play_launch

Resolve, run, and analyze ROS 2 launch executions with resource monitoring and interactive management.

Upgrading? CHANGELOG.md lists the breaking changes — 0.9.0 has several, and a few of them change behaviour rather than erroring.

Installation

Requires Ubuntu 22.04+ with ROS 2 (Humble, Jazzy, Kilted, Lyrical or Rolling), Python 3.10+, and:

sudo apt install libz3-dev
sudo apt install ros-${ROS_DISTRO}-rclcpp-components ros-${ROS_DISTRO}-class-loader

Install from PyPI:

pip install play_launch

One wheel serves every supported ROS distribution that shares your Python (Humble on 3.10; Jazzy and Kilted on 3.12; Lyrical and Rolling on 3.14), for x86_64 and aarch64. It uses the distribution you sourced (ROS_DISTRO), or the only one under /opt/ros when nothing is sourced; PLAY_LAUNCH_ROS_DISTRO overrides both, and python3 -m play_launch.payload says which one it picked and why. On Ubuntu 24.04 and later add --break-system-packages (or use a virtual environment with --system-site-packages, so ROS's Python packages stay visible).

This installs the play_launch command, which carries the whole surface: launch, run, up, resolve, dump, check, plot, contract, context, setcap, verify, measure.

Optional: grant the helper binaries the capabilities they need for per-process I/O monitoring and non-root RT scheduling (requires sudo):

play_launch setcap

This enables per-process I/O monitoring and non-root RT scheduling (--sched): play_launch setcap grants CAP_SYS_PTRACE to the I/O helper and CAP_SYS_NICE to the RT helper — both are small ROS-free binaries. The main play_launch binary is never given a file capability — that would put it in secure-execution mode and stop it finding its ROS libraries.

See the Real-Time Scheduling Guide for the scheduling spec format and how it maps onto processes, threads, and the kernel.

Quick Start

Launch any ROS 2 package with monitoring and Web UI enabled by default:

play_launch launch demo_nodes_cpp topics/talker_listener.launch.py

Access Web UI at http://127.0.0.1:8080 for real-time node management and log streaming.

The Rust parser is used by default for speed. For maximum compatibility, use --parser python.

Usage

Launch Files

Replace ros2 launch with play_launch launch:

play_launch launch <package> <launch_file> [arguments...]

Single Nodes

Replace ros2 run with play_launch run:

play_launch run <package> <executable> [arguments...]

Two-Step Workflow

Resolve once, bring the system up as often as you like:

# Resolve once — writes system_model.yaml
play_launch dump launch <package> <launch_file> [arguments...]

# Bring it up, as often as you like
play_launch up system_model.yaml

up takes the model path positionally or as --model <path>; it is required either way, since the model is the only thing up spawns from.

Checking Contracts

check parses a launch file and checks the contracts shipped beside it (<stem>.contract.yaml). It needs no ROS graph and is meant for CI:

play_launch check my_pkg bringup.launch.xml               # 0 clean, 1 errors, 3 a contract refused
play_launch check bringup.launch.xml --expect ladder-rung-budget   # a negative test
play_launch check bringup.launch.xml --explain --width 100         # the fault-reaction table, wrapped
play_launch --version   # play_launch 0.13.0 (v0.13.0-9-gabc1234, rlm v0.1.47)

Exit status 3 means a contract file did not parse (error[manifest-parse]), so it was not checked at all; the refusal names the checker's version and the grammar (rlm tag) it reads. Through a pipe the output is ASCII without colour (NO_COLOR is honoured on a terminal too). Exit codes, --expect, the JSON report and the --explain table: Checking contracts in CI.

Features

All features enabled by default:

• Resource monitoring: CPU, memory, I/O, GPU (2s interval)
• Diagnostic monitoring: the /diagnostics topic (add more via diagnostics.topics)
• Web UI: Interactive management at http://127.0.0.1:8080
• Observable containers: Composable nodes share their container's process, as under ros2 launch, with ComponentEvent load reporting (default; --container-mode isolated fork+execs each composable instead)

Disable Features

Disable specific features:

play_launch launch <package> <launch_file> --disable-monitoring
play_launch launch <package> <launch_file> --disable-diagnostics
play_launch launch <package> <launch_file> --disable-web-ui
play_launch launch <package> <launch_file> --disable-all

Container Mode

Control how composable nodes are managed (default: observable):

# Observable: ComponentEvent publishing, shared process -- as `ros2 launch` runs it (default)
play_launch launch <pkg> <file> --container-mode observable

# Isolated: fork+exec per-node process isolation (a crash takes down one node)
play_launch launch <pkg> <file> --container-mode isolated

# Stock: use original container from launch file, no override
play_launch launch <pkg> <file> --container-mode stock

Adjust Monitoring

Change sampling interval (default: 2000ms):

play_launch launch <package> <launch_file> --monitor-interval-ms 500

Configure Web UI

Change address or port (default: 127.0.0.1:8080):

play_launch launch <package> <launch_file> --web-addr 0.0.0.0:8080

Configuration File

Use YAML for advanced control:

# config.yaml
monitoring:
  enabled: true
  sample_interval_ms: 2000

processes:
  - node_pattern: "NODE 'rclcpp_components/component_container*"
    cpu_affinity: [0, 1]
    nice: 5

Apply configuration:

play_launch launch <package> <launch_file> --config config.yaml
play_launch up system_model.yaml --config config.yaml

Visualization

Generate interactive plots from monitoring data:

# Plot latest execution
play_launch plot

# Plot specific log directory
play_launch plot --log-dir play_log/2025-10-28_16-17-56

# Plot specific metrics
play_launch plot --metrics cpu --metrics memory

# List available metrics
play_launch plot --list-metrics

Output saved to play_log/<timestamp>/plot/:

