incorrect CPU utilization on node map

[lin-crl]

Describe the problem
The CPU utilization on node map is very different from the utilization on Hardware dashboard.
Please describe the issue you observed, and any steps we can take to reproduce it:

To Reproduce

1. Set up CockroachDB cluster
2. Run a workload such as movr or kv
3. Look at Nodemap CPU utilization and compare it w/ Hardware dashboard

Expected behavior
Expect consistent utilization on node map

Additional data / screenshots

Environment:

• CockroachDB version 22.1.6
• Server OS: Ubuntu
• Client app - DBConsole

Additional context
What was the impact? - confusion to users

Jira issue: CRDB-19464

Epic CRDB-21265

[koorosh]

Node Map and CPU Percent metrics on Hardware dashboard represent CPU usage in different forms and rely on different metrics underneath.

• CPU Percent metric on Hardware dashboard - Current user+system cpu percentage, normalized 0-1 by number of cores
• CPU metric on Node Map - Sum of system and user current cpu percentage. Current metric is not present on Metrics dashboards.

I'd suggest to extend Runtime dashboard with a new metric to represent the same info as on Node Map.
WDYT @thtruo

[thtruo]

I'd suggest to extend Runtime dashboard with a new metric to represent the same info as on Node Map.

@koorosh Do you mean the Hardware dashboard? I don't think we should extend that dashboard with the Node Map metric. The node map metric should instead be updated to report what's used in the CPU Percent metric from the Hardware dashboard. That's a more accurate reflection. The current Node Map metric has resulted in confusion from customers before, especially when they see >100% values from the Node Map. We should use the normalized metric instead.

[koorosh]

We should use the normalized metric instead.

Thanks for clarification!