run-tests

name	run-tests
description	Validate code changes by intelligently selecting and running the appropriate test suites. Use this when editing code to verify changes work correctly, run tests, validate functionality, or check for regressions. Automatically discovers affected test suites, selects the minimal set of venvs needed for validation, and handles test execution with Docker services as needed.
allowed-tools	["Bash","Read","Grep","Glob","TodoWrite"]

Test Suite Validation Skill

This skill helps you efficiently validate code changes by running the appropriate subset of the test suite. It uses scripts/run-tests to intelligently discover affected tests and run only what's necessary for validation.

When to Use This Skill

Use this skill when you have:

• Made changes to source code files and want to validate they work
• Fixed a bug and want to verify the fix
• Added a feature and need test coverage
• Modified test infrastructure or configuration
• Want to verify changes don't break existing functionality

Key Principles

1. Always use the run-tests skill when testing code changes - it's optimized for intelligent suite discovery
2. Never run pytest directly - bypasses the project's test infrastructure (use scripts/run-tests or riot via scripts/ddtest)
3. Minimal venvs for iteration - run 1-2 venvs initially, expand only if needed
4. Use --dry-run first - see what would run before executing
5. Follow official docs - docs/contributing-testing.rst is the source of truth for testing procedures

How This Skill Works

Step 1: Identify Changed Files

First, determine which files were modified:

• If you have pending edits, I'll identify the changed files from the current session
• I'll look at git status to find staged, unstaged, and untracked changes
• You can also specify files explicitly if working on specific changes

Step 2: Discover Available Test Suites

I'll use the scripts/run-tests script to discover what test suites match your changes:

scripts/run-tests --list <edited-files>

This outputs JSON showing:

• Available test suites that match your changed files
• All venvs (Python versions + package combinations) available for each suite
• Their hashes, Python versions, and package versions

Step 3: Intelligently Select Venvs

Rather than running ALL available venvs (which could take hours), I'll select the minimal set needed to validate your changes:

For Core/Tracing Changes (Broad Impact)

When you modify files like:

• ddtrace/internal/core/*, ddtrace/_trace/*, ddtrace/trace/*
• ddtrace/_monkey.py, ddtrace/settings/*
• ddtrace/constants.py

Strategy: Run core tracer + internal tests with 1 venv each

• Example: tracer suite with latest Python + internal suite with latest Python
• This validates broad-reaching changes without excessive overhead
• Skip integration suites unless the change directly affects integration code

For Integration/Contrib Changes (Targeted Impact)

When you modify files like:

• ddtrace/contrib/flask/*, ddtrace/contrib/django/*, etc.
• ddtrace/contrib/*/patch.py or integration-specific code

Strategy: Run ONLY the affected integration suite with 1-2 venvs

• Example: Flask changes → run contrib::flask suite with latest Python
• If change involves multiple versions (e.g., Django 3.x and 4.x), pick 1 venv per major version
• Skip unrelated integrations

For Test-Only Changes

When you modify tests/ files (but not test infrastructure):

• Run only the specific test files/functions modified
• Use pytest args with two separators: -- -- -k test_name (first -- ends scripts/run-tests parsing and starts riot args; second -- tells riot to forward remaining args to pytest), or direct pytest test paths (e.g., -- -- tests/contrib/flask/test_views.py)

For Test Infrastructure Changes

When you modify:

• tests/conftest.py, tests/suitespec.yml, scripts/run-tests, riotfile.py

Strategy: Run a quick smoke test suite

• Example: internal suite with 1 venv as a sanity check
• Or run small existing test suites to verify harness changes

Step 4: Execute Selected Venvs

I'll run the selected venvs. On the first invocation in a session, always run without -s to ensure the venv has dd-trace-py properly installed:

scripts/run-tests --venv <hash1> --venv <hash2>

On subsequent runs, use -s (riot's --skip-base-install flag, not to be confused with pytest's -s) to skip rebuilding dd-trace-py and save significant time:

scripts/run-tests --venv <hash1> --venv <hash2> -- -s

When to use -s (skip base install) on subsequent runs:

• Only Python files were modified (no native code changes)
• Iterating on test fixes within the same session
• Re-running tests after small code tweaks

When to omit -s even on subsequent runs (force rebuild):

• After merging or rebasing from main (dependencies or native code may have changed)
• C extensions, Cython (.pyx, .pxd), or CMake files were modified (e.g., under ddtrace/internal/, ddtrace/appsec/_iast/_taint_tracking/, src/native/)
• setup.py, pyproject.toml, or setup.cfg were modified
• riotfile.py or .riot/requirements/ files were modified

This will:

• Start required Docker services (redis, postgres, etc.)
• Run tests in the specified venvs sequentially
• Stop services after completion
• Show real-time output and status

Step 5: Handle Results

If tests pass: ✅ Your changes are validated!

