What data for this category looks like
Data for this category comes primarily from CDN consoles, DNS resolution records, edge node access logs, and monitoring alert systems. Documents linked to daily operations and tickets include domain name access forms, cache rule configuration documents, refresh and preheating operation records, and fault root cause analysis reports. Data fields include domain names, cache rule identifiers, edge node IP addresses, access delay values, alert levels, fault trigger times, and more. Update frequency is either daily synchronization or real-time updates when configuration changes are made.
Constraints on re-ranking and hybrid retrieval
The multi-source, scattered data characteristics of this category require hybrid retrieval to cover structured configuration data, semi-structured logs, and unstructured fault reports. Data update cycles vary widely. Real-time modified cache rules and daily aggregated access logs require differentiated timeliness weighting during retrieval to avoid interference from outdated results. Field differences across data types are significant. Matching rules must be set for exclusive fields such as domain names, alert levels, and node IP addresses to ensure retrieval results align strongly with operations scenarios. The retrieval logic must also adapt to different data structures to ensure compatibility between structured and unstructured content.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
usingReRank | Enabled | CDN operations scenarios require prioritizing matching of highly relevant fault and configuration data. Re-ranking can filter low-correlation recall results |
limit | Top 8–12 entries | Valid information for operation tickets is usually concentrated in a small number of highly relevant items. This range balances retrieval coverage and efficiency |
searchMode | hybrid | Both structured configuration fields and unstructured fault descriptions must be matched. Hybrid retrieval balances exact matching and semantic matching |
similarity | 0.72–0.88 | For strongly identifying fields such as CDN domain names and node IP addresses, this range can filter irrelevant cross-domain retrieval results |
| Model channel configuration | Bind vendor credentials for the corresponding re-ranking model | Ensure normal invocation of re-ranking services, and comply with FastGPT v4.17.0 model channel configuration requirements |
Three common mistakes
- Phenomenon: The re-ranking marker field in retrieval results is false, and no post-reranking results are generated. Cause: The
AIPROXY_API_ENDPOINT(AI Proxy service root address, format such as http://aiproxy:3000, no additional paths allowed) andAIPROXY_API_TOKEN(admin token matching the AI Proxy server configuration) environment variables are not correctly configured on the App/Admin server. This causes App/Admin service startup verification to fail, and the re-ranking service cannot be invoked normally. - Phenomenon: Commercial re-ranking models return empty content. Cause: Vendor credentials for the re-ranking model are not correctly filled in the model channel configuration page, or the model name does not match the console configuration.
- Phenomenon: The same model name is used for both embedding and re-ranking configurations, overwriting existing settings. Cause: In FastGPT v4.17.0, model configurations only allow unique names regardless of usage type. Adding a model with the same name automatically replaces the original configuration.
How to confirm configuration is complete
- Check that the App/Admin server has correctly set the
AIPROXY_API_ENDPOINT(AI Proxy service root address, format such as http://aiproxy:3000, no additional paths allowed) and non-emptyAIPROXY_API_TOKEN(admin token matching the AI Proxy server configuration) environment variables. - Check that vendor credentials and model name for the re-ranking model in the model channel configuration are correctly configured.
- Enter a test keyword related to CDN operations, and verify whether the retrieval result list includes post-reranking sorting markers.
- Attempt to add a re-ranking model with the same name as an existing configuration, confirm that the system blocks duplicate configurations, and verify that the uniqueness rule is active.
The values provided here are common starting points, and should be evaluated 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.