Jesu, Yes, normally we expect that one Ignite node will run on all machines where file system is accessed. The main reason for this is performance. For example, what if Hadoop sent a job to a node where HDFS is running, but there is no Ignite node? In this case if you access HDFS through IGFS, you will have to connect to some remote Ignite node and try fetching data form there, what is less than optimal due to additoinla network trips. This is why advised way to work with IGFS is to start Ignite node on each machine running HDFS.
However, this is not strict requirement. You can work with remote Ignite nodes through TCP endpoint of course. But if IGFS use some additional Hadoop configuration files like *core-site.xml* or *hdfs-site.xml*, then you must ensure that they are placed in the same directories on all machnies where you access IGFS. Please let me know if you have any further questions. Vladimir. On Mon, Sep 14, 2015 at 11:56 AM, Jesu <[email protected]> wrote: > Vladimir, > Yes, the processes are running in different machines. Does this mean ignite > needs to run on all the machines in which igfs is accessed? > > > > -- > View this message in context: > http://apache-ignite-users.70518.x6.nabble.com/How-to-configure-Hadoop-accelerator-for-Hadoop-High-Availability-tp1273p1378.html > Sent from the Apache Ignite Users mailing list archive at Nabble.com. >
