Data characteristics for this category
Data sources for this category include MES systems, online inspection equipment, laboratory testing systems, process parameter logs, customer quality standards, and historical defect databases. Data is collected continuously per production batch and during shift rotations. Daily processed documents include control plans, patrol inspection records, SPC statistical reports, finished product inspection reports, nonconforming product review forms, release approval forms, and monthly quality reports. Document fields include batch number, inspection item, measured value, judgment result, responsible person, release status, and other standard fields. Some inspection data includes physical units such as millimeters, pascals, and percentage.
Constraints on deployment and upgrade
Multi-source heterogeneous data access requires support for configuring multiple distinct storage connections during deployment. The system must be compatible with multiple storage types including relational databases, object storage, and cache services. High-frequency updated production data requires deployment nodes to support high-concurrency processing. Horizontal scaling of multiple nodes is needed to meet batch document processing demands. Diverse document formats including PDF, Excel, and structured reports require pre-installing format-specific parsing dependencies during deployment. Fields include physical units and batch identifiers. Configuration must preserve field extraction precision to avoid unit loss or incorrect batch association. The historical defect database scale grows continuously as production data accumulates. Deployment must reserve sufficient storage capacity to accommodate long-term data growth.
Configuration settings
These configuration instructions apply to FastGPT v4.17.0:
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
REDIS_URL | redis://:qcpass@redis-cluster:6379/0 | Enables cache synchronization of batch data across multiple deployment nodes, preventing state inconsistency |
PG_URL | postgresql://qcuser:qcpwd@db-server:5432/fastgpt_qc | Stores structured business data such as inspection reports and review records, supporting complex query requirements |
STORAGE_S3_ENDPOINT | http://s3.qc-site.com:9000 | Stores large volumes of attachments including inspection reports and PDF documents, compatible with object storage high scalability |
DB_MAX_LINK | 20–30 | Adapts to concurrent document processing requests per production batch, preventing database connection exhaustion |
| File parsing timeout | 120–180 seconds | Processes long documents such as large SPC reports and multi-page inspection reports, avoiding parsing interruptions |
| Batch processing concurrency limit | First 10 entries | Matches the batch processing rhythm of shift patrol inspection records, preventing system resource overload |
Three common configuration errors
- An error reporting model loading failure occurs when deploying the markpdf parsing container. The root cause is that the container image does not include pre-built model files. The pre-trained model must be manually downloaded to the container’s mount directory. Without this configuration, PDF documents cannot be parsed normally.
- After deploying an updated version, the inference output length is far shorter than expected. Full processing of long documents is not possible. The root cause is that the default context length parameter is reset to a smaller value during the version update. This does not align with the long-text processing needs of inspection reports.
- When calling the MCP service, the parameter type is fixed as string and cannot be modified. The root cause is that built-in service configurations in v4.9.6 and some older versions do not allow customization of field types. This results in parameter formats that do not meet business requirements.
How to verify successful configuration
- A user accesses the FastGPT admin backend, navigates to the system configuration page to confirm that the
REDIS_URL,PG_URL, andSTORAGE_S3_ENDPOINTvalues match the deployment environment, and activates the test connection button to verify normal connectivity. - A typical finished product inspection report PDF is uploaded, and the document cross-review workflow is triggered. Parsed fields are checked to confirm they include key information such as batch number, inspection item, and measured value, with no missing or incorrect data.
- Two or more deployment nodes are started, and batch patrol inspection record documents are submitted. Node logs are checked to confirm no errors related to database connection timeouts or storage write failures, and that batch data is synchronized consistently across nodes.
- The MCP service interface is called, and a custom parameter type is passed. The returned result is verified to match the expected field format, with no forced locking to the string type.
The values provided here are common starting points. They should be evaluated against the reader's own samples.