Understanding Safemode in Hadoop

Goals

Safemode is a representation that Hadoop is in read-only mode. It is ideally meant for transitioning the cluster from production to maintenance mode. In this mode, HDFS doesn’t allow any write-related and processing operations.

However, read is possible. When we start Hadoop services, HDFS service goes in safemode to perform the following activities:

  1. Flush edits to fsimage and load fsimage in-memory
  2. Get the block report from each DataNode and check the compliance of metadata
  3. Report and update replication parameters (under-replicated and over-replicated blocks)

However, a user can bring the Hadoop cluster in safemode for the following reasons:

  1. Check-pointing metadata
  2. Removing orphaned blocks
  3. Removing corrupted blocks and its metadata.

HDFS Files: none

Safe Mode

1. How to control safemode activity

To check whether the cluster is in safemode or not:

 hdfs dfsadmin -safemode get

To switch the cluster from production to maintenance mode:

hdfs dfsadmin -safemode enter

To switch the cluster from maintenance to production mode:

hdfs dfsadmin -safemode leave 

To check whether the HDFS is out of safe mode or not: 

```console
hdfs dfsadmin -safemode wait

How to do Checkpointing Manually?

2. Ensure the Hadoop cluster is in safemode

hdfs dfsadmin -safemode enter

3. Commit the edit logs to fsimage

hdfs dfsadmin -saveNamespace

4. Bring the cluster back to production mode

 hdfs dfsadmin -safemode leave

Results

Congrats!

You see how the nodes can be checkpointed manually.