The return of the NameNode controversy?
I thought that old criticism about Hadoop NameNode reliability, etc., had died down with changes made to core Apache Hadoop software, but MapR is still insisting that Hadoop can't be a "mission critical" with HDFS as-is. "People assume that 'snapshots' are point-in-time consistent snapshots across volumes and clusters, but with open source HDFS, whenever the file is closed, that's the data that's contained in that snapshot," MapR's Jack Norris told me last week. "So that means there are all storts of time stamps associated with a snapshot [if you use HDFS], and you can't recover an application that way."
With these vulnerabilities, talk of Hadoop as an "enterprise data hub," as espoused by Cloudera, is premature, says Norris, because you can't depend on the data that gives you higher-level analytical capabilities at the top of the stack. Cloudera and Hortonworks would obviously disagree with these assertions, but they're too busy throwing cold water on each other's strategies. Frankly, the more these three companies rail at each other, the less faith the whole world has in Hadoop -- and they're not distinquishing among anybody's distribution.