Add OPA-gated subprocess execution APIs (Deno.Command and child_process.exec)

[r33drichards]

Summary

This PR adds policy-gated subprocess execution capabilities to the JavaScript runtime, supporting both Deno's Command API and Node.js-compatible child_process.exec(). All subprocess invocations are evaluated against an OPA policy chain before execution.

Key Changes

• New subprocess module (server/src/engine/subprocess.rs):

  • Implements two async ops: op_subprocess_output (for Deno.Command.output()) and op_subprocess_exec (for child_process.exec())
  • Provides JavaScript wrappers that expose both APIs to user code
  • Supports command execution with arguments, working directory, and environment variable configuration
  • Returns structured output with exit code, stdout, stderr, and success status
  • Encodes binary output as base64 for safe transmission across the JS/Rust boundary

• OPA policy integration:

  • Subprocess operations are gated by OPA policies before execution
  • Policy input includes operation type, command, arguments, working directory, and environment variables
  • Supports three operation types: command_output, command_spawn, and exec
  • Added subprocess field to PoliciesConfig to configure subprocess policy chains

• Runtime integration:

  • Added SubprocessConfig to ExecutionConfig for passing policy chains to the runtime
  • Conditionally registers subprocess extension and injects JavaScript wrappers when subprocess policies are configured
  • Integrated subprocess policy chain initialization in main.rs

• Example policy (policies/subprocess.rego):

  • Provides template for defining allowed commands and shell command patterns
  • Demonstrates filtering by operation type, command name, and working directory

• Comprehensive testing:

  • Unit tests for policy input serialization, base64 encoding, and options deserialization
  • Integration tests verifying policy allow/deny behavior for subprocess operations

Implementation Details

• Binary output from subprocess calls is base64-encoded in Rust and decoded back to Uint8Array in JavaScript for the Deno API
• The child_process.exec() implementation supports configurable encoding (defaults to UTF-8 strings, can be set to "buffer" for binary)
• Shell execution for exec() is platform-aware (uses /bin/sh -c on Unix, cmd /C on Windows)
• Policy evaluation is async and occurs before any subprocess is spawned

https://claude.ai/code/session_011eLgvj7BYjWQzr42j3WHK5

[github-actions]

MCP-V8 Load Test Benchmark Report

Comparison of single-node vs 3-node cluster at various request rates.

Results

Topology	Target Rate	Actual Iter/s	HTTP Req/s	Exec Avg (ms)	Exec p95 (ms)	Exec p99 (ms)	Success %	Dropped	Max VUs
cluster-stateful	100/s	46.5	1905.6	2001.78	5119.54	5139.68	79%	2998	100
cluster-stateful	200/s	52.1	3740.5	3601.11	5141.81	5157.39	39.2%	8609	200
cluster-stateless	1000/s	366	1521.2	2596.89	10000.66	12607.7	98%	37310	1000
cluster-stateless	100/s	99.9	300.8	52.51	53.28	62.63	100%	0	10
cluster-stateless	200/s	199.8	602.2	53.26	54.28	102.75	100%	1	21
cluster-stateless	500/s	474	1775.8	262.91	966.33	3197.77	100%	814	500
single-stateful	100/s	19	1866.9	4996.37	5165.47	5177.75	6%	4763	100
single-stateful	200/s	37.6	3797.5	5152.44	5179.43	5191.76	0%	9601	200
single-stateless	1000/s	143.8	361.9	6547.84	11965.2	15550.17	73.4%	50765	1000
single-stateless	100/s	99.9	309.1	57.68	103.89	106.11	100%	2	12
single-stateless	200/s	198.3	650.2	91.64	211.14	391.91	100%	62	80
single-stateless	500/s	143.6	386.4	3340.45	7223.04	10302.02	99.8%	21128	500

P95 Latency

Topology	Rate	P95 (ms)
cluster-stateful	100/s	5119.54	███████████████████████████
cluster-stateful	200/s	5141.81	███████████████████████████
cluster-stateless	100/s	53.28	█████████████
cluster-stateless	200/s	54.28	█████████████
cluster-stateless	500/s	966.33	██████████████████████
cluster-stateless	1000/s	10000.66	█████████████████████████████
single-stateful	100/s	5165.47	███████████████████████████
single-stateful	200/s	5179.43	███████████████████████████
single-stateless	100/s	103.89	███████████████
single-stateless	200/s	211.14	█████████████████
single-stateless	500/s	7223.04	████████████████████████████
single-stateless	1000/s	11965.2	██████████████████████████████

Notes

• Target Rate: The configured constant-arrival-rate (requests/second k6 attempts)
• Actual Iter/s: Achieved iterations per second (each iteration = 1 POST /api/exec)
• HTTP Req/s: Total HTTP requests per second (1 per iteration)
• Dropped: Iterations k6 couldn't schedule because VUs were exhausted (indicates server saturation)
• Topology: single = 1 MCP-V8 node; cluster = 3 MCP-V8 nodes with Raft