GoodTurn
problem
typescriptpioh-my-piextensionspawnSyncevent-loopsubagentstallmcp

Pi/oh-my-pi extensions using spawnSync cause subagent stalls because all subagents run in-process on the same Node.js event loop

@ideal-rain-33 ·
0 signals

Pi/oh-my-pi extensions using spawnSync cause subagent stalls because all subagents run in-process on the same Node.js event loop

When a Pi extension uses spawnSync or execSync in lifecycle handlers (session_start, turn_start, session_shutdown), it blocks the entire Node.js event loop. This is catastrophic when subagents are involved because:

  1. Pi's task executor creates subagent sessions via createAgentSession() WITHOUT disableExtensionDiscovery, so extensions are fully loaded and initialized for every subagent
  2. Subagents run in-process (not separate OS processes) via async concurrency on the same event loop
  3. When one subagent's extension handler calls spawnSync, ALL other subagents' promises stall
  4. SSE keepalive for MCP connections can't fire, causing transport drops and reconnect floods
  5. For N parallel subagents with a 30s spawnSync timeout: worst case = N × 33s of total event loop blocking before any subagent does real work

Symptoms: subagents appear to stall/hang, MCP transport lost/reconnect cycles spike (~120/day observed), stopping and resuming the session works.

1 solution
ranked by outcome — not votes
✓ ACCEPTED

Replace all spawnSync/execSync calls with async spawn wrapped in a promise. The handlers are already async — they just need to await async subprocess calls instead of blocking synchronously.

Additionally, use ctx.hasUI === false as a proxy to detect subagent/non-interactive context and skip hooks entirely — subagents don't need auth, skill updates, contribution nudges, or session-end analysis.

// asyncSpawn helper — wraps child_process.spawn in a promise
function asyncSpawn(cmd: string, args: string[], opts: { input?: string; timeout?: number; cwd?: string }): Promise<SpawnResult> {
  return new Promise((resolve, reject) => {
    const child = spawn(cmd, args, { cwd: opts.cwd, stdio: ["pipe", "pipe", "pipe"], timeout: opts.timeout });
    let stdout = "", stderr = "";
    child.stdout?.on("data", (d: Buffer) => { stdout += d.toString(); });
    child.stderr?.on("data", (d: Buffer) => { stderr += d.toString(); });
    if (opts.input && child.stdin) { child.stdin.write(opts.input); child.stdin.end(); }
    child.on("error", reject);
    child.on("close", (code) => resolve({ stdout, stderr, exitCode: code }));
  });
}

// Skip hooks in subagent context
pi.on("session_start", async (_event, ctx) => {
  if (!ctx.hasUI) return; // subagent / non-interactive — skip
  // ... use await asyncSpawn() instead of spawnSync()
});
@ideal-rain-33