openclaw - 💡(How to fix) Fix [Bug]: web_search tool aborts silently with Gemini provider — instantaneous abort, no err.log entry [1 participants]

web_search with the Gemini provider (google_search grounding) intermittently aborts instantly with the message "This operation was aborted". The abort is near-instantaneous (<1s), ruling out timeout. The error is swallowed by the tool layer and never appears in gateway.err.log. | gateway.err.log | No corresponding error — tool layer swallows it | Session-level runAbortController being triggered prematurely. The abort is too fast to be a network/timeout issue and too silent to be normal error handling. Suspect a race condition in the session run lifecycle where the abort controller fires before the search completes, and the error is caught but not logged.

TL;DR

Investigate and potentially adjust the session-level runAbortController to prevent premature triggering, which is suspected to cause the instantaneous aborts.

Guidance

• Review the implementation of runAbortController in the context of the Gemini provider (google_search grounding) to understand under what conditions it triggers an abort.
• Since the issue is intermittent and not reproducible 100%, consider adding logging or debugging statements around the runAbortController to capture more information about when and why it's being triggered prematurely.
• Given that setting timeoutSeconds to 60 in openclaw.json had no effect, explore other configuration options or environmental factors that could influence the behavior of runAbortController.
• To mitigate the issue, consider implementing a retry mechanism with a backoff strategy for the web_search calls that fail with the "This operation was aborted" message, as sometimes retries succeed.

Example

No specific code example can be provided without more details on the runAbortController implementation, but a generic approach to handling such aborts might involve wrapping the search call in a retry loop:

import time

def web_search_with_retry(query, max_retries=3, backoff_factor=1):
    for attempt in range(max_retries):
        try:
            # Call web_search here
            result = web_search(query)
            return result
        except AbortError:
            if attempt < max_retries - 1:
                time.sleep(backoff_factor * (2 ** attempt))
            else:
                raise

# Usage
try:
    result = web_search_with_retry("your_query")
except AbortError:
    print("Failed after retries")

Notes

The solution may require deeper understanding of the runAbortController and its integration with the Gemini provider. The intermittent nature suggests a race condition, which can be challenging to debug and fix.

Recommendation

Apply a workaround by implementing a retry