What the data for this category looks like
Data for this category primarily comes from CMDB, cloud resource inventories, auto-discovery tools, and procurement ledgers. Four document types are regularly generated: configuration item lists, service relationship topologies, change records, and asset inventory tables. Data updates occur daily or weekly. Each individual entry includes fields such as configuration item unique identifier, affiliated business domain, dependent components, running status, and change time. Most fields link to asset attributes and service dependency relationships. No unified fixed units exist, as values vary based on the specific configuration item type.
What Constraints These Characteristics Place on Re-ranking and Hybrid Retrieval
The data includes topological structures with strong dependency relationships and frequently updated change records. Hybrid retrieval must match both configuration item attributes and their associated relationships, balancing semantic recall and retrieval logic for dependency associations. Real-time data updated daily requires controlled retrieval latency, to avoid delaying response efficiency for on-duty work orders. Many configuration item fields have complex associations. The re-ranking link must first align with business dependency priorities, adjust sorting logic based on configuration item association relationships, and adapt to field format differences across multiple data sources. This ensures the relevance of retrieval results meets the needs of operations and maintenance scenarios.
How to Set Up Configurations
The following configurations are formulated for FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the attribute semantics of CMDB configuration items and work order keywords, covering recall requirements for core associated relationships |
limit | Top 10 entries | Adapts to the need to quickly locate core configuration items in operations on-duty scenarios, avoiding excessive results that interfere with judgment |
searchMode | hybrid | Enables hybrid retrieval mode, recalling both configuration item attributes and associated topology information, matching the multi-dimensional query needs of work orders |
usingReRank | Enabled | Re-ranks recall results based on the business dependency priority of configuration items, prioritizing returning configuration items most closely associated with the work order |
Three Common Configuration Errors
- Symptom: Retrieval results show no associated configuration items or return irrelevant assets, and no recall records for associated topologies appear in logs. Cause:
searchModeis not set to hybrid; only basic semantic recall is used, which cannot match the dependency relationships of configuration items. - Symptom: The interface shows the "result re-ranking" function is not enabled, but retrieval results are still sorted by basic similarity. Cause: The
usingReRankconfiguration is not enabled, or configuration parameters are not synchronized to the official retrieval link. - Symptom: Container startup fails, and the log reports an "unsupported architecture" error. Cause: An amd architecture re-ranking model image is used, and the arm architecture runtime environment is not adapted.
How to Verify Successful Configuration
- View the FastGPT retrieval configuration page, confirm that
searchModeis set to hybrid andusingReRankis enabled. - Submit a test work order that includes a configuration item name and its dependency relationship, and verify that returned results are sorted to prioritize core associated dependency items.
- Check retrieval logs, confirm that there are call records for the re-ranking module, and that the number of returned results matches the
limitsetting. - Adjust the
similaritythreshold, observe changes in the relevance of retrieval results, and verify that the parameter takes effect.
The values provided are common starting points and should be evaluated against deployment samples specific to the target environment.
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.