If tests fail: 🔴 I'll:

• Show you the failure details
• Identify which venv failed
• Ask clarifying questions to understand the issue
• Offer to run specific failing tests with more verbosity
• Help iterate on fixes and re-run

For re-running specific tests (use -s since the venv is already built):

scripts/run-tests --venv <hash> -- -s -- -vv -k test_name

When Tests Fail

When you encounter test failures, follow this systematic approach:

1. Review the failure details carefully - Don't just skim the error, understand what's actually failing
2. Understand what's failing - Don't blindly re-run; analyze the root cause
3. Make code changes - Fix the underlying issue
4. Re-run with more verbosity if needed - Use -vv or -vvv for detailed output
5. Iterate until tests pass - Repeat the process with each fix

Venv Selection Strategy in Detail

Understanding Venv Hashes

From scripts/run-tests --list, you'll see output like:

{
  "suites": [
    {
      "name": "tracer",
      "venvs": [
        {
          "hash": "abc123",
          "python_version": "3.8",
          "packages": "..."
        },
        {
          "hash": "def456",
          "python_version": "3.11",
          "packages": "..."
        }
      ]
    }
  ]
}

Selection Rules

1. Latest Python version is your default choice

   • Unless your change specifically targets an older Python version
   • Example: if fixing Python 3.8 compatibility, also test 3.8

2. One venv per suite is usually enough for iteration

   • Only run multiple venvs per suite if:
     • Change impacts multiple Python versions differently
     • Testing package compatibility variations (e.g., Django 3.x vs 4.x)
     • Initial validation passed and you want broader coverage

3. Minimize total venvs

   • 1-2 venvs total for small targeted changes
   • 3-4 venvs maximum for broader changes
   • Never run 10+ venvs for initial validation (save that for CI)

4. Consider test runtime

   • Each venv can take 5-30 minutes depending on suite
   • With 2 venvs you're looking at 10-60 minutes for iteration
   • With 5 venvs you're looking at 25-150 minutes
   • Scale appropriately for your patience and deadline

Using --venv Directly

When you have a specific venv hash you want to run, you can use it directly without specifying file paths:

scripts/run-tests --venv e06abee

The --venv flag automatically searches all available venvs across all suites, so it works regardless of what files you have locally changed. This is useful when:

• You know exactly which venv you want to test
• You have unrelated local changes that would otherwise limit suite matching
• You want to quickly re-run a specific venv without file path arguments

Examples

Example 1: Fixing a Flask Integration Bug

Changed file: ddtrace/contrib/internal/flask/patch.py

scripts/run-tests --list ddtrace/contrib/internal/flask/patch.py
# Output shows: contrib::flask suite available

# Select output (latest Python):
# Suite: contrib::flask
# Venv: hash=e06abee, Python 3.13, flask

# First run: no -s to ensure venv is properly set up
scripts/run-tests --venv e06abee

# Subsequent runs: use -s since only Python files changed
scripts/run-tests --venv e06abee -- -s

Example 2: Fixing a Core Tracing Issue

Changed file: ddtrace/_trace/tracer.py

scripts/run-tests --list ddtrace/_trace/tracer.py
# Output shows: tracer suite, internal suite available

# Select strategy:
# - tracer: latest Python (e.g., abc123)
# - internal: latest Python (e.g., def456)

# First run: no -s
scripts/run-tests --venv abc123 --venv def456

# Subsequent runs: use -s since only Python files changed
scripts/run-tests --venv abc123 --venv def456 -- -s

Example 3: Fixing a Test-Specific Bug

Changed file: tests/contrib/flask/test_views.py

scripts/run-tests --list tests/contrib/flask/test_views.py
# Output shows: contrib::flask suite

# First run: no -s
scripts/run-tests --venv flask_py311 -- -- -vv tests/contrib/flask/test_views.py

# Subsequent runs: use -s to skip rebuild
scripts/run-tests --venv flask_py311 -- -s -- -vv tests/contrib/flask/test_views.py

Example 4: Iterating on a Failing Test

After the first run shows a test failing, use -s to iterate quickly:

scripts/run-tests --venv flask_py311 -- -s -- -vv -k test_view_called_twice
# Focused on the specific failing test with verbose output

Best Practices

DO ✅

• Use -s on subsequent runs: After the first run builds the venv, pass -- -s to skip rebuild when only Python files changed
• Start small: Run 1 venv first, expand only if needed
• Be specific: Use pytest -k filter when re-running failures
• Check git: Verify you're testing the right files with git status
• Read errors: Take time to understand test failures before re-running
• Ask for help: When unclear what tests to run, ask me to analyze the changes

DON'T ❌

• Use -s after merging from main: Native code or dependencies may have changed, requiring a rebuild
• Use -s when C/Cython/CMake files changed: Native extensions must be recompiled
• Run all venvs initially: That's what CI is for
• Skip the minimal set guidance: It's designed to save you time
• Ignore service requirements: Some suites need Docker services up
• Run tests without changes saved: Make sure edits are saved first
• Iterate blindly: Understand what's failing before re-running

Additional Testing Resources

For comprehensive testing guidance, refer to the contributing documentation:

