Skip to content

Technical3 min read

Technical dimension

Building observability that holds up in production. Signals, OpenTelemetry, pipelines, backends, SLOs, AI and integrity.

A doctor does not order every possible test for every patient. They pick the ones that help reach a diagnosis, without bankrupting the hospital. The technical dimension looks for the same thing for an IT system: how to get, at the right cost, the signal that lets you understand an outage.

Every chapter starts from a concrete operations problem and gives the full configuration, tested in a lab, a small environment you start on your own machine. The Labs page says which ones are published.

If you are new to the topic, start with the first three chapters of the table, in order: they set out the vocabulary used everywhere else.

ChapterWhat you learn to doStatus
Observability in 10 minutestell monitoring from observability, understand metrics, logs, traces, profiles and events, link them, reason about cardinalitypublished, for beginners; cardinality simulator
OpenTelemetry end to endfollow a trace step (a span) from the application to storage, instrument, configure a minimal Collector, apply the standard’s naming conventionspublished, for beginners
SLOs and alertingset service quality targets (SLIs and SLOs), compute the tolerated margin of error (the error budget), alert when it melts too fast (the burn rate) instead of on thresholdspublished, for beginners; error budget simulator
Cardinality explainedunderstand how a label with unlimited values multiplies series and storage, find the culprit, fix it and cap itpublished, with a video and the cardinality-cost-lab
Who is growing the metrics bill?find the metrics and labels that push the bill up, in three questions, then keep a monthly ranking shared with financepublished
Entities, Profiles, Injectorunderstand three OpenTelemetry pieces and their maturity: entity model, profiles signal, agent injectionpublished, in the OpenTelemetry section
OpenTelemetry and reversibilitytell data portability from operational reversibility, measure a time to exitpublished, in the OpenTelemetry section
Collection pipelinesdesign a Collector: filtering, transformation, sampling, routingplanned
Backends and storagechoose between Prometheus, VictoriaMetrics, Elastic, Loki, Tempo, ClickHouseplanned
AI observabilityobserve inference, RAG and MCP tools with an open source stackpublished, with a full course
Log integritybuild an audit log a third party can verifypublished
HPC and GPU infrastructuretie every GPU metric to a job and an account, catch silent degradationreference architecture, course in preparation
Agents, RAG and MCPper-component measurement grids, probabilistic quality SLOsGenAI method
VictoriaMetricsa metrics backend for high-cardinality workloadscourse
Edge and IoTobserve constrained, intermittent and heterogeneous fleetsplanned

To see these ideas move rather than read about them, try the trace explorer, the clickable architecture, the animated Collector pipeline, tail sampling or the quality SLO simulator. The rest are gathered on the tools and diagrams pages.