Buck2 Local Resources Skill

Overview

This skill guides you through creating Buck2 tests that depend on external processes or services. Buck2's local resources pattern manages broker processes automatically: starting them before tests, providing connection info via environment variables, and cleaning up afterward.

When to Use This Skill

Use this skill when:

• Creating integration tests that need databases (PostgreSQL, Redis, SQLite)
• Testing HTTP/REST APIs (need HTTP server running)
• Testing message queues (Kafka, RabbitMQ, NATS)
• Testing Unix socket communication
• Testing service orchestration (multiple services interacting)
• Any test requiring external processes that Buck2 should manage

Don't use for:

• Simple unit tests (no external dependencies)
• Tests with in-memory alternatives (prefer in-memory DBs)
• Heavy services like Docker-in-Docker (too slow for Buck2 tests)

Core Concepts

The Three Components

Every local resources test has three parts:

1. Broker Script: Shell script that starts the service and outputs JSON
2. Broker Rule: Buck2 rule providing LocalResourceInfo
3. Test Rule: Buck2 rule using ExternalRunnerTestInfo to consume the resource

How It Works

Test execution:
1. Buck2 identifies test needs broker (via local_resources)
2. Buck2 runs broker script → service starts, JSON output
3. Buck2 parses JSON, sets environment variables
4. Buck2 runs test script with those env vars
5. Test completes → Buck2 kills broker process (automatic cleanup)

Step-by-Step Implementation Guide

Step 1: Identify Resource Requirements

Before writing code, answer:

1. What service? (HTTP server, database, message queue, etc.)
2. What connection info? (port, URL, socket path, credentials?)
3. How to start it? (command to run, config needed?)
4. How to detect readiness? (health check, file exists, port open?)
5. Single or multiple resources? (one service or several?)

Example answers:

• Service: HTTP server
• Connection: Port number and URL
• Start: python3 -m http.server <port>
• Readiness: Port responds to HTTP request
• Resources: Single (just HTTP)

Step 2: Create Broker Script

Template:

#!/usr/bin/env bash
# SPDX-FileCopyrightText: © 2024-2026 Austin Seipp
# SPDX-License-Identifier: Apache-2.0

set -euo pipefail

# 1. Setup: Create temp directory for isolation
TMPDIR=${TMPDIR:-/tmp}
WORKDIR="$TMPDIR/my-service-$$"  # $$ = process ID for uniqueness
mkdir -p "$WORKDIR"

# 2. Start service in background
<command-to-start-service> &
SERVICE_PID=$!

# 3. Wait for service to be ready (CRITICAL!)
for i in {1..50}; do
    if <readiness-check>; then
        break
    fi
    sleep 0.1
done

# 4. Output JSON (ONLY this to stdout, everything else to stderr!)
echo "{\"pid\": $SERVICE_PID, \"resources\": [{\"<key1>\": \"<value1>\", \"<key2>\": \"<value2>\"}]}"

Important:

• Service MUST run in background (&)
• MUST output JSON to stdout
• JSON MUST include pid (Buck2 tracks this to kill process)
• JSON MUST include resources array with connection details
• Resource keys will map to environment variables
• ALL other output must go to stderr (>&2) or /dev/null

Examples:

#!/usr/bin/env bash
set -euo pipefail

TMPDIR=${TMPDIR:-/tmp}
WORKDIR="$TMPDIR/http-$$"
mkdir -p "$WORKDIR"

# Create test content
echo "Hello" > "$WORKDIR/index.html"

# Start server
cd "$WORKDIR"
python3 -m http.server 8080 > /dev/null 2>&1 &
PID=$!

# Wait for ready
for i in {1..30}; do
    if curl -s http://localhost:8080 > /dev/null 2>&1; then
        break
    fi
    sleep 0.1
done

echo "{\"pid\": $PID, \"resources\": [{\"port\": \"8080\", \"url\": \"http://localhost:8080\"}]}"

