What the data for this category looks like
This category's data mainly comes from cloud billing statements, monitoring systems, business forecast reports, and stress test results. Daily produced documents include capacity planning reports, scaling application forms, resource utilization reports, and budget calculation sheets. Data update cycles are mostly weekly or monthly. Each individual document contains fields such as server model, resource configuration parameters, estimated peak load, procurement budget, and more. Field units include CPU cores, storage capacity, monetary amount, request thresholds, and more. There is no unified fixed format, and adjustments are needed to match specific business scenarios.
What constraints these characteristics impose on the "knowledge base retrieval and recall" link
Data update cycles are mainly weekly or monthly. Incremental synchronization of the knowledge base must match this rhythm to avoid high-frequency updates occupying cluster resources. Document formats and fields are inconsistent. The retrieval link must support custom field mapping and multi-format text parsing to ensure that capacity planning documents from different sources can be effectively recalled. Numerical fields account for a high proportion. Retrieval must support parameter range matching logic to adapt to numerical query requirements such as resource utilization and peak load. Retrieval response must remain stable in operational on-call scenarios to avoid affecting internal work order processing efficiency due to recall delays.
How to set the configurations
The following configurations are formulated based on the retrieval logic of FastGPT v4.17.0.
| Configuration Item | Recommended Range | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches highly relevant numerical parameters and business descriptions in capacity planning documents, and avoids recalling irrelevant low-match content |
limit | Top 8–12 entries | Covers the core parameters of a single capacity planning report, while controlling the context length to avoid exceeding the model's processing limit |
searchMode | Comprehensive ranking | Combines keyword matching and semantic similarity, and prioritizes recalling core documents containing peak load and resource configuration |
usingReRank | Enabled | Performs secondary ranking on initial recall results, corrects sorting bias from initial retrieval, and matches the priority logic of business scenarios |
datasetSearchUsingExtensionQuery | Enabled | Automatically optimizes user work order questions, converts operation and maintenance terms into retrieval terms that match knowledge base fields, and improves recall accuracy |
Three common misconfigurations
- Retrieval result sorting does not meet expectations, and comprehensive ranking results do not prioritize highly matched capacity planning documents. The cause is that the
usingReRankconfiguration is not enabled. Initial recall only relies on basic semantic similarity and does not adjust sorting based on the priority logic of business scenarios. - A
getaddrinfo EAI_AGAIN mongoerror occurs during knowledge base import or retrieval, and recovery cannot be achieved after restarting the service. The cause is an abnormal database connection configuration associated with the knowledge base, or cluster network policies restricting database access, causing the retrieval link to fail to connect to the data source normally. - Resource utilization data embedded in images cannot be recalled from PDF-format capacity planning reports. The cause is that the document image parsing function is not enabled, and only the plain text content of the PDF is extracted, without covering structured data in images.
How to confirm that configurations are correctly set
- Submit a test retrieval containing the peak load keyword, check whether the sorting of recall results meets business expectations, and adjust related configurations until matching is achieved.
- View the knowledge base synchronization logs to confirm that the data update frequency matches the weekly or monthly business rhythm, and there are no high-frequency invalid synchronization records.
- Trigger a retrieval containing numerical parameters, verify that capacity planning documents containing corresponding fields can be recalled, and confirm that the custom field mapping configuration takes effect.
- Simulate daily retrieval requests, check whether the response delay meets the real-time requirements of the operational on-call scenario, and adjust the return result quantity control configuration.
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.