VictoriaMetrics best practices

1. Architecture #

Keeping it simple.

Principles #

• Single node as long as possible; cluster when volume or multi-tenancy requires it.
• vmagent for collection, with a disk buffer and two destinations for HA.
• SSD disks, at least 20% free space (part merging).
• Retention set explicitly, daily backups verified by a test restore.

2. Data #

Controlling cardinality.

Rules #

• Cardinality explorer checked every month.
• Relabeling in vmagent to drop useless labels and metrics.
• Stream aggregation for metrics where only the total is useful.
• Dashboards in standard PromQL if they must stay portable.

3. Operations #

Monitoring and upgrading.

Rituals #

• Official dashboards and alerts installed.
• Regular upgrades (changelog read, one instance at a time).
• vmalert with -remoteRead.url to keep alert state.
• Access protected by vmauth or a reverse proxy.

Reference diagram and quick check #

Checklist.

Checklist #

1. Retention and disk sized
2. vmagent with disk buffer
3. vmalert connected to Alertmanager
4. vmbackup backups and restore test
5. Monitoring with the official dashboards
6. Authenticated access