PCA study plan

Stand up a local Prometheus instance with a scrape config that targets at least three distinct job types and verify targets show UP in the UI

Write 15 PromQL instant queries across selectors, label matchers, and range vectors, and record the result type each returns

Explain out loud the difference between gauge, counter, and histogram metric types and give one concrete example of each from a running target

Draw the Prometheus data model on paper: metric name, labels, samples, and timestamps, then annotate how a scrape becomes a time series

Build rate() and increase() queries over counters and identify why a window shorter than two scrape intervals gives misleading results

Apply aggregation operators (sum, avg, topk, count by) to collapse a high-cardinality series set into a per-job summary

Configure two recording rules that precompute expensive aggregations and confirm they appear as new series

Diagnose and fix one query that returns no data by inspecting label mismatches with label_values and series matching

Write histogram_quantile() queries against a bucket metric and validate the result against a hand-computed estimate

Use offset, @ modifier, and subqueries in at least three queries to compare current versus past behavior

Compare pull versus push collection models and list two tradeoffs of each in the context of your own setup

Map the four golden signals onto specific metrics available in your local instance

Instrument a small application with a client library exposing a counter, a gauge, and a histogram with meaningful labels

Write a custom exporter or textfile collector that surfaces at least two business-level metrics

Explain cardinality risk and rewrite one metric to remove a high-cardinality label without losing analytical value

Describe how service discovery and relabeling change the target set, and apply one relabel rule that drops a label

Author three alerting rules covering a symptom-based condition, a saturation condition, and an absent-metric condition

Tune a for: duration and explain how it interacts with the evaluation interval to reduce flapping

Write a runbook entry for each alert naming the likely cause and first diagnostic query

Route alerts through Alertmanager with at least one inhibition rule and one grouping configuration

Build a dashboard with panels for rate, latency percentile, error ratio, and saturation, each backed by a saved query

Add template variables so the dashboard can be filtered by job and instance without editing panels

Apply unit formatting and thresholds so a panel communicates health at a glance

Review one dashboard for misleading axes or missing context and fix the issues found

Solve ten timed PromQL scenarios mixing aggregation, histogram quantiles, and offset comparisons

Debug three broken queries by reasoning about vector matching and label sets before testing

Convert one raw metric query into a recording rule feeding an alert to reduce evaluation cost

Explain when an alert should be based on a symptom versus a cause and justify your choices

Re-derive the core concepts of each domain from memory in a one-page cheat sheet

Re-run every deliverable from weeks 1 through 7 and fix anything that no longer works

Complete a full-length timed practice set and categorize every miss by domain

Close the two weakest domains by rewriting the missed concepts in your own words