What Data for This Category Looks Like
The data for this category primarily originates from cloud database consoles, slow logs, monitoring metrics, and internal work order systems. Updates occur either daily or in real time when instances are modified. Daily operation and maintenance documents include instance change orders, backup and recovery records, performance diagnosis reports, and parameter adjustment records. Each data entry contains fields such as instance ID, change time, operation content, monitoring metric values, and work order number. There is no unified fixed unit: metric data uses units matching their corresponding monitoring items, and document data is stored in plain text format.
Constraints Imposed by These Characteristics on Knowledge Base Retrieval and Recall
Decentralized data sources require the knowledge base to connect to multiple data types, with separate configuration needed for import rules and format adaptation for each. Differences in update frequency demand combining incremental and full synchronization to prevent data lag or redundant imports. Diverse document structures include both structured monitoring metric values and unstructured operation and maintenance instruction text, requiring adapted text splitting logic for different formats. Strong links between work orders and change records require associating fields like instance ID during retrieval to avoid invalid cross-instance recall. The timeliness of slow logs and performance reports means recall results must prioritize latest change and diagnosis records to maintain reference value.
Configuration Settings
The following configurations are written for FastGPT v4.17.0:
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the precision of professional terminology in cloud database operation scenarios, avoiding recall of irrelevant historical records |
limit | Top 8 entries | Covers the number of knowledge points for common operation scenarios such as slow SQL troubleshooting and parameter adjustment, avoiding overly long context |
searchMode | Exact Match | Prioritizes recalling content highly relevant to operation and maintenance work orders and change records, aligning with the requirement to return knowledge base responses verbatim |
usingReRank | Enabled | Re-ranks initially recalled results by relevance, improving sorting accuracy for backup and recovery, performance diagnosis content |
datasetSearchUsingExtensionQuery | Disabled | Prevents generation of additional queries that deviate from knowledge base original content, ensuring responses come entirely from imported question-answer pairs and documents |
Three Common Misconfigurations
- Phenomenon: Retrieval results include generated content not present in the knowledge base. Cause: The
datasetSearchUsingExtensionQueryconfiguration was not disabled, causing the system to generate additional queries that expand the recall scope. - Phenomenon: Some user questions fail to recall corresponding knowledge points. Cause: A reasonable retrieval strategy was not configured for the cloud database operation-specific scenario, leading to insufficient matching precision for professional terminology.
- Phenomenon: Recall results cannot be filtered by instance ID or work order number. Cause: No filtering fields were configured for the knowledge base, and business-related fields were not included in the retrieval matching scope.
How to Verify Successful Configuration
- Input the original text of an existing operation and maintenance question-answer pair in the knowledge base, and check if the returned result exactly matches the knowledge base response.
- View the retrieval log to confirm that all recalled results come from imported cloud database operation and maintenance-related documents, with no additional generated non-knowledge base content.
- Attempt retrieval using business fields such as instance ID and work order number, and verify that the filtering function can screen results by the specified fields.
- Adjust the
similaritythreshold, observe changes in the relevance of recall results, and confirm alignment with the precision requirements of the current scenario.
The values provided here are common starting points and should be evaluated against samples from the relevant deployment.
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.