Deep dives on distributed tracing, real-time metrics, OpenTelemetry, SRE practice, and building for production scale.
Head-based sampling makes a keep/drop decision before the trace is complete. Tail-based sampling waits — and that changes everything. We walk through the architecture, the trade-offs, and why obseria.io processes every span before deciding what to store.
Lena Hartmann
Staff Engineer
After operating the OTel Collector for 400+ customer tenants, we have opinions. Batching strategies, memory limiters, queue sizing, and the one config change that cut our CPU usage by 40%.
Marco Bietti
Platform Engineer
One of our customers was receiving 30 pages a day. After switching to obseria.io's ML anomaly detection, they're down to 12 pages per month — all of them actionable. Here's exactly how the model works.
Priya Mehta
ML Engineer
The ecosystem has improved dramatically — eBPF auto-instrumentation, native OTLP in kubelet, and proper sidecar-less tracing. But cross-namespace correlation is still a mess. Our field notes from production.
Sam Okonkwo
DevRel Engineer
N+1 query problems are invisible in conventional APM — they look like normal latency. Distributed tracing exposes them as repeated child spans. We built an automated N+1 detector on top of our span tree, and here's how it works.
Lena Hartmann
Staff Engineer
They solve similar problems differently. Prometheus pulls; OTel pushes. Prometheus has a richer ecosystem; OTel has vendor neutrality. In practice, you'll run both. Here's our recommended architecture.
Marco Bietti
Platform Engineer
Most postmortem templates are too long and get skipped. Ours is three sections, takes 20 minutes to fill out, and links directly to the obseria.io trace from the incident. We're sharing the template and the reasoning.
Yuki Tanaka
SRE Lead
Most tools charge per unique series. High-cardinality labels like user_id or request_path can explode your bill overnight. We explain our per-data-point pricing model and show you exactly how to structure labels for cost efficiency.
Ana Sousa
Product Engineer
eBPF lets us attach probes directly to kernel network calls — meaning we can trace Go, Rust, JVM, and native binaries without SDK changes. We shipped this in obseria.io 3.2. Here's how it works under the hood.
Carlos Vega
Systems Engineer
Browse by topic
We publish 2–3 deep dives per month on observability engineering, SRE practice, and OpenTelemetry. No marketing, no noise.
