fix: structog instead of logging, also updated middlewares

Merged
HA
8mo ago
+921-2318

Summary by cubic

Switched backend to structlog and OpenTelemetry tracing with a single OTLP traces endpoint and tail sampling. Removed correlation IDs across APIs, events, and storage; request size limits remain enforced by RequestSizeLimitMiddleware using MAX_REQUEST_SIZE_MB.

  • Refactors

    • Replaced logging.Logger with structlog BoundLogger; logs bind OTEL span context, sanitize sensitive data, and use key=value fields.
    • Removed CorrelationMiddleware and all correlation_id usage; events link via aggregate_id; replays use replay_id with updated indexes, models, and responses.
    • Added a DI-managed Tracer (app/core/tracing/tracer.py) initialized from Settings; instruments httpx/pymongo/logging with ParentBased + TraceIdRatioBased sampling; removed custom tracing/adaptive sampling code.
    • Providers inject BoundLogger and Tracer; services/workers no longer initialize tracing.
    • Settings/CORS: dropped ENABLE_TRACING/Jaeger fields; added OTLP_TRACES_ENDPOINT; stopped exposing X-Correlation-ID and X-Request-ID.
    • Updated otel-collector to include tail_sampling (error, slow, probabilistic) in the traces pipeline.
    • Pod monitor config is provided via DI with defaults (namespace=integr8scode, kubeconfig_path=None).
    • Reverted request size enforcement to RequestSizeLimitMiddleware (dependency approach removed).

  • Migration

    • Set OTLP_TRACES_ENDPOINT and deploy the updated otel-collector with tail sampling.
    • Stop sending/expecting X-Correlation-ID and X-Request-ID; rely on OTEL trace context.
    • Update any dashboards/parsers that used correlation_id (events, indexes, exports, user settings, pod labels) to use aggregate_id or replay_id.
    • Configure MAX_REQUEST_SIZE_MB as needed; enforcement is via RequestSizeLimitMiddleware.
    • If pod monitoring relied on K8S_NAMESPACE/KUBECONFIG env vars, adjust deployment or provider configuration to the new defaults.

Summary by CodeRabbit

  • New Features

    • Structured logging rolled out for clearer, consistent logs.
    • Trace collector tail-sampling added to reduce noise (focus on error/slow traces).

  • Configuration Changes

    • Tracing config simplified to a single OTLP endpoint.

  • Refactor

    • Correlation ID removed across UI, filters, exports and APIs; replay_id introduced for replay flows and shown in replay responses.