#!/usr/bin/env bash
set -euo pipefail

TMPDIR=${TMPDIR:-/tmp}
SOCKET="$TMPDIR/my-socket-$$"

# Start socket server (using socat or custom server)
socat UNIX-LISTEN:$SOCKET,fork EXEC:'/bin/cat' &
PID=$!

# Wait for socket file to exist
for i in {1..50}; do
    if [[ -S "$SOCKET" ]]; then
        break
    fi
    sleep 0.1
done

echo "{\"pid\": $PID, \"resources\": [{\"socket_path\": \"$SOCKET\"}]}"

#!/usr/bin/env bash
set -euo pipefail

TMPDIR=${TMPDIR:-/tmp}
DB_PATH="$TMPDIR/test-db-$$.sqlite"

# Initialize database
sqlite3 "$DB_PATH" "CREATE TABLE test (id INTEGER PRIMARY KEY, value TEXT);"

# No background process needed for SQLite, but we still need a PID
# Use sleep as a dummy process to keep broker alive
sleep infinity &
PID=$!

echo "{\"pid\": $PID, \"resources\": [{\"db_path\": \"$DB_PATH\"}]}"

Step 3: Create Broker Rule (in defs.bzl)

Template:

# SPDX-FileCopyrightText: © 2024-2026 Austin Seipp
# SPDX-License-Identifier: Apache-2.0

_<service>_broker_rule = rule(
    impl = lambda ctx: [
        DefaultInfo(),
        RunInfo(args = cmd_args(ctx.attrs._script[DefaultInfo].default_outputs[0])),
        LocalResourceInfo(
            setup = cmd_args(ctx.attrs._script[DefaultInfo].default_outputs[0]),
            resource_env_vars = {
                "<ENV_VAR_NAME>": "<json_key>",  # Map JSON key → env var
                # Add more mappings as needed
            },
        ),
    ],
    attrs = {
        "_script": attrs.default_only(
            attrs.exec_dep(default = "<cell>//<package>:<broker-script-target>")
        ),
    },
)

Key points:

• LocalResourceInfo is the provider that marks this as a broker
• setup: The command to run (the broker script)
• resource_env_vars: Dict mapping env var names (keys) to JSON keys (values)
  • Example: {"HTTP_URL": "url"} means JSON's resources[0].url → test's $HTTP_URL
• _script: Reference to the broker script target (defined in BUILD file)

Example:

_http_broker_rule = rule(
    impl = lambda ctx: [
        DefaultInfo(),
        RunInfo(args = cmd_args(ctx.attrs._script[DefaultInfo].default_outputs[0])),
        LocalResourceInfo(
            setup = cmd_args(ctx.attrs._script[DefaultInfo].default_outputs[0]),
            resource_env_vars = {
                "HTTP_PORT": "port",
                "HTTP_URL": "url",
            },
        ),
    ],
    attrs = {
        "_script": attrs.default_only(
            attrs.exec_dep(default = "depot//my/package:http-broker-script")
        ),
    },
)

Step 4: Create Test Script

Your test script will receive environment variables from the broker:

#!/usr/bin/env bash
# SPDX-FileCopyrightText: © 2024-2026 Austin Seipp
# SPDX-License-Identifier: Apache-2.0

set -euo pipefail

# Environment variables are automatically set by Buck2
echo "Service available at: $<ENV_VAR>"

# Run your tests
<test-commands-here>

# Exit with appropriate code
if <tests-passed>; then
    echo "✓ Tests passed"
    exit 0
else
    echo "✗ Tests failed"
    exit 1
fi

Example:

#!/usr/bin/env bash
set -euo pipefail

echo "Testing HTTP server at $HTTP_URL"

# Make request
RESPONSE=$(curl -s "$HTTP_URL/index.html")

if [[ "$RESPONSE" == "Hello" ]]; then
    echo "✓ Test passed"
    exit 0
else
    echo "✗ Test failed"
    exit 1
fi

