On Wed, Oct 8, 2014 at 3:30 AM, mufy <[email protected]> wrote: > Posting the query again to get attention :-) > > *[CONCERN 2]* > > What's the advantage of specifying the ZK ensemble in the sqlline > connection string? With the way PStore is handled now (directly on the > HDFS), is it relevant anymore? >
In the current scheme of things, HDFS does not replace rather supplement [1] ZooKeeper as the PStore. Smaller data, like storage plugin configuration and system settings, which require quick access is still stored in ZooKeeper whereas the large data is offloaded to the filesystem (distributed or local). However, PStore is not the only or the primary thing ZooKeeper is used within Drill. It is used as the cluster coordinator in the distributed mode and can not be done away with. This is how any Drill client, including sqlline, finds the drillbits to connect to. [1] https://issues.apache.org/jira/browse/DRILL-1414
