[Bug] Ralph Loop tick() uses Promise.all — partial assignTasks failure corrupts task queue state

[noahwaldner]

Severity: High
Category: Logic Bug
File: src/ralph-loop.ts

Description

tick() runs checkTimeouts() and assignTasks() concurrently via Promise.all. assignTasks() iterates idle sessions and assigns tasks one by one in a loop. Each iteration mutates task state (task.assign(), session.assignTask()) before calling the async sendInput(). If assignTasks() throws partway through the loop, Promise.all re-throws immediately — leaving the queue in a partially-mutated state with no rollback.

Code

src/ralph-loop.ts:281–312:

private async tick(): Promise<void> {
  await Promise.all([
    this.checkTimeouts(),
    this.assignTasks(),   // ← if this throws mid-loop, earlier mutations are not rolled back
  ]);
  ...
}

private async assignTasks(): Promise<void> {
  const idleSessions = this.sessionManager.getIdleSessions();
  for (const session of idleSessions) {
    const task = this.taskQueue.next();  // ← task dequeued
    if (!task) break;
    await this.assignTaskToSession(task, session);
    // ↑ task.assign() + session.assignTask() happen inside here
    // if this throws on iteration 2, iteration 1's state mutations stand
  }
}

Concurrency Concern

checkTimeouts() and assignTasks() run simultaneously. A task that times out at the exact moment it is being assigned can be processed by both functions concurrently — checkTimeouts() calls task.fail() while assignTasks() is mid-way through task.assign(). The resulting task state is undefined.

Impact

After a partial failure:

1. Some tasks are in_progress with sessions assigned — correct.
2. Other tasks may be dequeued from the priority queue but never assigned — they are silently dropped.
3. Sessions from successfully-processed iterations remain busy but may have no prompt sent (if the failure occurred before sendInput()).
4. The task queue has no self-healing mechanism for this state — manual intervention or a full restart is required.

This is especially dangerous during high-frequency autonomous runs where many tasks are being assigned across many sessions simultaneously.

[Ark0N]

Already fixed — tick() now runs checkTimeouts() and assignTasks() sequentially (not via Promise.all), and assignTasks() has per-iteration try/catch so one failed assignment doesn't abort the loop.