Category
observability
6 posts in this category.
OTel agent to Datadog: logs without the Datadog agent
Shipping logs to Datadog through an OpenTelemetry Collector: the exporter config, the batching that causes 413s, the hostname bug that restarts your pods, and keeping the bill in your control.
OTel agent to Loki: logs without a log shipper
The Loki exporter is gone from the Collector. What replaced it, how OTLP attributes become Loki labels, and the two limits that will reject your logs in production.
Monitoring setup for PySpark applications
Spark applications are ephemeral, so the usual scrape model does not fit. Wiring the Prometheus servlet, catching the Python memory that JVM metrics never show, and keeping evidence after the driver exits.
Monitoring setup for Airflow
Airflow 3 emits metrics through StatsD or OpenTelemetry, and the choice changes your Prometheus config. The pipeline, the metrics that predict failures, and the alerts worth paging on.
OpenTelemetry with Jaeger
Instrumentation, the Collector, and tail sampling. How the pieces fit for distributed tracing on Kubernetes, and the two things that decide whether traces are actually useful.
Grafana + Prometheus: Kubernetes monitoring that earns its keep
kube-prometheus-stack, the Operator CRDs, and how to build alerts people act on instead of a wall of dashboards nobody opens.