obtrace-sdk-python 1.2.0

Obtrace Python SDK

obtrace-sdk-python

Python backend SDK for Obtrace telemetry transport and instrumentation.

Scope

• OTLP logs/traces/metrics transport
• Context propagation
• HTTP instrumentation (requests/httpx)
• Framework helpers (FastAPI, Flask)

Design Principle

SDK is thin/dumb.

• No business logic authority in client SDK.
• Policy and product logic are server-side.

Install

pip install obtrace-sdk-python

Configuration

Required:

• api_key
• ingest_base_url
• service_name

Optional (auto-resolved from API key on the server side):

• tenant_id
• project_id
• app_id
• env
• service_version

Quickstart

Simplified setup

The API key resolves tenant_id, project_id, app_id, and env automatically on the server side, so only three fields are needed:

from obtrace_sdk import ObtraceClient, ObtraceConfig

client = ObtraceClient(
    ObtraceConfig(
        api_key="obt_live_...",
        ingest_base_url="https://ingest.obtrace.io",
        service_name="my-service",
    )
)

Full configuration

For advanced use cases you can override the resolved values explicitly:

from obtrace_sdk import ObtraceClient, ObtraceConfig, SemanticMetrics

client = ObtraceClient(
    ObtraceConfig(
        api_key="<API_KEY>",
        ingest_base_url="https://inject.obtrace.ai",
        service_name="python-api",
        env="prod",
    )
)

client.log("info", "started")
client.metric(SemanticMetrics.RUNTIME_CPU_UTILIZATION, 0.41)
client.span(
    "checkout.charge",
    attrs={
        "feature.name": "checkout",
        "payment.provider": "stripe",
    },
)
client.flush()

Canonical metrics and custom spans

• Use SemanticMetrics for the product-wide metric catalog.
• Custom spans use client.span(name, attrs=...).
• Keep free-form metric names only for truly product-specific signals that are not part of the shared catalog.

Frameworks and HTTP

• Framework helpers: FastAPI and Flask
• HTTP instrumentation: requests and httpx
• Reference docs:
  • docs/frameworks.md
  • docs/http-instrumentation.md

Production Hardening

1. Keep api_key only in server-side secret storage.
2. Use one key per environment and rotate periodically.
3. Keep fail-open behavior (telemetry must not break request flow).
4. Validate ingestion after deploy using Query Gateway and ClickHouse checks.

Troubleshooting

• No telemetry: validate ingest_base_url, API key, and egress connectivity.
• Missing correlation: ensure propagation headers are injected on outbound HTTP.
• Short-lived workers: call flush() before process exit.

Documentation

• Docs index: docs/index.md
• LLM context file: llm.txt
• MCP metadata: mcp.json

Reference