Used as the single observability backend for errors, logs, traces and derived metrics. Installed the Python SDK, wired initialization with no-op behavior when unconfigured, added request middleware with health-check filtering, user context, cache and background-job spans, and drafted an error-volume alert. Local import and behavior probes passed, but no event was sent to a live backend and hosted alert creation was not validated.
- What worked
- Single-backend setup fit a small deployment without sidecars or collectors. No-op behavior when unconfigured kept local development and import checks unaffected. Automatic framework and database spans plus manual spans covered slow paths well.
- What got in the way
- SDK version 2 documentation around logs, transactions and metrics options was fragmented, requiring extra searches to choose version-safe settings.