I have an HBase cluster with 4 RegionServers that was serving ~3600 requests per second. After adding two RegionServers / DataNodes at once, two of the pre-existing RegionServers went up to ~55000 and ~96000 requests per second, and totaling about ~153000 requests per second across all 6 RegionServers. The newly added RegionServers had the lowest requests per second.
After leaving it overnight, the total number of requests has gone back down to ~3600 requests per second across all 6 RegionServers. As well, HDFS has distributed the blocks evenly so that the new DataNodes now host a similar amount of blocks as the pre-existing DataNodes. Not sure if relevant, but the "hbase:meta" table regions was not hosted on one of the RegionServers that had the spike in requests per second. Is it normal behaviour for existing RegionServers to have to work harder when new RegionServers come online? If so, what is the reason for the (massive) increase? Thanks, Mark
