What data for this category looks like
Core data for database operation and maintenance scenarios comes from database monitoring, audit logs, SQL execution plans, and ticketing systems. Daily produced documents include standardized database change orders, slow SQL reports, backup records, and permission application forms. Document fields include change order number, SQL statement, execution duration, backup path, permission level, and others. Units include milliseconds, GB, calendar days, and others. Data update frequency is triggered by database changes or synchronized daily, with no fixed batch update cycle.
Constraints imposed by these characteristics on knowledge base retrieval and recall
Database operation and maintenance documents have both structured and unstructured features. Fields such as SQL statements and change order numbers require precise matching. Retrieval cannot rely solely on semantic recall, and must also consider keyword matching. Data update frequency is triggered by changes, so the knowledge base synchronization mechanism must support real-time or change-triggered updates. This avoids recalling outdated backup records or permission application forms. Individual document lengths vary greatly. Slow SQL reports and SQL execution plans may contain long text content. Chunking and recall logic must adapt to different document lengths to avoid truncating key information. Additionally, data in this category involves permission and performance-sensitive content. Retrieval scope must be strictly limited to exclusive knowledge base collections to prevent cross-business line data confusion.
Configuration settings
The following configurations are based on FastGPT v4.17.0, and are adapted to the knowledge base retrieval and recall requirements of database operation and maintenance scenarios:
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches the precision requirements of database operation and maintenance documents, and avoids recalling irrelevant general operation and maintenance documents or non-business-related content |
limit | Top 8 entries | Adapts to the number of associated documents for a single user query, controls context length, and avoids exceeding the LLM context window limit |
searchMode | hybrid | Combines keyword and semantic retrieval, and matches both structured fields such as SQL statements and change order numbers, and unstructured content such as slow SQL fault descriptions |
usingReRank | Enabled | Performs secondary ranking on recalled results such as slow SQL reports and change orders, and prioritizes matches for specific fault or approval scenarios in user queries |
datasetSearchUsingExtensionQuery | Enabled | Generates extended queries based on original user questions to supplement matching of abnormal SQL fragments and permission application details in audit logs |
collectionIds | Bind exclusive database operation and maintenance knowledge base collection ID | Strictly limits the retrieval scope to prevent recalling knowledge base data from other business lines or categories |
Three common configuration mistakes
- Phenomenon: An embedding error occurs when uploading database change orders or slow SQL reports. Cause: The long text structure of database documents is not adapted, the chunk length exceeds the model's supported range, or unescaped SQL special characters in the document cause parsing failure.
- Phenomenon: When using the knowledge base search plug-in, the specified database operation and maintenance knowledge base cannot be selected via variables, and the retrieval results include documents from unrelated categories. Cause: The
collectionIdsparameter is not correctly configured to bind the exclusive collection, or the variable is not mapped to the correct collection ID. - Phenomenon: Retrieval results include general operation and maintenance content outside the knowledge base, and the LLM response is not strictly limited to the recalled document range. Cause: Context filtering configurations other than
datasetSearchUsingExtensionQueryare not enabled, or the retrieval scope is not clearly limited to the specified collection.
How to confirm correct configuration
- Upload a single database change order or slow SQL report, check that the upload status shows normal with no embedding error reports, and confirm that the embedding process is completed.
- Enter a query related to slow query troubleshooting, verify that the recalled results include slow SQL reports and SQL execution plan documents, and that the
similarityscore falls within the preset 0.75–0.85 range. - Test calling the knowledge base search plug-in via variables, confirm that only documents from the current exclusive database operation and maintenance collection are recalled, and no cross-category data is mixed in.
- After initiating a query, check that the LLM response only references content from the recalled knowledge base documents, and no unauthorized information outside the knowledge base appears.
The values provided here are common starting points, and should be evaluated 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.