Step 5: Create Test Rule (in defs.bzl)

Template:

_<service>_test_rule = rule(
    impl = lambda ctx: [
        DefaultInfo(),
        RunInfo(args = cmd_args([ctx.attrs.script[DefaultInfo].default_outputs[0]])),
        ExternalRunnerTestInfo(
            type = "custom",
            command = [cmd_args([ctx.attrs.script[DefaultInfo].default_outputs[0]])],
            local_resources = {
                '<resource-name>': ctx.attrs.<broker_attr>.label,
                # Add more resources if needed
            },
            required_local_resources = [
                RequiredTestLocalResource("<resource-name>", listing = False, execution = True),
                # Add more if needed
            ],
        ),
    ],
    attrs = {
        "script": attrs.source(),
        "<broker_attr>": attrs.exec_dep(providers = [LocalResourceInfo]),
        # Add more broker attrs for multiple resources
    },
)

Key points:

• ExternalRunnerTestInfo marks this as a test with external dependencies
• type: Usually "custom" for local resources
• command: The test script to run
• local_resources: Dict of resource name → broker target label
  • Names are arbitrary (for reference)
  • Must match names in required_local_resources
• required_local_resources: List of resources needed
  • listing: Usually False (not needed during buck2 targets)
  • execution: Usually True (needed during test execution)

Example:

_http_test_rule = rule(
    impl = lambda ctx: [
        DefaultInfo(),
        RunInfo(args = cmd_args([ctx.attrs.script[DefaultInfo].default_outputs[0]])),
        ExternalRunnerTestInfo(
            type = "custom",
            command = [cmd_args([ctx.attrs.script[DefaultInfo].default_outputs[0]])],
            local_resources = {
                'http': ctx.attrs.http_broker.label,
            },
            required_local_resources = [
                RequiredTestLocalResource("http", listing = False, execution = True),
            ],
        ),
    ],
    attrs = {
        "script": attrs.source(),
        "http_broker": attrs.exec_dep(providers = [LocalResourceInfo]),
    },
)

Step 6: Wire Everything in BUILD File

Template:

# SPDX-FileCopyrightText: © 2024-2026 Austin Seipp
# SPDX-License-Identifier: Apache-2.0

load("@root//buck/shims:shims.bzl", depot = "shims")
load(":defs.bzl", "<exports-struct>")

# 1. Export broker script
depot.export_file(
    name = "<broker-script-name>",
    src = "<broker-script-file>",
)

# 2. Create broker target
<exports-struct>.<broker_rule>(
    name = "<broker-resource-name>",
)

# 3. Create test target
<exports-struct>.<test_rule>(
    name = "<test-name>",
    script = "<test-script-file>",
    <broker_attr> = ":<broker-resource-name>",
)

Example:

# SPDX-FileCopyrightText: © 2024-2026 Austin Seipp
# SPDX-License-Identifier: Apache-2.0

load("@root//buck/shims:shims.bzl", depot = "shims")
load(":defs.bzl", "my_resources")

# Export broker script
depot.export_file(
    name = "http-broker",
    src = "http-broker.sh",
)

# Create broker target
my_resources.http_broker(
    name = "http-broker-resource",
)

# Create test target
my_resources.http_test(
    name = "http-test",
    script = "http-test.sh",
    http_broker = ":http-broker-resource",
)

Don't forget to export rules in defs.bzl:

my_resources = struct(
    http_broker = _http_broker_rule,
    http_test = _http_test_rule,
)

Step 7: Test and Debug

1. Test broker script standalone:

bash <broker-script>.sh

Should output valid JSON. Validate:

bash <broker-script>.sh | jq .

