What the data for this category looks like
Data for this category primarily comes from pressure testing platforms, APM monitoring tools, database slow logs, and production traffic baselines. Daily generated documents include pressure test plans, capacity assessment reports, pressure test data records, and optimization suggestion sheets. Document fields include throughput rate, latency, resource usage rate, and resource margin, with corresponding units of tps, ms, %, and GB. Updates are performed monthly or as a full refresh when major version iterations occur. The volume of content updated each time fluctuates based on business scale.
What constraints do these characteristics impose on the knowledge base retrieval and recall link
Multi-source heterogeneous data sources require retrieval to cover documents and logs in different formats, and need to support hybrid retrieval modes. Fixed document fields and units require that professional terms and indicator dimensions are prioritized during retrieval to avoid mixing irrelevant content. The monthly or version-level update rhythm requires that the knowledge base collection is strongly bound to the business domain, to avoid cross-domain content interfering with retrieval results. The professional nature of pressure test data requires that the precision of recall results matches business needs; otherwise, bottleneck positioning deviations or resource judgment errors may occur.
How to set the configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the precision requirement for professional terms in pressure test reports, and avoids mixing irrelevant non-performance content |
limit | Top 8 entries | Balances the number of retrieval results and context length, and adapts to the multi-paragraph, multi-indicator structure of pressure test plans |
searchMode | Hybrid retrieval | Covers multi-source heterogeneous content such as raw data from pressure testing platforms, APM monitoring logs, and database slow logs |
usingReRank | Enabled | Reranks recalled pressure test latency and throughput data, and prioritizes matching the core need of bottleneck positioning |
datasetSearchUsingExtensionQuery | Enabled | Supplements professional term queries for pressure test scenarios, and improves the recall rate of slow log-related content |
collectionIds | Bound by business domain | Associates exclusive knowledge base collections for pressure test plans and capacity reports, to avoid cross-domain content interfering with retrieval |
Three common mistakes
- Symptom: No referenced knowledge base ID is returned when calling the chat interface. Cause: The metadata return switch for the chat interface is not enabled, making it impossible to trace the source collection of retrieved content.
- Symptom: Format errors occur when importing Excel-format pressure test data records. Cause: The mapping relationship between table columns and knowledge base fields is not configured according to category document requirements, causing unit fields such as tps and latency to fail to be correctly identified.
- Symptom: Uploaded pressure test plan files cannot be deleted from the knowledge base. Cause: The deletion operation is performed in the knowledge base collection of a different business domain instead of the corresponding one.
How to confirm the configuration is complete
- In the FastGPT v4.17.0 knowledge base management interface, confirm that the
collectionIdsconfiguration bound to the corresponding business domain is set. - Initiate a query containing category-specific terms such as "pressure test latency" and "throughput rate", and check whether the returned result fields include the corresponding indicators.
- Call the chat interface, and check whether the returned result contains metadata information associated with the knowledge base.
- Adjust the
similaritythreshold, and verify whether the relevance of the recall results meets the requirements of the business scenario.
The configuration 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.