What the data for this category looks like
The typical data subjects for this category are SREs, cloud platform operations staff, and architects. Daily produced documents include capacity planning reports, scaling application forms, resource utilization reports, and budget calculation sheets. Data sources cover cloud bills, server monitoring systems, business traffic forecast reports, and stress test result files. Update frequencies are mostly weekly or monthly. Document fields include resource type, peak utilization, estimated scaling volume, budget amount, and corresponding units: CPU cores, memory GB, storage TB, and monetary yuan.
What constraints these characteristics impose on the knowledge base retrieval and recall link
Because data update frequencies are weekly or monthly, the knowledge base incremental sync cycle must match this rhythm. This avoids unnecessary resource consumption from frequent syncs. Most documents are structured reports and application forms. Retrieval must support both field matching and full-text search to cover different types of business documents. Data from multiple sources is scattered across different knowledge base collections. Retrieval scope must be limited by specifying a collection ID to avoid irrelevant data interfering with results. Units for resource-related fields have strict unified requirements. Retrieval results must retain original unit information. This ensures calculation accuracy for core subsequent steps such as peak prediction and resource bottleneck identification.
How to configure settings
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the relevance between structured fields and business terms in capacity planning documents, filters low-relevance monitoring alert content |
limit | Top 8–12 entries | Controls the number of documents retrieved per single request. Adapts to the batch processing volume of capacity planning documents, avoids overly long context affecting analysis |
searchMode | hybrid | Supports both full-text search and vector search. Covers retrieval needs for structured reports and unstructured reports |
usingReRank | Enabled | Performs reordering on retrieved results. Prioritizes returning high-value documents that include peak prediction and resource utilization data |
datasetSearchUsingExtensionQuery | Enabled | Generates extended query terms based on user questions. Matches indirectly related content in capacity planning documents, such as linking "scaling requirements" to stress test reports |
collectionIds | Configured according to actual business grouping | Limits retrieval scope to the capacity planning knowledge base for the corresponding business line. Avoids invalid data from cross-business line interference |
Three common configuration mistakes
- Phenomenon: Retrieval results do not return documents from the target knowledge base, only irrelevant monitoring log content. Cause: The
collectionIdsparameter is not used to specify the target knowledge base collection, so retrieval scope covers non-business-related datasets. - Phenomenon: Documents from low-priority business lines appear before high-priority capacity planning reports in retrieval results. Cause: The
usingReRankreordering function is not enabled, or retrieval order is not configured according to business weight. - Phenomenon: Context references only display plain text, and markdown formatting such as original document titles and lists is lost. Cause: The original format retention setting for document content is not enabled, so split text blocks lose formatting information.
How to confirm configuration is complete
- View the knowledge base retrieval configuration page, confirm that the
collectionIdsparameter has been filled with the knowledge base ID for the corresponding business line. - Initiate a test query, verify that the number of returned results matches the value set by the
limitparameter. - Check the similarity score of retrieval results, confirm that it falls within the interval configured by the
similarityparameter. - After enabling the
usingReRankfunction, initiate a query again, verify that high-value capacity planning reports are returned first.
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.