Background :

HBase reads seem to fall under the 'strong consistency' model, as : All reads are served from the master where the data has already been committed. As a result, the clients seem to always see the latest view of the committed data

Question :

How does the replication model (typically asynchronous at HDFS level) within the same HBase cluster ensures that this 'strong consistency' is always honoured. From my undertstanding, there could be cases when a new write gets lost due to this (asynchronous replication + hdfs data-node failover), and clients stop seeing the committed data which was acked to them.

If the above assumption holds true, how can we still guarantee HBase to be always 'strongly consistent' or in general, Any database which is backed by asynchronous replication, how can we consider it as strongly consistent. ??