What this category of data looks like
Data for this category primarily originates from advertising platform APIs, attribution systems, CRM lead databases, and landing page statistics tools. Data update cycles are divided into real-time (such as real-time impression and click data) and daily (such as daily conversion and customer acquisition cost reports). Daily produced documents include daily and weekly advertising reports, keyword adjustment records, material A/B test sheets, and optimization effect review reports. Fields include impressions, clicks, conversions, budget amounts, material click-through rates, and similar metrics. Units include counts, yuan, percentages, and other standard units.
Constraints on deployment and upgrade
Multi-data-source access requirements mandate configuration of stable cross-system connection parameters during deployment, to ensure normal data pulling from advertising platforms, attribution systems, and other data sources. Real-time data update cycles require adjustment of vector index refresh thresholds during deployment, to avoid data lag impacting optimization judgments. Daily documents include tabular adjustment records and material test sheets, so multi-format file parsing configuration must be adapted to ensure completeness of knowledge base training. The operations on-duty and ticket scenario requires configuration of permission isolation rules to distinguish access scopes for different roles such as optimizers and analysts. During upgrade processes, real-time data links must be maintained without interruption to prevent delays in advertising adjustments.
Configuration settings
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Approach | Rationale |
|---|---|---|
MONGODB_URI | mongodb://username:password@host:port/dbname?authSource=admin | Stores advertising historical data, ticket records, and knowledge base documents; requires a highly available connection |
REDIS_URL | redis://:password@host:port/0 | Caches real-time advertising data, conversation contexts, and ticket status; recommended timeout setting is 300 seconds |
STORAGE_S3_ENDPOINT | Fill in according to the local object storage service address, such as http://192.168.1.100:9000 | Stores knowledge base uploaded resources such as material files and report documents; adapts to multi-format file reading requirements |
FILE_TOKEN_KEY | Randomly generated 32-bit string | Encrypts verification tokens for file uploads, prevents unauthorized submission of knowledge base files |
ROOT_KEY | Custom complex string | Configures the system's highest-privilege access key, used for background operations in maintenance scenarios and ticket review |
FE_DOMAIN | Public or internal domain name for the deployed service, such as http://fastgpt.example.com | Configures the frontend access address to ensure normal redirection for internal tickets and knowledge bases |
Three common misconfigurations
- Knowledge base file upload results in empty data processing. The cause is incorrect configuration of
STORAGE_S3_ENDPOINTor insufficient permissions for the local storage path, making it impossible to read uploaded documents such as material test sheets and adjustment records during file parsing. - Vector indexes operate normally when run locally, but vector scores remain consistent after packaging the Docker image. The cause is failure to synchronize environment variables for
REDIS_URLandMONGODB_URI, causing the container to fail to read cached real-time advertising data, with missing data sources relied on for vector calculation. - First response works normally after calling a local large model, but subsequent multi-turn context conversations throw errors. The cause is failure to adjust the session timeout configuration for
REDIS_URL; the default timeout is too short, causing context cache to be cleared prematurely, making it impossible to load historical conversations and advertising data.
How to verify correct configuration
- Conduct a knowledge base file upload test: upload an adjustment record spreadsheet, verify that data processing progress completes within 120 seconds and vector indexes are generated.
- Launch a multi-turn conversation test: input three consecutive questions containing different advertising metrics, confirm that conversation contexts load normally and responses are associated with corresponding data.
- Check the connection status of
REDIS_URL: use maintenance tools to view whether the number of cached sessions and expiration times meet configuration requirements. - Verify permission isolation for the ticket system: use keys for different roles to attempt accessing sensitive budget adjustment tickets, confirm that unauthorized roles cannot view corresponding content.
The values provided are common starting points and should be evaluated against samples relevant to the deployment context.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-10-19.