That should have almost no difference to Drill other than performance and whether the data being queried is up-to-date or delayed.
Other features like atomic snapshots and mirrors have operational significance, but Drill should not much care or even know that it is querying data in a snapshot or mirror. One significant area of difference is that MapR exposes the cluster storage not only via the HDFS API, but also via NFS (distributed NFS in M5 or M7). That will allow prototype implementations to use normal Java I/O systems to read and write data rather than having to use the HDFS API's. THis will mean that things like the scanner that reads JSON data from a normal file in the reference interpreter will work against a cluster without change. That may speed development, of course. MapR specific questions can be directed to http://answers.mapr.com or to me directly. This question is a bit different in that it asks if Drill will work on MapR. On Wed, Jul 10, 2013 at 6:22 AM, Alexandre BECHE <[email protected]>wrote: > Dear drill dev, > > As discussed yesterday during the Hangout, I am currently setting up a > cluster using the M3 distribution. > I went through the MapR documentation for the installation and I found that > there is no Namenodes but an HDFS compliant API. What are the impact on > DRILL for that? Is DRILL compatible for both system (native HDFS and MapR > custom HDFS)? > > Cheers, > Alex >