• cpu_timeline.html - CPU usage over time
• memory_timeline.html - Memory usage over time
• io_timeline.html - I/O read/write rates
• cpu_distribution.html - CPU distribution box plot
• memory_distribution.html - Memory distribution box plot
• statistics.txt - Top 10 rankings for all metrics

Web UI Features

• Node management: Start/Stop/Restart individual or all nodes
• Container controls: Load/Unload composable nodes
• Real-time logs: Stream stdout/stderr with log level coloring and filtering
• Diagnostics panel: View /diagnostics messages with level filtering
• Status monitoring: Color-coded node states
• Auto-restart: Per-node automatic restart configuration
• Search & filter: Find nodes in large deployments

Output Structure

play_log/<timestamp>/
├── node/<node_name>/
│   ├── metadata.json
│   ├── metrics.csv       # Resource metrics (when enabled)
│   ├── out/err           # Process logs
│   ├── pid/status/cmdline
│   └── params_files/     # ROS parameter files
├── load_node/<name>/
│   └── out/err           # Per-composable-node logs (isolated mode)
├── system_stats.csv      # System-wide metrics
├── diagnostics.csv       # Diagnostic messages (when enabled)
└── plot/                 # Generated visualizations

Command Reference

# Launch (all features enabled by default)
play_launch launch <package> <launch_file> [args...]
play_launch run <package> <executable> [args...]

# Dump and up (dump emits the SystemModel by default — one artifact)
play_launch dump launch <package> <launch_file> [args...]
play_launch up system_model.yaml            # or: play_launch up --model system_model.yaml

# Parser selection (Rust is default)
play_launch launch <pkg> <file> --parser rust     # Default, fast
play_launch launch <pkg> <file> --parser python   # Maximum compatibility

# Container mode
play_launch launch <pkg> <file> --container-mode observable  # Default
play_launch launch <pkg> <file> --container-mode isolated
play_launch launch <pkg> <file> --container-mode stock

# Disable features
play_launch launch <pkg> <file> --disable-monitoring
play_launch launch <pkg> <file> --disable-diagnostics
play_launch launch <pkg> <file> --disable-web-ui
play_launch launch <pkg> <file> --disable-all
play_launch launch <pkg> <file> --disable-respawn

# Reload a composable node whose process crashed (default: off)
play_launch launch <pkg> <file> --composable-respawn on-crash  # always
play_launch launch <pkg> <file> --composable-respawn inherit   # only under a respawn="true" container

# Enable only specific features
play_launch launch <pkg> <file> --enable monitoring
play_launch launch <pkg> <file> --enable web-ui --enable diagnostics

# Adjust settings
play_launch launch <pkg> <file> --monitor-interval-ms 500
play_launch launch <pkg> <file> --web-addr 0.0.0.0:8080
play_launch launch <pkg> <file> --config config.yaml

# Logging
play_launch launch <pkg> <file> --verbose              # Enable INFO level
RUST_LOG=play_launch=debug play_launch launch <pkg> <file>  # DEBUG level

# Contracts
play_launch check <pkg> <file> [args...]           # exit 0 / 1 errors / 3 refused
play_launch check <pkg> <file> --expect <rule-id>  # pass only on exactly these errors
play_launch check <pkg> <file> --format json       # diagnostics as JSON on stdout
play_launch check <pkg> <file> --explain --ascii --width 120

# Visualization
play_launch plot
play_launch plot --log-dir <dir>
play_launch plot --metrics cpu --metrics memory --metrics io --metrics gpu
play_launch plot --list-metrics

Development

Prerequisites

• ROS 2 (Humble on Ubuntu 22.04; Jazzy or Kilted on 24.04; Lyrical or Rolling on 26.04), sourced: the build targets $ROS_DISTRO
• libclang (libclang-dev): rcl_interception_sys generates its bindings from the ROS headers
• Rust toolchain (stable)
• Python 3.10+
• just command runner
• cargo-nextest test runner
• uv Python package manager

Install Dependencies

just install-deps

This installs colcon-cargo-ros2 and runs rosdep install for all ROS dependencies.

Build

just build          # Full build: colcon + interception .so + wheel
just build-cpp      # C++ only (msgs + container)
just build-rust     # Rust only (assumes C++ install/ exists)
just build-wheel    # Bundle + wheel only (no colcon rebuild)

Optional: grant the I/O helper its monitoring capability and the RT helper its scheduling capability (requires sudo). After this, RT scheduling (--sched) works without root:

play_launch setcap

Re-run it after every rebuild or upgrade. A file capability is bound to the exact contents of a binary, so replacing that binary drops it — deliberately, since carrying privilege across a change would grant it to code nobody vetted. play_launch says so when it happens, naming the grant it is replacing rather than telling you to run a command you already ran.

(A just setcap recipe also exists. It is a DEVELOPER convenience that avoids the password prompt via a throwaway container, it needs a source checkout, and it does not work under rootless Docker. play_launch setcap is the supported path.)

Run

just run launch <package> <launch_file> [arguments...]
just run run <package> <executable> [arguments...]

Test

just test               # Parser unit + fast integration (~3s)
just test-all           # Parser unit + all integration (~30s)
just test-unit          # Parser unit tests only
just test-integration   # All integration tests
just test-autoware      # Autoware integration tests

Code Quality

just check      # clippy + rustfmt + ruff + cpplint + clang-format
just format     # Auto-format Rust + Python + C++
just quality    # Format then check

Install Wheel Locally

just install-wheel

License

MIT License. See LICENSE.txt.