Principle:Ollama Ollama Anthropic Trace Logging
| Knowledge Sources | |
|---|---|
| Domains | Observability, Debugging |
| Last Updated | 2025-02-15 00:00 GMT |
Overview
Anthropic Trace Logging is the principle of capturing full HTTP request and response payloads at the API compatibility layer for diagnostic and debugging purposes. This observability pattern enables developers to inspect the exact data flowing through protocol translation boundaries, identify mismatches between expected and actual API behavior, and diagnose issues in multi-layer proxy architectures.
Core Concepts
HTTP Trace Capture
HTTP trace capture involves recording the complete request and response lifecycle at a specific interception point. This includes HTTP method, URL, headers, request body, response status code, response headers, and response body. For streaming responses, trace capture must handle chunked transfer encoding and Server-Sent Events, potentially buffering or tee-ing the response stream so that logging does not interfere with data delivery to the client. Trace capture is typically gated behind a debug flag or environment variable to avoid performance overhead and sensitive data exposure in production.
Structured Diagnostic Output
Raw HTTP dumps are difficult to parse visually. Structured diagnostic output formats trace data in a human-readable way, separating request and response sections, pretty-printing JSON bodies, and annotating timestamps and durations. This may involve deserializing JSON payloads for indented formatting, truncating excessively large bodies, and highlighting key fields such as model name, token counts, and error codes to accelerate debugging workflows.
Boundary Logging
In systems with protocol translation layers, boundary logging captures data at the exact point where one protocol is converted to another. This is particularly valuable because it allows developers to see both the original incoming format and the translated outgoing format side by side, making it straightforward to identify translation errors, missing fields, or semantic mismatches between API specifications.
Implementation Notes
In the Ollama codebase, Anthropic trace logging is implemented as a dedicated trace utility that can be enabled to log the full HTTP request and response payloads passing through the Anthropic compatibility layer. When active, it captures the incoming Anthropic Messages API request body, the translated Ollama-native request, the native response, and the translated Anthropic-format response. This four-point capture enables developers to pinpoint whether issues originate in the incoming request, the request translation, the inference backend, or the response translation.