openclaw - 💡(How to fix) Fix Circuit breaker: auto-reset sessions after N consecutive aborts [1 participants]

TL;DR

Implement a circuit breaker mechanism to auto-reset sessions after a configurable number of consecutive aborts to prevent infinite loops.

Guidance

• Introduce a counter to track consecutive aborts for each session, resetting it upon successful completion.
• Set a configurable threshold (e.g., 3) for the number of consecutive aborts allowed before triggering a session reset.
• Upon reaching the threshold, discard the current session context and initiate a new session.
• Consider logging or notifying when the circuit breaker is triggered for monitoring and debugging purposes.

Example

class SessionManager:
    def __init__(self, threshold=3):
        self.threshold = threshold
        self.abort_count = 0

    def handle_abort(self):
        self.abort_count += 1
        if self.abort_count >= self.threshold:
            self.reset_session()
            self.abort_count = 0

    def reset_session(self):
        # Discard current session context and start fresh
        pass

    def handle_success(self):
        self.abort_count = 0

Notes

This approach assumes that the session context can be safely discarded and restarted without significant side effects. The threshold value may need to be adjusted based on the specific application requirements and failure patterns.

Recommendation

Apply workaround: Implement the proposed circuit breaker mechanism to prevent infinite abort loops and reduce waste of resources. This approach directly addresses the problem by introducing a mechanism to detect and reset poisoned sessions, thereby preventing the replay of toxic context.