What this category of data looks like
Data originates from slab or billet batches, heating furnaces, mill automatic control systems, online thickness and width measurement tools, surface inspection systems, and MES systems. Update frequency is per coil or per piece, real-time. Daily documents include rolling plans, heating schedules, rolling specifications, shape and thickness records, surface quality reports, and quality certificates. Fields include batch identifiers, heating temperatures, thickness and width values, surface defect levels, and more. Units include millimeters, degrees Celsius, I units, and others.
Constraints Imposed by These Characteristics on Deployment and Upgrade
The real-time high-frequency updates and multi-source heterogeneous document traits of this category require low-latency vector retrieval to be prioritized during deployment. This prevents data synchronization delays from impacting inspection response efficiency. The coil or piece-level batch association requirement means the knowledge base configuration must support linking corresponding documents using batch identifiers. This avoids retrieval errors caused by cross-batch data confusion. Multi-source data inputs require unified indexing rules during deployment to ensure accurate recall of documents from different equipment and inspection systems. During upgrades, the original batch association logic must be preserved to prevent historical data retrieval failures from configuration changes. Data rollback nodes must also be reserved to handle exceptions occurring after upgrades.
Configuration Settings
The following configurations apply to FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
DB_MAX_LINK | 10–20 connections | Adapts to real-time high-frequency equipment inspection data retrieval, avoids connection pool exhaustion |
MONGODB_URI | Cluster address with read-write separation suffix | Stores multi-source heterogeneous rolling-related documents, supports high-concurrency read and write operations |
REDIS_URL | Address with single-instance memory no less than 8 GB | Caches high-frequency retrieved shape and thickness data, reduces vector database pressure |
STORAGE_S3_ENDPOINT | Enterprise intranet object storage address | Stores large-volume surface quality reports and rolling specification documents, reduces local storage load |
FILE_TOKEN_KEY | Randomly generated 32-bit string | Ensures token verification security for file upload and knowledge base access |
FE_DOMAIN | Enterprise intranet domain name | Limits front-end access scope, adapts to intranet deployment requirements for production environments |
Three Common Mistakes
- Phenomenon: After upgrade, the number of recall results for equipment inspection Q&A is less than before upgrade, and document fragment relevance is insufficient. Cause: The original coil or piece-level document batch association configuration was not retained during upgrade, causing retrieval to fail to accurately match documents of the corresponding batch.
- Phenomenon: After locally deploying a large model, adding a rolling-related knowledge base triggers an error, and retrieval cannot be completed. Cause: The
DB_MAX_LINKconfiguration was not adjusted, and high-frequency real-time data retrieval exhausted the database connection pool. - Phenomenon: A 403 token unauthorized error is returned when calling the knowledge base, and the third-party service can pass verification normally when testing the model alone. Cause: The
FILE_TOKEN_KEYconfiguration was not set to the same key as the third-party service, causing FastGPT to fail token verification.
How to Verify Proper Configuration
- Run a database connection test, confirm that the number of connections configured in
DB_MAX_LINKcan cover the concurrent retrieval requirements of the production environment, and adjust the configuration to match the actual concurrency volume. - Upload a shape and thickness record document, retrieve the heating schedule document corresponding to the batch, verify that associated content can be accurately recalled, and confirm that the batch association configuration is effective.
- Check the object storage access logs, confirm that the
STORAGE_S3_ENDPOINTconfiguration can normally store and read large-volume documents. - Verify the token verification logic, confirm that the
FILE_TOKEN_KEYmatches the key of the third-party service, to avoid permission errors.
The values provided are common starting points and should be evaluated against the reader's own samples.