What data for this category looks like
Data sources include slab or billet batches, heating furnaces, mill automation equipment, online thickness and width measurement devices, surface inspection systems, and MES systems. Update frequency is per coil or per steel piece batch level, real-time. Daily documents include rolling plans, heating schedules, rolling specifications, shape and thickness records, surface quality reports, and quality certificates. Document fields cover heating temperature, rolling reduction parameters, thickness tolerances, surface defect grades, and other parameters, using industry-standard process parameter units.
What constraints these characteristics impose on deployment and upgrade
This category’s data is real-time and updated per coil or batch. During deployment, storage configurations must support high-frequency vector ingestion and retrieval to avoid retrieval latency. Coil-level traceability requirements mandate that the system binds billet batch IDs to process documents, so precise retrieval rules configured by batch fields are necessary. For scenarios with multiple types of process documents, deploy a preprocessing flow that supports multi-format parsing. Since process adjustments directly affect product quality, the upgrade phase must retain a historical version rollback mechanism to enable fast recovery during anomalies.
How to set configurations
The configuration rules on this page apply to FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
REDIS_URL | redis://:custom password@host address:6379/0 | Real-time data caching for this category requires stable memory storage. Port 6379 is the default Redis port, and database 0 is used for caching process retrieval contexts. |
PG_URL | postgresql://custom username:custom password@host address:5432/fastgpt_steel_roll | Coil-level batch data requires transactional storage from a relational database. Port 5432 is the default PostgreSQL port, and the specified database name simplifies business scenario management. |
STORAGE_S3_ENDPOINT | http://custom storage host address:9000 | Large documents such as shape and thickness records and surface quality reports require S3-compatible object storage. Port 9000 is the default MinIO service port. |
DB_MAX_LINK | 20–30 connections | This category has high concurrent real-time data writes and retrievals. A reasonable connection count prevents database connection exhaustion. |
| Single retrieval return limit | Top 10 entries | There are many process specification documents. Limiting return entries avoids context overflow and aligns with the precise needs of coil-level retrieval. |
| Document chunk length | 1000–1500 characters | Documents such as rolling specifications contain continuous process parameters. Too large a chunk causes semantic fragmentation, while too small a chunk increases retrieval overhead. |
Three common mistakes
- When running
docker compose up -d, the return isError response from daemon: connection refused. The cause is that the host address and port inPG_URLorREDIS_URLare not filled correctly, preventing dependent services from initializing connectivity. - Knowledge base search responses time out. The cause is that a reasonable
DB_MAX_LINKconnection count is not configured. The database connection pool is exhausted during concurrent retrievals, preventing timely request responses. - After enabling the thinking toggle, locally deployed model thinking content cannot be collapsed. The cause is that
FE_DOMAINis not correctly configured to the actual frontend access domain name, leading to abnormal loading of frontend interaction scripts.
How to confirm configuration is complete
- Run the
docker compose pscommand to confirm all service containers are running with no abnormal exit records. - Upload a rolling process document to verify the parsing process has no errors and chunking results match preset rules.
- Submit a process parameter query to confirm retrieval results include matching process documents and retrieval response meets preset business performance requirements.
- Enable the thinking toggle and submit a query to verify frontend interaction components load correctly and support collapsing thinking content.
All configuration values listed here are common starting points. Adjust them based on your own specific samples and requirements.