-
Prometheus
Metrics and alertingPull-based scraping and a query language worth learning properly.
Start with
promtool check config prometheus.ymlThe gotcha
Label cardinality is the failure mode. One unbounded label will take the server down.
-
Grafana
DashboardsVisualization over Prometheus, Loki and almost anything else.
Start with
docker run -p 3000:3000 grafana/grafana-ossThe gotcha
A dashboard nobody opens is not monitoring. Alerts are the product; dashboards are for the investigation.
-
OpenTelemetry Collector
Telemetry pipelineOne vendor-neutral path for traces, metrics and logs, so switching backends is a config change.
Start with
otelcol --config config.yamlThe gotcha
Sampling decisions made at the collector are hard to undo later. Decide what you keep before you need it.
-
Loki
Log aggregationIndexes labels rather than content, which keeps it affordable at volume.
Start with
logcli query '{app="api"} |= "error"'The gotcha
Because it does not index content, a bad label schema makes queries slow in a way more indexing will not fix.
Other categories