What data looks like for this category
Data sources for this category include incident timelines, operation logs, monitoring metrics, and code commit records. The update frequency is once per major incident or once per month. Regularly produced documents include RCA reports, incident tickets, review minutes, and rectification tracking forms. RCA reports contain fields such as incident ID, occurrence time, impact scope, root cause items, and rectification measures. Code commit records contain fields such as change ID, submitter, review comments, and associated incident ticket numbers. The data structure centers on multi-source information associated with incidents, and all fields are strongly tied to the incident troubleshooting and root cause analysis workflows.
What constraints these characteristics impose on tool calling and plugins
The low data update frequency requires tool calling to be triggered only when bound to an incident ticket or monthly analysis cycle, to avoid invalid calls. Multi-source data requirements mean plugins must adapt to pull logic for different formats such as logs and code change records, to ensure data integrity. The fixed document structure requires the output of tool calling to match the required fields of documents such as RCA reports, to enable direct insertion into existing templates. The core of root cause analysis relies on accurate associated data, so tool calling must accurately link incident tickets with corresponding code changes and log information, to avoid irrelevant data interfering with the analysis process. The most time-consuming evidence collection link can be automated via tool calling, so the trigger timing of tool calling must align with the process nodes of incident review.
How to set configurations
The following configurations are formulated based on FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
toolChoice | auto | Adapts to automatic calling logic for multi-source data, no manual specification of tool type required |
functionCall | ["fetch_fault_logs", "fetch_code_changes", "generate_rca_section"] | Covers three types of tools required for root cause analysis: log pulling, code change query, and RCA content generation |
| Tool calling timeout | 600 seconds | Root cause analysis requires pulling multi-source data, which takes a long time; 600 seconds covers most data pulling scenarios |
| Tool calling cache duration | 7 days | Data update frequency is once per month or major incident, so a one-week cache is sufficient to cover a single analysis cycle |
| Tool calling retry count | 3 times | Addresses network fluctuations or interface rate limiting, avoiding resource consumption from repeated calls |
| Associated field matching rules | Incident ID, change ID | Ensures pulled data is strongly bound to the current incident ticket, complying with the field requirements of RCA reports |
Three common mistakes
- Phenomenon: Tool calling returns a 429 Request rate increased too quickly status code. Cause: No reasonable calling interval and retry count are set, triggering interface rate limiting in a short period.
- Phenomenon: API key configurations of different tool instances are overwritten by each other. Cause: No exclusive key is bound to each tool individually, and global configuration items are reused.
- Phenomenon: Tool calling results cannot match the required fields of RCA reports. Cause: No associated field matching rules are configured, and pulled data is not bound to the current incident ticket.
How to confirm the configuration is complete
- A test incident ticket is submitted to trigger tool calling, and automatic pulling of associated logs and code change data is verified.
- Tool calling logs are reviewed to confirm that the number of rate limiting errors conforms to the configured retry rules, with no repeated invalid calls.
- Fields of tool calling results are checked to confirm inclusion of associated fields such as incident ID and change ID, enabling direct insertion into the RCA report template.
- API key configurations of multiple tool instances are verified, confirming that keys of different instances do not overwrite each other.
The values provided are common starting points and should be measured against the reader's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-10-19.