Details
-
Improvement
-
Status: Open
-
Minor
-
Resolution: Unresolved
-
None
-
None
-
None
Description
Right now we expose a metric Hadoop:service=HBase,name=RegionServer,sub=Server,attribute=averageRegionSize, which is calculated as (the sum of all Store#getStoreFilesSize() for all stores in all regions, plus the current memStore size) divided by the number of regions. We also expose under Hadoop:service=HBase,name=RegionServer,sub=Regions,attribute=$tableName_$region_metric_storeFileSize. I posit that the average size metric is too coarse-grain, yet the metric per region is too granular to be casually consumed. I propose we change the averageRegionSize to be a histogram with our standard percentiles. I suspect this change would be obliged to omit the memStore size from the calculation, but I think the memStore is adequately monitored elsewhere.