Right-Size Your Prometheus Memory: Calculator & Rule of Thumb
Right-Size Your Prometheus Memory: Calculator & Rule of Thumb
Is Your Prometheus Scaled Correctly?
Prometheus OOM’d again? Or worse—you’re burning cloud budget on 4x the memory you need?
The rule of thumb I use: allocate 7-9 KiB of memory per active time series. Add 1 CPU core per 4GiB of memory (minimum 2 cores). This simple formula has helped clients right-size everything from 100K to 10M+ series deployments.
Prometheus Memory Calculator
Use this calculator to size your deployment. The optimum goal I use is 7.5KiB of memory per time series.
Common Issues
This is a rule of thumb. Depending on your usage patterns it may need some adjusting. High frequency (15s interval) metrics will need more memory. Histogram work may require both more memory and CPU. Numerous and complicated recording rule expressions or alert expressions may require more CPU. In non-production environments you might get away with less resources. So use your monitoring of Prometheus to see how well Prometheus fits in these guidelines with your usage.
Need Expert Help?
This rule works for 80% of deployments. But if you’re dealing with high-cardinality explosions, multi-tenant Prometheus, or scaling past 10M series—that’s where my expertise comes in.
Schedule a 30-minute consultation or email jjneely@cardinality.cloud.