• docs/contributing-testing.rst - Detailed testing guidelines

  • What kind of tests to write (unit tests, integration tests, e2e tests)
  • When to write tests (feature development, bug fixes)
  • Where to put tests in the repository
  • Prerequisites (Docker, uv)
  • Complete scripts/run-tests usage examples
  • Riot environment management details
  • Running specific test files and functions
  • Test debugging strategies

• docs/contributing.rst - PR and testing requirements

  • All changes need tests or documented testing strategy
  • How tests fit into the PR review process
  • Testing expectations for different types of changes

• docs/contributing-design.rst - Test architecture context

  • How products, integrations, and core interact
  • Where different types of tests should live
  • Testing patterns for each library component

When to reference these docs:

• First time writing tests for this project → Read contributing-testing.rst
• Understanding test requirements for PRs → Read contributing.rst
• Need context on test architecture → Read contributing-design.rst

Troubleshooting

Docker services won't start

# Manually check/stop services:
docker compose ps
docker compose down

Can't find matching suites

• Verify the file path is correct
• Check tests/suitespec.yml to understand suite patterns
• Your file might not be covered by any suite pattern yet

Test takes too long

• You may have selected too many venvs
• Try running with just 1 venv
• Use pytest -k to run subset of tests

Technical Details

Architecture

The scripts/run-tests system:

• Maps source files to test suites using patterns in tests/suitespec.yml
• Uses riot to manage multiple Python/package combinations as venvs
• Each venv is a self-contained environment
• Docker services are managed per suite lifecycle
• Use -- <riot args> -- <pytest args> for mixed passthrough. The first -- is consumed by scripts/run-tests, and the second -- is consumed by riot before forwarding remaining args to pytest. If you only need pytest args, use -- -- <pytest args>.

Supported Suite Types

Primary suites for validation:

• tracer: Core tracing functionality tests
• internal: Internal component tests
• contrib::*: Integration with specific libraries (flask, django, etc.)
• integration_*: Cross-library integration scenarios
• Specialized: telemetry, profiling, appsec, llmobs, etc.

Environment Variables

Some suites require environment setup:

• DD_TRACE_AGENT_URL: For snapshot-based tests
• Service-specific variables for Docker containers
• These are handled automatically by the script

Resource Limiting (for testing under constrained environments)

You can limit CPU and memory resources to simulate resource-constrained CI environments where multiple jobs run in parallel. This helps reproduce flaky tests that fail due to timing issues, race conditions, or resource exhaustion.

Environment Variables:

• DD_TEST_CPUS: CPU limit (e.g., 0.25, 0.5, 1.0, 2.0)
• DD_TEST_MEMORY: Memory limit with unit (e.g., 512m, 1g, 2g)

Usage:

# Run tests with resource constraints
DD_TEST_CPUS=0.5 DD_TEST_MEMORY=1g scripts/run-tests --venv <hash>

# Run specific test file with heavy constraints
DD_TEST_CPUS=0.25 DD_TEST_MEMORY=1g scripts/run-tests tests/path/to/test.py

# Multiple runs to catch intermittent failures
for i in {1..10}; do
  DD_TEST_CPUS=0.5 DD_TEST_MEMORY=1g scripts/run-tests --venv <hash> -- --randomly-seed=$RANDOM
done

Recommended Resource Limits:

• Moderate Load (Typical Shared CI):

  DD_TEST_CPUS=2.0 DD_TEST_MEMORY=4g
  

  Simulates a CI runner with some other jobs running. Good for initial testing.

• Heavy Load (Busy CI Server):

  DD_TEST_CPUS=1.0 DD_TEST_MEMORY=2g
  

  Simulates a heavily loaded CI server with many concurrent jobs. Recommended starting point for reproducing flaky tests.

• Extreme Load (Stress Testing):

  DD_TEST_CPUS=0.5 DD_TEST_MEMORY=1g
  

  Simulates extreme resource contention. Good for surfacing timing issues and race conditions.

• Critical Failure Conditions:

  DD_TEST_CPUS=0.25 DD_TEST_MEMORY=512m
  

  Forces maximum resource pressure. Use this to find the breaking point or reproduce worst-case scenarios.

When to Use Resource Limits:

1. Investigating flaky tests - Tests that pass locally but fail in CI
2. Timing-sensitive tests - Tests involving async operations, network calls, or multiprocessing
3. Resource exhaustion issues - Tests that fail under memory pressure or CPU throttling
4. Race condition detection - Slower execution can expose timing bugs
5. Reproducing CI failures - When you have a seed that failed in CI

Verifying Limits Are Applied:

# Check configuration before running
DD_TEST_CPUS=0.5 DD_TEST_MEMORY=1g docker compose config | grep -A 5 testrunner

# Monitor actual resource usage during test run (in another terminal)
docker stats

Example: Testing a Flaky Test

# Run a known flaky test 20 times with resource constraints
DD_TEST_CPUS=0.5 DD_TEST_MEMORY=1g scripts/run-tests \
  tests/appsec/integrations/flask_tests/test_iast_flask_testagent.py::test_iast_unvalidated_redirect \
  -- --count=20