There can be rare cases where this will return the wrong measure if cpu_all is available in a given timestamp where lpar is not (and lpar is available)
This can be replaced by a more robust condition:
| eventstats count(eval(vp_usage>0)) as count_lpar | eval usage=if(count_lpar>0, vp_usage, pct_usage)
In some dashboards, queries will return both lpar and cpu_all results and filter out the unrequired measurement if lpar is available:
| eval usage=if(isnum(vp_usage), vp_usage, pct_usage)
There can be rare cases where this will return the wrong measure if cpu_all is available in a given timestamp where lpar is not (and lpar is available)
This can be replaced by a more robust condition:
| eventstats count(eval(vp_usage>0)) as count_lpar | eval usage=if(count_lpar>0, vp_usage, pct_usage)