What data this category includes
This category includes structured log entries, distributed tracing Span information, and time-series monitoring metric samples. Data sources are application instrumentation SDKs, centralized logging systems, distributed tracing platforms, and time-series metric systems. Update frequency is daily or aligned with application version iterations. Supporting daily documentation includes logging collection specifications, tracing instructions, metric dimension dictionaries, and dashboard configuration records. Fields include timestamps, service identifiers, log levels, trace IDs, metric values and corresponding units. Metric units vary by type: request volume uses "count", latency uses "milliseconds".
What constraints these characteristics impose on model access and configuration
Multi-source heterogeneous data sources require models to adapt to different formats of logs, tracing data, and metric data during access. The configuration phase must identify core fields associated with data (such as trace IDs) to enable context aggregation. Data is updated alongside version iterations, so context rules for model calls must remain consistent with the latest collection specifications and metric dictionaries. Single log and tracing data entries have high token usage, so the total context length for a single input must be limited to avoid exceeding model support limits. Test case and defect analysis rely on complete contextual evidence, so configuration must retain sufficient associated information while avoiding redundant data that interferes with model inference.
The following configuration items are designed for the logging platform and observability scenario with FastGPT v4.17.0.
How to set configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
AIPROXY_API_ENDPOINT | Root address of the AI Proxy service, for example http://aiproxy:3000. Do not append /v1 or /chat/completions paths | Valid root address configured for the AI Proxy service, ensuring correct model call routing |
AIPROXY_API_TOKEN | Admin token matching the AI Proxy service configuration, cannot be empty | Used for identity authentication with the AI Proxy service, unrelated to vendor credentials configured for model channel settings |
maxContext | 8000–12000 tokens | Balances contextual completeness of log and tracing data with model inference efficiency, adapted for single-batch defect analysis scenarios |
maxTokens | 2000–3000 characters | Limits model output length, preventing defect analysis results from exceeding display or storage thresholds |
quoteMaxToken | 1500 tokens | Retains sufficient original log and tracing fragments as reference evidence, supporting traceability requirements for test cases |
temperature | 0.3–0.5 | Reduces output randomness, ensuring results for defect analysis and test case generation meet the rigor requirements of operations scenarios |
Three common configuration errors
- Incorrectly filling the
AIPROXY_API_ENDPOINTwith a single-model service address (such as a VLLM/Ollama/Xinference address) or appending paths like /v1, which causes AI Proxy service startup verification failures or model call exceptions - Failing to re-verify context-related parameters after upgrading versions, leading to context length exceeding the support thresholds of the new model version. This manifests as knowledge base queries returning data but the model producing no output, consistent with common scenario issues.
- Not setting a reasonable value for
quoteMaxToken, resulting in truncated referenced log fragments that prevent complete defect localization. This manifests as missing critical trace association information in test cases.
How to confirm proper configuration
- Check App/Admin service startup verification logs to confirm
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKENare correctly configured and not empty - Import log and tracing data from a single test case, trigger model inference, and check if the output includes associated trace IDs and log level information
- Adjust the
temperatureparameter value, observe consistency in model output, and confirm results meet the rigor requirements of operations scenarios - Review model call log records to confirm that
maxContextandquoteMaxTokenvalues do not trigger model token limit error messages
The values provided are common starting points and should be evaluated against the reader's own samples.
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.