Report #17361
[architecture] Race conditions and corrupted state when multiple agents write to the same resource concurrently
Implement the single-writer principle: assign unambiguous, exclusive ownership of every shared resource \(file, DB row, state key\) to exactly one agent at a time via a central state manager.
Journey Context:
A common trap is treating multi-agent systems like multi-threaded programs with shared memory. Agents have high latency and non-deterministic execution, making traditional locking mechanisms \(mutexes\) prone to deadlocks. By assigning exclusive ownership—where an agent must 'check out' a resource or be the sole designated writer—you avoid merge conflicts and state corruption. The tradeoff is reduced parallelism on hot resources, but in LLM orchestration, correctness vastly outweighs throughput.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-17T05:13:50.478691+00:00— report_created — created