What the data for this category looks like
Data for this category comes from CMDB, cloud resource inventories, auto-discovery tools, and procurement ledgers. Four types of documents are produced regularly: configuration item lists, service dependency topologies, change records, and asset inventory sheets. Data is updated daily or weekly. Core fields include configuration item unique identifiers, asset classifications, cross-component dependency relationships, change timestamps, and inventory statuses. The most time-consuming steps are configuration item collection, cross-asset relationship verification, and post-change data synchronization. Data anomalies can directly lead to impact analysis deviations, delayed fault location, or discrepancies between asset book values and actual values.
What constraints do these characteristics impose on the knowledge base retrieval and recall workflow
Data sources for this category are scattered and have different update frequencies. A clear retrieval scope must be defined during searches to avoid mixing in irrelevant asset data. The semantic complexity of configuration item dependency relationships is high. General semantic recall models may not accurately match cross-component association queries, so recall parameters must be adjusted specifically. Daily updated data has high real-time requirements. The retrieval link must adapt to the synchronization rhythm of frequently updated datasets. Weekly updated static inventory data, by contrast, can reduce real-time verification overhead. Additionally, the time-consuming nature of configuration item collection and relationship verification requires that the retrieval link avoid overly long preprocessing steps to prevent timeouts.
How to set the configuration
| Configuration Item | Recommended Approach | Rationale |
|---|---|---|
collectionIds | Create datasets separately for CMDB, cloud resource inventories, and procurement ledgers, then fill in the corresponding dataset ID list | Precisely limit the retrieval scope to avoid irrelevant asset data interfering with search results |
similarity | 0.72–0.85 | Adapt to the semantic matching accuracy of configuration item dependency relationships, avoid false recalls or missed recalls |
limit | Top 10–15 entries | Adapt to the information density of on-duty operations scenarios, reduce the result filtering cost for on-duty personnel |
usingReRank | Enabled | Improve the Top 3 result accuracy of semantic matching for configuration item fields, adapt to the complex semantics of dependency relationships |
datasetSearchUsingExtensionQuery | Enabled | Cover vague asset name descriptions in operation and maintenance tickets, improve recall coverage |
searchMode | Semantic retrieval priority | Adapt to the unstructured characteristics of configuration item dependency relationships, improve matching effects |
Three common mistakes
- Symptom: Semantic retrieval fails to recall target configuration items, but full-text retrieval returns results normally. Cause:
datasetSearchUsingExtensionQueryis not enabled, or thesimilaritythreshold is set too high, resulting in insufficient semantic matching accuracy to match dependency relationship descriptions of configuration items. - Symptom: Slow operation logs are continuously reported when uploading knowledge base files, and MongoDB response is delayed. Cause: The
limitparameter for single retrieval is not restricted, or no reasonable re-ranking candidate set quantity is set afterusingReRankis enabled, causing the retrieval link to exceed the time threshold. - Symptom: Retrieval results only include cloud disk files, and do not cover configuration item documents from Feishu native knowledge bases. Cause: Feishu official knowledge base data source is not correctly associated; only cloud disk migrated files are synchronized, and the native synchronization channel for Feishu knowledge bases is not configured.
How to confirm the configuration is complete
- Navigate to the dataset management page of FastGPT v4.17.0, confirm that the associated data sources include CMDB, cloud resource inventories, procurement ledgers, and other corresponding sources, and that the
collectionIdsparameter has been filled with the correct dataset IDs. - Initiate a test query that includes configuration item dependency relationships, check whether the
similarityscore of the returned results falls within the 0.72–0.85 range, and whether the Top 5 results include the target configuration item. - Upload a configuration item change record file, wait for parsing to complete, check whether slow operation logs appear in the retrieval link, and confirm that MongoDB responses do not have continuous delays.
- After enabling
usingReRank, compare the Top 3 result matching accuracy before and after enabling to confirm that the re-ranking function is working correctly.
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.