2. Test manually:

# Run broker
bash <broker-script>.sh
# Note the PID and resource values

# Set env vars manually
export <ENV_VAR>=<value-from-json>

# Run test
bash <test-script>.sh

# Kill broker
kill <pid-from-json>

3. Run with Buck2:

buck2 test //<package>:<test-name>

4. Debug failures:

• Check broker output: buck2 test ... -v 2
• Validate JSON: bash broker.sh | jq .
• Check env vars: Add env | grep <PREFIX> to test script
• Check processes: ps aux | grep <service>
• Clean up leaked processes: pkill -f <service>

Multiple Resources Pattern

For tests needing multiple services:

Broker Rules

Create separate broker rules for each service (as shown above).

Test Rule

_multi_test_rule = rule(
    impl = lambda ctx: [
        DefaultInfo(),
        RunInfo(args = cmd_args([ctx.attrs.script[DefaultInfo].default_outputs[0]])),
        ExternalRunnerTestInfo(
            type = "custom",
            command = [cmd_args([ctx.attrs.script[DefaultInfo].default_outputs[0]])],
            local_resources = {
                'http': ctx.attrs.http_broker.label,
                'db': ctx.attrs.db_broker.label,
                'redis': ctx.attrs.redis_broker.label,
            },
            required_local_resources = [
                RequiredTestLocalResource("http", listing = False, execution = True),
                RequiredTestLocalResource("db", listing = False, execution = True),
                RequiredTestLocalResource("redis", listing = False, execution = True),
            ],
        ),
    ],
    attrs = {
        "script": attrs.source(),
        "http_broker": attrs.exec_dep(providers = [LocalResourceInfo]),
        "db_broker": attrs.exec_dep(providers = [LocalResourceInfo]),
        "redis_broker": attrs.exec_dep(providers = [LocalResourceInfo]),
    },
)

BUILD File

my_resources.multi_test(
    name = "integration-test",
    script = "integration-test.sh",
    http_broker = ":http-broker-resource",
    db_broker = ":db-broker-resource",
    redis_broker = ":redis-broker-resource",
)

Test script will have access to all environment variables from all brokers.

Common Patterns and Templates

Dynamic Port Allocation

Avoid hardcoded ports by using port 0 (OS assigns random port):

# Start with port 0
python3 -m http.server 0 &
PID=$!

# Discover assigned port
sleep 0.2
PORT=$(lsof -ti -sTCP:LISTEN -a -p $PID)

echo "{\"pid\": $PID, \"resources\": [{\"port\": \"$PORT\"}]}"

Unix Socket Pattern

Avoids port allocation entirely:

SOCKET="$TMPDIR/service-$$"
my_service --socket="$SOCKET" &
PID=$!

# Wait for socket file
for i in {1..50}; do
    if [[ -S "$SOCKET" ]]; then
        break
    fi
    sleep 0.1
done

echo "{\"pid\": $PID, \"resources\": [{\"socket_path\": \"$SOCKET\"}]}"

Health Check Patterns

Port-based:

for i in {1..30}; do
    if curl -s http://localhost:$PORT/health > /dev/null 2>&1; then
        break
    fi
    sleep 0.1
done

Socket-based:

for i in {1..50}; do
    if [[ -S "$SOCKET" ]]; then
        break
    fi
    sleep 0.1
done

Process-based:

for i in {1..30}; do
    if kill -0 $PID 2>/dev/null; then
        break
    fi
    sleep 0.1
done

Best Practices

1. Isolation via $TMPDIR

   • Use $TMPDIR for all temp files
   • Add $$ (PID) to paths for uniqueness
   • Example: $TMPDIR/my-service-$$

2. Always wait for readiness

   • Don't output JSON until service is ready
   • Use health checks, file existence, port listening
   • Timeout after reasonable attempts

3. Dynamic resource allocation

   • Port 0 for random ports
   • Unix sockets with unique paths
   • Avoid hardcoded ports/paths

4. Clean JSON output

   • Only JSON to stdout
   • All logs to stderr or /dev/null
   • Validate with jq .

