June 1, 2026
The OpenTelemetry GenAI semantic conventions reached stable status in early 2026, and if you are running LLM-backed features in production, you now have a vendor-neutral way to capture token counts, model latency, prompt content, and cost …
May 25, 2026
OpenTelemetry reached CNCF graduation in May 2026. That's a meaningful signal — the project is stable, the APIs are locked, and the ecosystem is mature enough that betting your observability pipeline on it is no longer …
May 18, 2026
Grafana's Adaptive Logs feature, which shipped to Grafana Cloud recently, uses a sampling model to automatically drop high-volume, low-signal log lines. The pitch is compelling: cut your log ingestion bill without writing a single drop rule …
May 11, 2026
Skyscanner's OpenTelemetry write-up landed last week and it's one of the more honest operational accounts of running OTel at scale that I've seen from a large engineering org. They're managing 24 production Kubernetes clusters with a …
May 4, 2026
Skyscanner published a detailed post this week on how they manage OpenTelemetry Collectors across 24 production Kubernetes clusters. The architecture decisions they made — and the ones they avoided — are a useful stress test for …
April 24, 2026
Buyers who type this query into Google want three things: a defensible architecture, retention they can explain to Finance, and a cost model that includes S3 and query path, not only ingest. We start from a …
April 24, 2026
This is an opinionated cost teardown, not a vendor hit piece. We name the mechanisms that inflate hosted observability bills: custom metrics, indexed logs, APM cardinality, and seat-adjacent pricing. Then we show what moves when you …
April 24, 2026
Tail sampling is where OpenTelemetry configs go to die: policy mistakes show up as missing traces in prod, not as a failing unit test. This is a working collector shape we use when buyers search for …
April 24, 2026
Numbers are anonymized; failure modes are not. The useful part of a vendor exit is rarely the architecture diagram alone. It is what broke in week six when cardinality, ownership, and on-call habits collided with the …
April 24, 2026
Between one and five terabytes of logs per day, every storage engine becomes a finance and operations problem, not a benchmark chart. Here is how we compare Grafana Loki, Elasticsearch, and ClickHouse after running all three …
April 24, 2026
How we size Grafana Mimir clusters for 10M active series on AWS EKS — replica counts, cost tradeoffs, and the failure modes to plan for.
July 15, 2025
Observability is not free, but the cash story is usually MTTR, fully loaded engineer hours in bridges, and AWS lines nobody charts until Finance asks. We separate real savings from license theater for teams leaving SaaS …
July 15, 2025
Monitoring is the coverage you wired in advance. Observability is whether you can answer the next question without shipping a new metric for every hypothesis. Here is how we use both when we move teams to …
January 14, 2025
Logs feel definitive; traces feel modern. Metrics still anchor SLOs, autoscaling, and cardinality discipline on Prometheus and Mimir. This is the case we make when a team wants to log everything 'just in case.'