I'm using OpenTSDB against MapR-DB (OpenTSDB probably isn't overly relevant to the question though).
When analyzing the main table's regions, I can see that nearly half of them are over the region size of 4GB, and quite a few are ending up to 5.6GB. The largest one so far is 5.8GB.
Note that this is all based on physical size. I have LZ4 compression enabled and auto-split enabled. The logical table sizes are up to 11GB.
- Is this expected? If so, what's the upper bound on what the table size can actually be? When exactly is the auto-splitting enabled to bring the region size back down?
- I know MapR-DB has a lot of architectural improvements over HBase. If I set a smaller region size (e.g. 1GB), will it tend to help or hurt my lookup performance?
- Can I change the region size of a table that already has data (and is it okay to do)?