codex - 💡(How to fix) Fix Remote compaction intermittently fails with stream disconnected before completion [1 comments, 2 participants]

Codex for macOS intermittently fails when attempting remote compaction of a long thread. The visible error is: Error running remote compact task: stream disconnected before completion: error sending request for url (https://chatgpt.com/backend-api/codex/responses/compact) stream disconnected before completion: error sending request for url (https://chatgpt.com/backend-api/codex/responses/compact) remote compaction failed ... last_api_response_total_tokens=230780 ... all_history_items_model_visible_bytes=893315 ... model_context_window_tokens=258400 ... failing_compaction_request_model_visible_bytes=908574 ... compact_error=stream disconnected before completion: error sending request for url (https://chatgpt.com/backend-api/codex/responses/compact) remote compaction failed ... last_api_response_total_tokens=234880 ... all_history_items_model_visible_bytes=958429 ... model_context_window_tokens=258400 ... failing_compaction_request_model_visible_bytes=973688 ... compact_error=stream disconnected before completion: error sending request for url (https://chatgpt.com/backend-api/codex/responses/compact)

TL;DR

• The issue can be mitigated by adjusting the model context window or implementing a retry mechanism for remote compaction requests.

Guidance

• Investigate the correlation between the model context window tokens (model_context_window_tokens=258400) and the failure to compact, considering adjustments to this parameter.
• Review the API request for https://chatgpt.com/backend-api/codex/responses/compact to ensure it is properly formatted and that the server is not disconnecting due to request size or timeout issues.
• Consider implementing a retry mechanism with exponential backoff for the remote compaction request to handle intermittent failures.
• Analyze the last_api_response_total_tokens and all_history_items_model_visible_bytes values in the log lines to understand their impact on the compaction process.

Example

• No specific code snippet can be provided without more context on the Codex App's internal implementation, but a potential workaround could involve modifying the compaction request to handle larger payloads or to retry upon failure.

Notes

• The issue seems specific to the remote compaction request path and is intermittent, suggesting a potential problem with the request handling or server-side processing.
• The provided log lines indicate a possible correlation between the failure and the size of the data being compacted, but more investigation is needed to confirm this.

Recommendation

• Apply workaround: Implementing a retry mechanism for the remote compaction requests could help mitigate the issue until a more permanent fix is available. This approach can help improve the reliability of the compaction process without requiring significant changes to the underlying system.