oneport-debug-tracer

Cross-service distributed-trace root cause analysis. Give it a trace ID; it maps every span to the exact line of code across all your repositories — even when a single request crosses a React frontend, a Go API gateway, and a Python payment service — and tells you what broke and why.

This is the gap Claude Code and Copilot can't close: they trace code within one repo. A real production request fails across many.

Try it in 60 seconds (no Datadog, no API key)

pip install oneport-debug-tracer
oneport-tracer demo

The demo runs the real span-to-code engine against a bundled sample trace and a mini three-language monorepo (TypeScript → Go → Python). You'll see each span resolved to its exact file:line and a full root-cause analysis of a realistic cross-service bug — a checkout 500 whose true cause is a broken contract with the auth service.

  Span (call tree)     Service           Duration   Status    Matched code location
 ─────────────────────────────────────────────────────────────────────────────────
  submitCheckout       frontend           1240 ms   ● error   frontend/src/checkout.ts:7
    └─ RouteCharge     api-gateway        1150 ms   ● error   api-gateway/handlers/charge.go:6
      └─ charge_card   payment-service       8 ms   ● error   payment-service/src/payment.py:8

Add --format json to get machine-readable output for Jira/CI pipelines.

Use it on your own traces

# 1. Authenticate (stores your license key)
oneport-auth login

# 2. Point it at your trace platform + repos
export DD_API_KEY=…  DD_APP_KEY=…
export ANTHROPIC_API_KEY=…           # or run fully local: see oneport-debug-local

oneport-tracer analyze \
  --trace-id <id-from-pagerduty> \
  --platform datadog \
  --repos repos.yaml

repos.yaml maps service names (as they appear in your traces) to git remotes or local clones:

payment-service: git@github.com:acme/payment-service.git
api-gateway:     git@github.com:acme/api-gateway.git
frontend:        git@github.com:acme/frontend.git

Supported platforms: datadog, splunk, otel, jaeger, tempo. Supported languages for code mapping: Python, Java, Kotlin, Go, TypeScript/JavaScript.

How it works

1. Ingest all spans for the trace from your APM platform.
2. Index every repo's symbols and HTTP routes (Python AST + tree-sitter for Java/Kotlin/Go/TS, with a regex fallback when tree-sitter isn't installed).
3. Match each span to its source line by operation name, HTTP route, and qualified Class.method notation.
4. Analyze the failing path with an LLM (cloud or fully on-prem) and emit a structured RCAResult — ready for the terminal, Jira, Slack, or a SIEM.

Secrets are scanned and redacted from every prompt before it leaves your machine, and every analysis is written to a tamper-evident, hash-chained audit log.

Output

A structured root-cause analysis: severity, root cause, confidence, the exact affected code locations (repo · file · line · function · git-blame author), and a recommended fix. Part of the OnePort Debug enterprise debugging suite.

License

Apache-2.0