#community-help

Monitoring & Upgrading Cloud-Hosted Cluster: Metrics Method

TLDR Ross asked about the best metric aggregation method for monitoring a cloud-hosted cluster. Jason recommended looking at p50 and explained certain aspects of burst capacity.

Powered by Struct AI

1

Jan 27, 2023 (11 months ago)
Ross
Photo of md5-faf0fdba0b6739a6706f05c15b6738c6
Ross
03:26 PM
When monitoring a cluster for health/need to upgrade (Cloud-hosted) -- is there a metric aggregation method you'd recommend (i.e. max vs avg vs p95)? My avg are looking fine, while my max is a bit spikey-looking. Will those spikes be handled by the burst capacity?
Jason
Photo of md5-8813087cccc512313602b6d9f9ece19f
Jason
04:26 PM
I’d recommend looking at p50. If p50 crosses 70% then it’s a good time to upgrade.

Burst capacity is counted based on how long CPU is spiked to 100%. It doesn’t increase the number of cores when bursting.
Jan 28, 2023 (10 months ago)
Ross
Photo of md5-faf0fdba0b6739a6706f05c15b6738c6
Ross
12:12 AM
Thanks!

1