HDFS BlockPlacementPolicy, is there an alternative that considers available disk space?

ThomasLarsson — Mon, 17 Oct 2016 18:25:35 GMT

Hello,

I am wondering if there is an BlockPlacementPolicy that in addition to storing replicas safely on different racks as the default one does, also can consider how much disk space that is available on different nodes?

In case where you have a cluster that consists of two sets of machines with a big difference in the amount of available disk space, the default policy will lead to the disks of the set with a smaller amount of disk space running out of disk space long before you actually reach your total HDFS capacity.

Is there any such policy ready to be used?

Best Regards

Thomas

Re: HDFS BlockPlacementPolicy, is there an alternative that considers available disk space?

ThomasLarsson — Mon, 17 Oct 2016 18:34:54 GMT

I just found that something like this was added somewhat recently:

https://github.com/apache/hadoop/blob/f67237cbe7bc48a1b9088e990800b37529f1db2a/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/blockmanagement/AvailableSpaceBlockPlacementPolicy.java

This seems to be what I was looking for.

question HDFS BlockPlacementPolicy, is there an alternative that considers available disk space? in Archives of Support Questions (Read Only)

HDFS BlockPlacementPolicy, is there an alternative that considers available disk space?

Re: HDFS BlockPlacementPolicy, is there an alternative that considers available disk space?