5. Error handling

   • set -euo pipefail in all bash scripts
   • Check service started successfully
   • Output error messages to stderr

6. Testing

   • Test broker standalone first
   • Test script manually with env vars
   • Run with Buck2 last

Troubleshooting Guide

"Invalid JSON from broker"

Cause: Broker outputs non-JSON or malformed JSON.

Fix:

bash broker.sh | jq .

Look for error messages. Ensure only JSON on stdout.

"Environment variable not set"

Cause: Mismatch between JSON keys and resource_env_vars.

Fix: Check mapping:

• Broker: {"resources": [{"my_key": "value"}]}
• Rule: resource_env_vars = {"MY_VAR": "my_key"}
• Test: $MY_VAR should be value

"Resource not available"

Cause: Name mismatch in local_resources and required_local_resources.

Fix:

local_resources = {
    'myservice': ctx.attrs.broker.label,  # Name must match below
},
required_local_resources = [
    RequiredTestLocalResource("myservice", ...),  # Must match above
],

"Process already running" or port conflicts

Cause: Previous test didn't clean up or hardcoded ports conflict.

Fix:

• Use dynamic ports (port 0)
• Use Unix sockets
• Add $$ to temp directories
• Manually kill: pkill -f my-service

Test hangs

Cause: Broker never becomes ready (infinite wait loop).

Fix: Add timeout to health check:

for i in {1..30}; do  # Max 3 seconds
    if <ready>; then
        break
    fi
    sleep 0.1
done

# Verify ready
if ! <ready>; then
    echo "Service failed to start" >&2
    kill $PID 2>/dev/null
    exit 1
fi

Examples

See:

• buck/tests/local-resources/ - Simple examples (HTTP, socket, multi-resource)
• buck/third-party/qemu-static/ - Real-world example (QEMU with TPM)
• docs/buck2.md - Comprehensive documentation

Quick Reference

Broker Script Checklist

• SPDX headers
• set -euo pipefail
• Use $TMPDIR/service-$$ for isolation
• Start service in background (&)
• Capture PID ($!)
• Wait for service ready (health check)
• Output ONLY JSON to stdout
• JSON includes pid and resources array
• Resource keys will become env var names

Broker Rule Checklist

• SPDX headers
• Provides LocalResourceInfo
• setup points to broker script
• resource_env_vars maps JSON keys → env vars
• _script references broker script target

Test Rule Checklist

• SPDX headers
• Provides ExternalRunnerTestInfo
• type = "custom"
• local_resources dict with broker labels
• required_local_resources list matches dict keys
• Attributes for script and broker(s)

BUILD File Checklist

• SPDX headers
• Load defs.bzl and shims
• export_file for broker script
• Broker target using broker rule
• Test target using test rule
• Test script passed as script =
• Broker passed as <broker_attr> = :<broker-target>

Command Reference

# Test broker script
bash <broker>.sh
bash <broker>.sh | jq .

# Test manually
bash <broker>.sh  # Note PID and resources
export VAR=value  # Set env vars from JSON
bash <test>.sh
kill <pid>

# Run with Buck2
buck2 test //<package>:<test>
buck2 test //<package>:<test> -v 2  # Verbose

# Debug
buck2 targets //<package>:  # List targets
buck2 build //<package>:<broker>  # Build broker
ps aux | grep <service>  # Find processes
pkill -f <service>  # Kill leaked processes

Related Skills

• buck2-test-workflow - Testing workflows and best practices
• buck2-build-troubleshoot - Debugging Buck2 build failures
• buck2-query-helper - Querying Buck2 build graph

Summary

Creating local resource tests:

1. Write broker script (outputs JSON with PID and connection info)
2. Create broker rule (LocalResourceInfo with resource_env_vars)
3. Create test rule (ExternalRunnerTestInfo with local_resources)
4. Wire in BUILD file (export script, create broker, create test)
5. Test standalone, then with Buck2

Key principles:

• Broker manages service lifecycle
• Buck2 handles cleanup
• Environment variables pass connection info
• Isolation via temp directories and dynamic allocation
• Hermetic and parallel-safe