Add support for multiple service endpoints and process monitoring modes

[thinkingfish]

Problem

The rezolus-capture scripts currently support only a single service metrics endpoint and require a fixed duration for capture. This limits flexibility for:

• Monitoring multiple services simultaneously
• Capturing metrics while running a command or monitoring an existing process
• Validating that endpoint/source pairs are properly configured

Solution

Enhanced both scripts/rezolus-capture and docker/rezolus-capture with the following changes:

1. Multiple Service Endpoints: Changed ENDPOINT and SOURCE from single variables to arrays (ENDPOINTS and SOURCES), allowing users to specify multiple --endpoint and --source pairs. Each endpoint is recorded to a separate temporary parquet file.

2. Flexible Capture Modes: Replaced the required --duration flag with three mutually exclusive capture modes:

   • --duration <DURATION>: Capture for a fixed time period (original behavior)
   • --command <CMD...>: Run a command and capture while it executes
   • --pid <PID>: Monitor an existing process and capture until it exits

3. Improved Validation:

   • Enforce exactly one capture mode is specified
   • Validate endpoint/source pairing (each endpoint must have a corresponding source)
   • Check for duplicate source names
   • Verify target process exists when using --pid

4. Enhanced Recording Management:

   • Track all recorder processes in RECORDER_PIDS array
   • Added stop_recorders() helper function for graceful shutdown
   • Properly handle cleanup of multiple service parquet files
   • Support command execution with proper signal handling and exit code tracking

5. Flexible Output Combination:

   • Dynamically combine all non-empty parquet files (system + all services)
   • Handle single file case by copying instead of combining
   • Improved error handling when no recordings are produced

6. Documentation Updates:

   • Updated help text to reflect new capture modes and multiple endpoint support
   • Added examples for multiple endpoints, command execution, and PID monitoring
   • Clarified that sources must be specified once per endpoint in order

Result

Users can now:

• Monitor multiple services in a single capture session
• Capture metrics while running benchmarks or tests with --command
• Attach to running processes with --pid for dynamic monitoring
• Specify multiple endpoints with proper validation of source pairing
• Benefit from improved error messages and validation

The changes maintain backward compatibility for the --duration mode while adding powerful new capabilities for flexible metric collection scenarios.

https://claude.ai/code/session_01WV3WMHXPjqKaWrNo6KJN2n