What the data for this category looks like
Chemical raw material investment research data primarily comes from MSDS safety data sheets, bulk commodity spot trading platforms, industry association monthly monitoring reports, patent public databases, and downstream manufacturer supply and demand meeting minutes. Data update rhythms vary: spot price data updates daily, industry compliance standard documents update quarterly, and patent data syncs in real time. Individual document lengths vary widely, from hundreds of words of ingredient descriptions to tens of thousands of words of full-process compliance reports. Fields include CAS registry number, molecular formula, production batch number, purity index, transaction unit price, compliance number, and more. The unit for unit price is yuan per ton, and purity is marked as mass percentage.
What constraints do these characteristics impose on conversation logging and auditing?
The multi-source update rhythms and varied document lengths of chemical raw material data create multiple constraints for the conversation logging and auditing process. First, long documents and multi-turn conversations extend context length. Conversation logs must support large-field storage and paginated retrieval to avoid storage overflow or retrieval lag. Second, version differences across multiple data sources require the auditing process to associate data version identifiers. This ensures that compliant or spot data from the correct time node is used, preventing expired standards from being applied. Third, dedicated fields such as CAS numbers and compliance numbers must be included as core log fields, enabling quick session tracing by category. In addition, some data is subject to compliance and regulatory requirements. Audit logs must retain the called document version number to meet compliance auditing needs.
How to set the configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
LOG_STORAGE_ENABLE | Enabled | Chemical raw material investment research sessions involve compliance data tracing, so complete conversation records must be retained. When enabled, sessions can be retrieved via the backend audit panel |
LOG_STORAGE_MAX_SIZE | 1500–2500 MB | A single compliance document is approximately 20,000 to 50,000 words, corresponding to approximately 100–400 KB of log field usage. This range can cover storage requirements for at least 5000 sessions |
AUDIT_FIELD_INCLUDE | ["cas_no", "trade_price", "compliance_code"] | Dedicated chemical raw material fields including CAS number, transaction unit price, and compliance number must be included in audit logs to enable tracing by category and data dimension |
SESSION_HISTORY_PAGE_SIZE | 20–30 | In multi-user scenarios, historical sessions are pulled in pages by customUid. This range balances query efficiency and the volume of data returned in a single request |
SSE_CONNECTION_TIMEOUT | 600 seconds | Long document parsing and multi-turn conversations may take a long time. This setting prevents client-side SSE interruptions from causing incomplete session log storage |
MONGO_CONNECTION_TIMEOUT | 30 seconds | Chemical raw material knowledge bases have large data volumes. Sufficient time must be reserved for database connections to prevent interface login failures after startup |
The parameter values provided on this page are common starting points for configuration. Actual values are affected by material form, data volume, and business rules. Specific issues require case-by-case analysis. It is recommended to test on your own samples before finalizing settings.
Three common mistakes
- Issue: Conversation logs are not fully written to the database after the client actively disconnects the SSE connection. Cause: The
SSE_CONNECTION_TIMEOUTparameter was not adjusted. The default timeout period is too short, and the session is interrupted before writing is complete. - Issue: The interface cannot be logged in after the container starts, logs prompt MongoDB connection failure, and manual verification of MongoDB account passwords has no response. Cause: The
MONGO_CONNECTION_TIMEOUTparameter is set too small. Chemical raw material knowledge bases have large data volumes, and database connection handshake time exceeds the threshold. - Issue: The conversation history interface returns all application sessions, and cannot filter target user sessions by customUid. Cause: The
custom_uidfield is not configured inAUDIT_FIELD_INCLUDE, or the customUid parameter is not passed when calling the interface.
How to confirm the configuration is complete
- Enter the FastGPT backend audit panel, retrieve sessions for a specified customUid, and confirm that the returned results only include conversation records for the corresponding user.
- Manually disconnect the SSE connection, then check whether complete conversation logs and associated chemical raw material dedicated fields are retained in the database.
- View the container startup logs, confirm that there are no timeout errors for MongoDB connections, and that the interface can log in normally.
- Call the conversation history interface with the customUid parameter, and verify that the returned results match the expected number of sessions.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-14.