Data characteristics for this use case
Data for this category comes primarily from unfolded drawings exported from CAD/CAM software, CNC cutting programs, bending process cards, order BOMs, sheet metal inventory ledgers, operation data from laser cutters and bending machines, and process flow records from MES systems. Data updates align with order delivery timelines, daily production schedules, or each equipment changeover. Each document includes fields such as sheet material specifications, bending compensation coefficients, cutting dimensions, process flow nodes, and first-piece inspection results. Units include millimeters, seconds, square meters, and others, with a mix of structured and semi-structured formats.
Constraints for database and operations
The high-frequency update pattern of this data creates sudden peaks in database write requests, requiring support for non-fixed batch data write scenarios. Semi-structured process drawings, cutting programs, and structured order and inventory data require databases to support both precise queries on structured fields and index management for unstructured files. Multi-source heterogeneous device and system data have significant format differences, so unified data preprocessing rules must be built to avoid missing fields or format incompatibilities. Multi-dimensional data association needs also require database table structure designs that balance query efficiency and data consistency.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
PG_URL | postgresql://username:password@host:5432/fastgpt?sslmode=disable | Stores structured order and process data for sheet metal cutting and bending scenarios. Uses default PostgreSQL port 5432, disables SSL to reduce unnecessary overhead, and meets database connection requirements for FastGPT v4.17.0. |
DB_MAX_LINK | 15–20 | This scenario has large batch data write volumes. Connection count must cover concurrent needs for batch tasks and daily queries, calibrated via on-site testing. |
REDIS_URL | redis://:password@host:6379/0 | Caches frequently queried sheet metal inventory and bending coefficient data. Uses default Redis port 6379, configure database index according to deployment environment. |
| Reranking | Enable and set the top 3–5 candidate results | Retrieval results for this scenario need to match precise process parameters. Limiting the number of candidates reduces reranking latency, and adapts to 8c16G non-GPU deployment environments. |
| Query Optimization | Enable and set maximum context to 8000 token | Process descriptions for sheet metal cutting and bending are lengthy. Sufficient context ensures accurate question understanding, and adapts to resource constraints of non-GPU hardware. |
| Single Search Limit | Top 10–15 results | Data sources for this scenario include multiple types of process documents. Expanding the candidate range improves retrieval recall rate, while avoiding excessive results slowing overall workflow. |
Three Common Mistakes
- Symptom: When deploying a PostgreSQL database via Docker, knowledge base index creation fails, and the interface displays "database connection timeout" or "field format mismatch" errors. Cause: Failed to adapt to semi-structured document fields for sheet metal cutting and bending scenarios, failed to pre-create structured fields for sheet thickness and bending coefficients in the database, or connection parameters were not correctly configured.
- Symptom: After enabling query optimization and result reranking, message response time exceeds 30 seconds. Cause: The number of reranking candidate results was not limited, and the configuration was not adapted to 8c16G non-GPU hardware, causing reranking calculations to occupy excessive CPU resources.
- Symptom: Database connection pool is exhausted, displaying a "too many connections" error. Cause: The
DB_MAX_LINKparameter was not adjusted based on the scenario's batch data write requirements, and the connection count was set too low to cover concurrent write requests.
How to Verify Proper Configuration
- Run a database connection test to verify that configuration items such as
PG_URLandMONGODB_URIcan establish normal connections, with no authentication or network errors. - Upload a sheet metal cutting and bending process card document to trigger an indexing task, and check whether the index status completes normally with no missing field prompts.
- Enable query optimization and result reranking functions, initiate multiple process-related queries, and observe whether response times meet expected ranges for the deployed hardware.
- Simulate batch order data import, check database write speed and connection count usage, and confirm no connection exhaustion or write timeouts occur.
The values provided are common starting points and should be measured against the reader's own samples.