What this category of data looks like
Core data for this category includes version release-related documents: Release Notes, change logs, version tags, and release checklists. Primary data sources are Git tags, merged commit records, CI build artifacts, and Issue milestones. Data updates occur once per version. A single update includes categorized descriptions of all merged changes under that version, destructive change alerts, artifact verification information, and related content. The data structure uses primarily structured fields, including identifying items such as version number, change type, impact scope, and artifact hash. It also includes semi-structured change detail text to support verification and troubleshooting during operations on-duty.
Constraints on Knowledge Base Retrieval and Recall
Data sources are tied to code repositories and CI workflows. Retrieval must accurately associate version identifiers and artifact information to avoid recalling change content from non-current versions. Update frequency is once per version, so the knowledge base must support incremental synchronization to avoid resource consumption and delays caused by full reconstruction. The change categorization process requires significant time, so retrieval recall results must cover all key change items to support rapid verification during operations on-duty. Additionally, release artifact verification requires association with verification results returned by external CI tools. Retrieval must support access to non-native external data sources without manual data organization to complete verification.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches semantic relevance between change descriptions in Release Notes and operations issues, avoids recalling irrelevant changes or missing critical content |
limit | Top 8–12 entries | Covers core change items under the version, supports full verification needs for operations on-duty, while controlling context length |
searchMode | Semantic search + keyword matching | Balances structured fields such as version numbers and artifact hashes with semi-structured text of change descriptions, adapts to the mixed structure characteristics of this category of data |
usingReRank | Enabled, re-rank top 5 entries | Filters low-correlation recall results, focuses on content of concern to operations such as destructive changes and critical fixes |
datasetSearchUsingExtensionQuery | Enabled | Supports accessing artifact verification results returned by CI as extended data sources, meets the needs of release artifact verification |
collectionIds | ID of the dedicated Release Notes collection | Limits retrieval scope to only release documents for the current version, avoids recalling content from other versions or non-release categories |
Three Common Configuration Errors
- Phenomenon: Retrieval results include change documents from non-target versions or irrelevant PR review content. Cause: The
collectionIdsconfiguration item is not set to limit the knowledge base collection, resulting in recall of documents from other versions or non-release categories. - Phenomenon: The number of retrieval results returned is insufficient to cover all key changes for the current version. Cause: The
limitvalue is set too low, with insufficient recall quantity reserved to support full change verification. - Phenomenon: Artifact verification results returned by CI tools cannot be included in the retrieval data source. Cause:
datasetSearchUsingExtensionQueryis not enabled, and external data source synchronization rules are not configured, making it impossible to access non-native release artifact data.
How to Verify Correct Configuration
- Access the knowledge base management page of FastGPT v4.17.0, confirm that the ID of the target Release Notes collection has been correctly entered into the
collectionIdsconfiguration item. - Initiate a test retrieval, input "Destructive change description for version v2.3.0", check whether the returned results only include destructive change content for this version, with no irrelevant documents included.
- View the retrieval log, confirm that parameters such as
similaritythreshold andusingReRankhave taken effect, and the correlation of recall results meets expectations. - Call the FastGPT chat API, pass the specified
collectionIdsparameter, confirm that the results returned by the API only come from the target Release Notes knowledge base collection.
The values provided are common starting points and should be evaluated against local test 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.