5.1万- Resources
- Engineering
ClickHouse engineering resources
Percentiles vs averages: why your latency dashboard lies
Latency distributions are right-skewed, so the mean hides the slow tail users actually feel. What p50, p95, and p99 mean, why you cannot average percentiles, and how to compute them at scale.
Last updated: 2026年7月29日
OpenTelemetry's profiles signal enters public alpha
OpenTelemetry profiles, the fourth telemetry signal, entered public alpha in March 2026, with OTLP support, collector pipelines from v0.148.0, and a whole-system eBPF profiler.
Last updated: 2026年7月29日
OpenTelemetry semantic conventions explained
OpenTelemetry semantic conventions standardize telemetry attribute names like http.request.method and db.system.name, so data from any service is portable and queryable across tools.
Last updated: 2026年7月29日
OpenTelemetry histograms explained: buckets, exponential histograms, and percentiles
How OpenTelemetry histograms summarize distributions: explicit-bucket vs base-2 exponential encodings, how percentiles are computed from buckets and their error bounds, and how histogram data is stored and queried.
Last updated: 2026年7月29日
Observability vs visibility: the terms teams conflate
Visibility means seeing predefined surfaces; observability means interrogating internal state for unknowns. Why network teams say one and software teams say the other.
Last updated: 2026年7月29日
Observability vs monitoring: what''s actually different
Monitoring watches known indicators for known failures; observability lets you ask new questions of a live system. How they differ in data, cardinality, and cost.
Last updated: 2026年7月29日
Observability architecture: how the pieces fit together
The six layers of an observability architecture, from OpenTelemetry SDKs to alerting, and the four design decisions that determine what debugging costs at scale.
Last updated: 2026年7月29日
What is MTTR? MTTD, MTBF, and the incident metrics that matter
MTTR is the average time from incident start to service restoration. Learn the formulas behind MTTR, MTTD, MTTA, and MTBF, what a good target looks like, and how observability shrinks each one.
Last updated: 2026年7月29日
LLM inference latency: TTFT, tokens per second, and what to measure
LLM inference latency breaks down into TTFT, time per output token, throughput, and end-to-end latency. What each metric means, what moves it, and how to measure it with OpenTelemetry.
Last updated: 2026年7月29日
LLM guardrails: what they are and how to run them in production
LLM guardrails are programmatic controls that validate model inputs and outputs to block prompt injection, PII leaks, and off-topic responses, and every verdict they emit is an event worth monitoring.
Last updated: 2026年7月29日
The incident response process: from alert to postmortem
The incident response process is the structured sequence teams follow when production breaks: detect, declare, mobilize, mitigate, resolve, and learn, with severity levels, defined roles, and blameless postmortems.
Last updated: 2026年7月29日
What are the four golden signals of monitoring?
The four golden signals (latency, traffic, errors, and saturation) are the minimum metric set the Google SRE book recommends for monitoring any user-facing system.
Last updated: 2026年7月29日
Gauge vs counter: Prometheus metric types explained
A counter only increases and is read through rate(); a gauge moves up and down and is read directly. How the four Prometheus metric types differ, why counters reset, and how to pick between them.
Last updated: 2026年7月29日
AI SRE agents: can AI run your on-call?
AI SRE agents are LLM-based systems that investigate production alerts by querying logs, metrics, and traces. What they do well today, where they still fail, and what they need underneath.
Last updated: 2026年7月29日
How to engineer cost-efficient open source observability with ClickHouse (ClickStack) - 2026 technical playbook
A technical guide for engineers to build a cost-efficient observability stack. Learn how to use ZSTD codecs, Materialized Views, and Tiered Storage with ClickHouse to reduce observability data footprint by 10x.
Last updated: 2026年1月7日