run/docker.sock:/var/run/docker.sock) - Allows Nextflow to spawn sibling containers
2. Matching host paths (-v "${HOST_PATH}:${HOST_PATH}") - Work directories resolve correctly between containers
3. Symlink - Makes /workspaces/training/... paths work locally

Running Commands with DooD

docker exec -e LANG=C.UTF-8 -e LC_ALL=C.UTF-8 -e USER=testuser \
  -w "${HOST_PATH}/<working-dir>" \
  nf-training \
  nextflow run <script.nf> [options]

When DooD is Needed

Any tutorial where processes specify containers:

• hello_nextflow (later lessons with containers)
• nf4_science/genomics and other domain modules
• Side quests: essential_scripting_patterns, metadata, etc.

---

Apple Silicon (ARM) Macs

Most bioinformatics containers are built for x86_64/amd64. On ARM Macs, create a platform config:

docker exec nf-training bash -c 'cat > /tmp/platform.config << EOF
docker.runOptions = "--platform linux/amd64"
EOF'

Include when running:

docker exec -e LANG=C.UTF-8 -e LC_ALL=C.UTF-8 -e USER=testuser \
  -w "${HOST_PATH}/<working-dir>" \
  nf-training \
  nextflow run <script.nf> -c /tmp/platform.config

Note: Platform emulation uses more memory. For OOM errors (exit code 137), increase Docker Desktop memory in Preferences → Resources.

---

Troubleshooting

Error	Cause	Solution
Cannot connect to Docker daemon	Socket not mounted	Add -v /var/run/docker.sock:/var/run/docker.sock
.command.sh: No such file or directory	Path mismatch	Use matching paths: -v "${HOST_PATH}:${HOST_PATH}"
exec format error	ARM/x86 mismatch	Add --platform linux/amd64 to docker.runOptions
Exit code 137 (OOM)	Insufficient memory	Increase Docker Desktop memory allocation
Malformed input or unmappable chars	Missing UTF-8	Add -e LANG=C.UTF-8 -e LC_ALL=C.UTF-8
Error: No such container: nf-training	Container stopped	Restart container (see below)

---

Container Restart Procedure

The container may stop during long sessions. To restart:

# 1. Check if container is running
docker ps | grep nf-training

# 2. If not running, restart with DooD setup
docker stop nf-training 2>/dev/null; docker rm nf-training 2>/dev/null

HOST_PATH="${PWD}"
NXF_VER=$(grep -o '"NXF_VER":\s*"[^"]*"' .devcontainer/devcontainer.json | cut -d'"' -f4)
TRAINING_IMAGE=$(grep -o '"image":\s*"[^"]*"' .devcontainer/devcontainer.json | cut -d'"' -f4)
docker pull "${TRAINING_IMAGE}"

docker run -d --name nf-training \
  -e NXF_VER=${NXF_VER} \
  -e LANG=C.UTF-8 \
  -e LC_ALL=C.UTF-8 \
  -v /var/run/docker.sock:/var/run/docker.sock \
  -v "${HOST_PATH}:${HOST_PATH}" \
  -w "${HOST_PATH}" \
  "${TRAINING_IMAGE}" \
  sleep infinity

# 3. Recreate symlink (critical!)
docker exec nf-training bash -c "rm -rf /workspaces/training && mkdir -p /workspaces && ln -sf ${HOST_PATH} /workspaces/training"

# 4. Recreate platform config if needed (ARM Macs)
docker exec nf-training bash -c 'cat > /tmp/platform.config << EOF
docker.runOptions = "--platform linux/amd64"
EOF'

---

Cleanup

When done with testing:

docker stop nf-training && docker rm nf-training

---

Notes

• Always verify you're in the repository root before starting (check for docs/en/mkdocs.yml)
• The container uses sleep infinity so it persists across multiple command executions
• Symlink must be recreated each time the container restarts
• For long sessions, periodically check container is still running: docker ps | grep nf-training
• The docker run/docker exec commands above only set NXF_VER, LANG, and LC_ALL inside the container, so it does not inherit the host shell's CLAUDECODE variable and Nextflow 26.04+'s agent-mode console output is not triggered here. See Console Output Mode in repo-conventions — do not add -e CLAUDECODE or otherwise forward the host environment into this container, as that would reintroduce the problem.