I disagree with the notion of "No single point of failure" with HBase. While it's true that they got rid of the old hard SPOFs, if a data ("region") node fails, there's a decent chance you lost some data (there's a short period of time where data hasn't been replicated to HDFS and is ONLY only the master for that region) at least until you can bring that node back up (hope you've got reliable RAID/backups or it wasn't destroyed). HBase is a CP system.
There's also a region master re-election/recovery period that depends on the size of the database, network bandwidth, load, etc. It can be anywhere from 30 seconds to tens of minutes. An outage of a region node makes it's key range inaccessible. While that might not be a problem for some, especially in read-only situations, I can think of many applications where that would effectively translate into a total outage.
Comments
I disagree with the notion of "No single point of failure" with HBase. While it's true that they got rid of the old hard SPOFs, if a data ("region") node fails, there's a decent chance you lost some data (there's a short period of time where data hasn't been replicated to HDFS and is ONLY only the master for that region) at least until you can bring that node back up (hope you've got reliable RAID/backups or it wasn't destroyed). HBase is a CP system.
There's also a region master re-election/recovery period that depends on the size of the database, network bandwidth, load, etc. It can be anywhere from 30 seconds to tens of minutes. An outage of a region node makes it's key range inaccessible. While that might not be a problem for some, especially in read-only situations, I can think of many applications where that would effectively translate into a total outage.