What the data for this category looks like
Data involved in inspection and process record audit for petroleum refining comes from DCS/APC control systems, LIMS laboratory information management systems, tank farm metering systems, crude oil evaluation databases, equipment management systems, and energy management systems. Data update frequencies fall into three categories: real-time, shift-based, and daily. Daily documents include operating procedures, production plans, material balance sheets, distillate export quality reports, interlock ledgers, shift handover records, and change management documents. Core fields include unit number, raw material property parameters, product fraction indicators, interlock trigger conditions, and execution records.
What constraints these characteristics impose on deployment and upgrade
The data sources for this category are scattered, and update rhythms vary significantly. The deployment phase must adapt to access protocols for multiple heterogeneous systems. The upgrade phase must be compatible with data synchronization logic for different update frequencies. The complex document structure and multi-dimensional fields require configuring storage rules adapted to long text and multiple fields during deployment. This avoids data truncation or index loss. Additionally, the real-time requirements of process fluctuations and interlock events require reserving sufficient concurrent processing resources during deployment. Upgrades must ensure synchronization of core business data is not interrupted. This prevents abnormalities in production links caused by configuration changes.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
DB_MAX_LINK | 200–300 connections | Adapts to concurrent query and real-time data synchronization needs across multiple systems and departments, avoiding connection exhaustion |
MONGODB_URI | Points to a dedicated replica set address for LIMS and tank farm metering data | Isolates production business data from other business databases, ensuring data security and query efficiency |
REDIS_URL | Configured as a dedicated cache instance for local private deployment | Caches real-time process parameters and shift handover records, reducing overhead from repeated database queries |
STORAGE_S3_ENDPOINT | Points to the intranet address of the enterprise’s internal object storage service | Stores large-volume crude oil evaluation reports and interlock ledgers, reducing public network transmission latency |
FE_DOMAIN | Configured as the enterprise intranet domain name | Restricts access scope, adapting to private deployment scenarios without external network access |
ROOT_KEY | Generates a strong key in accordance with enterprise internal permission management rules | Secures access permissions for system background configuration, preventing unauthorized modifications |
Three common mistakes
- Symptom: After private deployment without external network access, the workflow editing interface cannot be opened, and a client application error is displayed on the page. Cause: The
FE_DOMAINconfiguration uses a public network domain name, which cannot be resolved in the intranet environment, leading to failure to load front-end resources. - Symptom: Request timeouts occur during concurrent model calls, with single concurrency exceeding the preset threshold. Cause: The
DB_MAX_LINKconfiguration value is too low, failing to support simultaneous queries of process data and inspection records across multiple departments. - Symptom: The Qwen2.5 model deployed using Ollama cannot trigger tool calls, and workflow tests have no response. Cause:
REDIS_URLis not configured as a cache middleware, resulting in failure to persist session states for tool calls.
How to confirm the configuration is correct
- Complete the basic deployment using the FastGPT v4.17.0 deployment script, then access the address configured by
FE_DOMAINin the intranet environment. Confirm that the front-end interface loads normally without resource loading errors. - Organize personnel from multiple departments to simultaneously initiate process data query requests. Verify that the
DB_MAX_LINKconfiguration supports concurrent requests, with no connection exhaustion-related errors. - Upload large-volume crude oil evaluation reports. Confirm that the storage service configured via
STORAGE_S3_ENDPOINTcan normally receive and read files. - Trigger a tool call test. Confirm that session states can be cached normally without state loss issues.
The values provided are common starting points and should be measured against the reader's own samples.