VictoriaMetrics Quiz
Review baseline: VictoriaMetrics 1.151.0 · stack chart 0.92.1
- Which statement is correct when migrating Prometheus queries to VictoriaMetrics?
- A) Every workload has exactly 7× better compression
- B) All installations have a fixed 100M-series limit
- C) Compare actual queries because MetricsQL intentionally differs from PromQL
- D) Prometheus cannot retain more than 15 days
Show Answer
Answer: C
Familiar syntax does not establish identical rate/increase, NaN, scalar or API behavior. Benchmarks need actual versions, data and hardware; retention defaults are not maxima.
- Which is not one of the cluster storage/query data-plane components?
- A) vminsert
- B) vmstorage
- C) vmselect
- D) VictoriaMetrics Operator
Show Answer
Answer: D
vminsert routes writes, vmstorage stores data, and vmselect queries it. The Operator reconciles Kubernetes resources; it is a separate control component.
- What does vmagent do?
- A) Replaces all durable storage
- B) Renders dashboards
- C) Collects metrics and forwards them to remote-write destinations
- D) Replaces Alertmanager notification routing
Show Answer
Answer: C
Its queue buffers outages but is finite: the configured disk cap drops oldest queued data, and an emptyDir is lost with Pod replacement. Persistence and monitoring remain separate design choices.
- What does keep_last_value(q) do, and what needs care?
- A) Returns the maximum stored value
- B) Fills evaluated gaps with an earlier value, which can conceal missing telemetry
- C) Restores all lost scrape samples
- D) Guarantees healthy availability alerts
Show Answer
Answer: B
Use gap filling only when its meaning is intended. Keeping an old healthy value can hide missing collection; monitor missing data separately.
- What does dedup.minScrapeInterval control?
- A) The target scraper scheduling interval
- B) Keeping one sample per discrete interval for the same series
- C) Lossless generic compression
- D) Alert evaluation frequency
Show Answer
Answer: B
The interval can remove legitimate higher-resolution samples, not just byte-identical copies. Align vmstorage/vmselect settings and the chosen storage/scraper replication design.
- How should single-node and cluster modes be selected?
- A) Always choose cluster mode
- B) Measure load, active series/churn, queries, retention and recovery requirements
- C) Use 100M samples/day as a universal cutoff
- D) Single-node mode has no query API
Show Answer
Answer: B
Assess one-server capacity and the cost of independently scaling insert/storage/select. Replicated storage or two processes alone does not provide tested service HA.
- What does vminsert replicationFactor=2 request?
- A) Exactly two total storage members
- B) Two copies on distinct storage members, subject to actual availability and history
- C) Queries on exactly two members
- D) Twice the compression
Show Answer
Answer: B
Maintaining two copies during one storage failure needs at least three members plus the relevant capacity and failure-domain conditions. Configure vmselect/dedup consistently; the flag does not backfill history, make degraded writes lossless, or replace backups.
- What is the scope of the MetricsQL default operator?
- A) It always creates correct service labels
- B) It fills missing points from the right-hand expression; missing data is not automatically zero traffic
- C) It guarantees that every ratio is meaningful
- D) It changes the metric ingestion interval
Show Answer
Answer: B
Do not append default 0 as a universal error-rate fix. Aggregate status labels consistently, fill a missing numerator only for known denominator series, and exclude zero traffic. The native fixture yields 0, 1 and 0.1 for healthy, all-failed and 10%-failed traffic; absent/zero-traffic services remain absent.
- What is vmalert responsible for?
- A) Collecting every application metric
- B) Storing all long-term samples
- C) Evaluating alert/recording rules and sending alerts to a configured notifier
- D) Rendering Grafana dashboards
Show Answer
Answer: C
Rule inputs/exporters, datasource URLs, state persistence and notifier/routing configuration must exist. A high restart count is not by itself proof of CrashLoopBackOff. Recording derived series does not compact or delete raw data.
- What is the correct vmbackup workflow?
- A) Back up only vminsert configuration
- B) Read a snapshot from the matching storage directory and protect a distinct backup destination
- C) Mount any RWO PVC on any node without checking
- D) Treat storage replication as a complete backup
Show Answer
Answer: B
Back up every vmstorage member to a separate prefix, or the intended single-node store. S3/GCS/Azure/local destinations are supported. Verify co-location/PVC identity, workload credentials, retention and actual restore; a local snapshot is not an independent copy.