What data for this category looks like
Data for this category primarily comes from ad platform APIs, attribution systems, CRM lead databases, and landing page analytics tools. Update cadence falls into two categories: real-time bidding data and daily aggregated reports. Daily documentation includes daily/weekly campaign reports, keyword bid adjustment logs, material A/B test spreadsheets, and optimization review documents. Data fields include impressions, clicks, conversions, customer acquisition cost per acquisition, budget balance, and more. Units include counts, yuan, entries, and others. Data structure primarily uses structured tables, with a small volume of unstructured review notes.
What constraints do these characteristics impose on the database and operations link
Multi-source heterogeneous data sources require the database to support both unstructured storage and structured querying. Real-time update requirements demand maintaining high concurrent connection counts to avoid data latency. Structured reports and bid adjustment logs require standardized table structures to ensure data consistency. Non-standard documents such as material test spreadsheets need flexible parsing and adaptation capabilities. Frequent operations including abnormal indicator positioning, budget allocation, and material A/B testing require database indexes to cover core query fields, reducing positioning time. At the same time, data anomalies may directly lead to increased customer acquisition costs or early budget exhaustion, so the operations link requires real-time monitoring of database load and data integrity.
How to configure settings
This document is written based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
MONGODB_URI | mongodb://ads_user:secure_pass@db-host:27017/ads_db?authSource=admin | Adapts to unstructured raw impression and click data storage from ad platforms |
PG_URL | postgresql://opt_user:db_pass@pg-host:5432/ads_optimize | Stores structured standardized tabular data such as bid adjustment logs, daily and weekly reports |
DB_MAX_LINK | 20–30 connections | Handles high-concurrency query requests for real-time bidding data, prevents connection exhaustion |
REDIS_URL | redis://:redis_pass@cache-host:6379/0 | Caches real-time hot data to reduce query pressure on the primary database |
Three common configuration errors
- Phenomenon: Importing Excel format files prompts an unsupported file error, or uploading CSV files only recognizes the first two columns of data. Cause: Multi-column data parsing rules are not configured, or the built-in table parsing and adaptation function of FastGPT is not enabled.
- Phenomenon: After calling a database query in a workflow, the generated response does not reference the query results. Cause: The query results are not bound to the context output field in the workflow node, or the target field returned by the database query is empty.
- Phenomenon: Parsing table files larger than 10MB triggers a
timeout of 360000ms exceedederror. Cause: The cache timeout configuration corresponding toREDIS_URLis not adjusted, orDB_MAX_LINKis set too low, causing connection wait timeouts.
How to confirm configuration is complete
- Run the database connection test script to verify connectivity for
MONGODB_URIandPG_URL, and check that system logs show no connection failure errors. - Upload a CSV or Excel bid adjustment record file containing multiple columns of data, confirm that all fields can be read and displayed normally after parsing.
- Trigger a real-time ad data query task, observe that the database connection count does not exceed the threshold set by
DB_MAX_LINK, with no connection overflow logs. - Run a workflow that includes a database query node, confirm that the generated response contains the specific data content returned by the query.
The values listed are common starting points and should be measured against samples from the reader’s own use cases.
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.