# Sre
All Sre notes →1 — Observability Foundations
The distinction between monitoring known failure modes and observability — being able to ask new questions of a system you didn't instrument in advance for that exact question.
10 — Grafana
Turning raw telemetry into a dashboard that actually answers 'is this system healthy' at a glance instead of requiring someone to already know what's wrong.
11 — Loki
Index-light, label-based log aggregation built to pair with Prometheus's label model — and the trade-off it makes against full-text search to get there.
12 — Tempo
Object-storage-backed trace storage designed for the exemplar-driven workflow — jump from a metric spike straight to the trace that explains it.
13 — Alerting Philosophy
Alerting on symptoms a human needs to act on right now, not on every cause — the design discipline that determines whether on-call trusts the pager.
14 — Alert Fatigue
How a noisy alerting system trains engineers to ignore the pager, and why that's a more dangerous failure mode than having no alerting at all.
15 — Dashboard Design
The three-question test for a vanity panel, and the top-down layout that mirrors how an actual investigation drills down from symptom to cause.
16 — High-Cardinality Metrics
Why an unbounded label — user ID, request ID, raw URL — turns a cheap metric into a production incident for the observability pipeline itself.
17 — Sampling Strategies
Head vs. tail sampling for traces, and how to keep the interesting 1% — errors, outliers — without paying to store 100% of uninteresting requests.
18 — Cost Optimization
Ingest volume, retention, and cardinality as the three levers that actually control an observability bill, and the FinOps discipline of tuning them without losing signal.
2 — Telemetry Signals
Metrics, logs, and traces as three different projections of the same underlying system behavior, and why you need more than one to actually diagnose most incidents.
3 — Metrics
Counters, gauges, and histograms as the aggregate signal — cheap at scale, but only as useful as the cardinality budget and label schema behind them.
4 — Logs
The highest-cardinality, highest-detail signal, and the structured-logging discipline that determines whether logs are searchable evidence or just noise at 2am.
5 — Distributed Tracing
Following a single request across every service it touches, and why trace context propagation is the one piece of plumbing the rest of tracing quietly depends on.
6 — OpenTelemetry
The vendor-neutral instrumentation standard that decouples how you emit telemetry from where it ends up, and why that's the whole point.
7 — Context Propagation
How trace and baggage context survives a hop across a network boundary, a queue, or an async job — and everywhere that propagation silently breaks.
8 — Instrumentation Strategies
Deciding what to instrument, at what cardinality, before you write the code — because retrofitting observability into an incident you're already in is the expensive way to learn this.
9 — Prometheus
Pull-based scraping, the metric data model, and the local-storage limits that are exactly why Prometheus federates or remote-writes at any real scale.