I was testing Clint Morgan's tableindexed region server patch on top of 0.19 on
our 25 node cluster but had to back out after, over time, a number of regions
went offline. Occasionally when the balancer would do a region reassignment a
region would get lost. As the region load therefore dropped over time, the
frequency of lost regions increased as more nodes became imbalanced causing
more reassignment. I had not gotten to the point of actually using any indexed
or transactional table functionality so I do not think that factors in.
Anybody seen anything like this on 0.19?
- Andy
