What data for this category looks like
Data sources include the power grid enterprise marketing business application system, equipment operation and maintenance archive database, and administrative approval docking platform. Updates are triggered by business processes. Data is updated synchronously when a single declaration is submitted. Closed cases are archived quarterly. Each declaration package includes structured fields: declaration number, power consumption capacity, voltage level, and applicant information. It also includes accompanying unstructured attachments: engineering design drawings, equipment test reports, and approval node circulation records. All units use standard power industry units such as kilovolt, megawatt-hour, and square meter.
What constraints these characteristics impose during deployment and upgrade
There are many structured fields and fixed business codes. During deployment, matching the field mapping rules of the corresponding vector model is required to avoid retrieval result deviations. Most accompanying attachments are large PDF or CAD drawings. Resource usage for storage and vector extraction is high. Sufficient memory and storage resources must be reserved during upgrades. Most power grid services are deployed in intranet environments. All dependent images must be packaged in advance for offline deployment to prevent failed dependency pulls during upgrades. Case review involves compliance checks. Compatibility with historical data must be ensured during upgrades to avoid interrupting approval workflows.
How to set configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
MONGODB_URI | mongodb://[intranet IP]:27017/fastgpt?authSource=admin | Power grid business data is stored in an intranet MongoDB cluster. The authentication source and intranet access address must be specified |
REDIS_URL | redis://:[password]@[intranet IP]:6379/0?socketTimeout=60000 | Used to cache vector recall results and session data. Setting the timeout to 60 seconds adapts to large attachment processing |
STORAGE_S3_ENDPOINT | http://[intranet storage server IP]:9000 | Adapts to the intranet object storage service for power grids. Replaces public network storage to ensure data security |
DB_MAX_LINK | 200–300 | Meets database connection requirements for concurrent declaration queries and case retrieval |
| FILETOKENKEY | Generate a 32-bit random string based on the business scenario | Used to encrypt access tokens for declaration materials to prevent unauthorized access |
Three common misconfigurations
- No organized, chaotic output of knowledge base query results after offline upgrade: The vector database metadata synchronization script was not executed during upgrade, leading to incorrect
REDIS_URLconfiguration. Cached vector recall results are not properly associated, and the large language model cannot extract valid information. - Failed to load the application historical version list: The database migration script was not executed before upgrade. Older historical version documents in MongoDB lack the
tmbIdfield, which does not match the version verification rules of FastGPT v4.17.0. - External calling interfaces return connection timeouts:
FE_DOMAINwas not configured as the power grid intranet access domain name, leading to cross-domain request interception and failure of external calls to access the service normally.
How to confirm configurations are correct
- Run
docker compose logs -f mongodbto view database connection logs, and confirm there are no connection timeout or authentication failure errors. - Upload a power grid engineering design drawing PDF under 100 MB, and check that the corresponding file is generated in the storage path corresponding to
STORAGE_S3_ENDPOINT. - Initiate a simulated declaration material query, and confirm that the number of vector recall results matches the concurrency limit specified by
DB_MAX_LINK. - Restart the service, and verify that in the offline environment of FastGPT v4.17.0, knowledge base queries can properly organize and output matching case content.
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 2026-09-28.