All Products
Search
Document Center

E-MapReduce:Local disk directories used by HDFS components

Last Updated:Mar 26, 2026

Hadoop Distributed File System (HDFS) components store critical data on the local disks of each node. This page lists the directories for both high availability (HA) and non-HA EMR clusters so you can protect or exclude them during disk maintenance.

Warning

If the directory that is used by a component is deleted for no reason, service exceptions and data loss may occur.

Non-HA cluster

In a non-HA cluster, a single active NameNode manages the namespace. SecondaryNameNode handles periodic checkpointing — it merges the EditLog into a new FsImage to prevent the EditLog from growing indefinitely.

ComponentDirectoryStores
NameNode/mnt/disk1/hdfs/nameFsImage files of the NameNode in the non-HA cluster.
NameNode/mnt/disk1/hdfs/edit/EditLog files of the NameNode in the non-HA cluster.
SecondaryNameNode/mnt/disk1/hdfs/secondary/Relevant data of the SecondaryNameNode in the non-HA cluster. SecondaryNameNode merges EditLog files and generates a new FsImage.
DataNode/mnt/disk1/hdfs through /mnt/diskN/hdfsData blocks, one directory per data disk. N depends on the number of data disks on a DataNode. For example, a two-disk DataNode uses /mnt/disk1/hdfs and /mnt/disk2/hdfs.

HA cluster

ComponentDirectoryStores
ZKFailoverController (ZKFC)NoneZKFC does not use a local disk directory.
NameNode/mnt/disk1/hdfs/nameFsImage files of a NameNode in the HA cluster.
NameNode/mnt/disk1/hdfs/edit/EditLog files of a NameNode in the HA cluster.
JournalNode/mnt/disk1/hdfs/journal/EditLog files of a JournalNode in the HA cluster.
DataNode/mnt/disk1/hdfs through /mnt/diskN/hdfsData blocks, one directory per data disk. N depends on the number of data disks on a DataNode. For example, a two-disk DataNode uses /mnt/disk1/hdfs and /mnt/disk2/hdfs.