Support Questions

Find answers, ask questions, and share your expertise

What Happens if a Failed NameNode is Back On Again after the Standby Namenode became Active?

avatar
New Contributor

Hello Community,

Suppose the Following Scenario:

1) we have a HA hadoop Cluster

2) Active Namenode Fails for some reason (EX: POWEROFF)

3) Standby Namenode Becomes Active

4) The Failed Namenode becomes ON back again

the question is:

what happens then?

does the failed Namenode becomes a standby node, or what?

enlighten me please

thank you

1 ACCEPTED SOLUTION

avatar

Hi @Jay Coder,

Please refer to automatic failover section in this blog : https://hadoop.apache.org/docs/r2.7.1/hadoop-project-dist/hadoop-hdfs/HDFSHighAvailabilityWithQJM.ht...

also you can refer this jira which explains how HDFS handles the failover and choose the Namenode to be active. : https://issues.apache.org/jira/browse/HDFS-2185

Please accept this answer if it clarifies your doubt.

View solution in original post

5 REPLIES 5

avatar

Hi @Jay Coder,

Please refer to automatic failover section in this blog : https://hadoop.apache.org/docs/r2.7.1/hadoop-project-dist/hadoop-hdfs/HDFSHighAvailabilityWithQJM.ht...

also you can refer this jira which explains how HDFS handles the failover and choose the Namenode to be active. : https://issues.apache.org/jira/browse/HDFS-2185

Please accept this answer if it clarifies your doubt.

avatar

@Jay Coder,

Can you see if this answer helps you .Please accept this answer if it did.

avatar
New Contributor

@Akhil S Naik

Thank you for your answer

I found the answer to my question after taking a look at the Failover_controller state machine diagram found in the Jira issue

it was not mentioned anywhere else

After all, The answer is TRUE

avatar
Contributor

@Jay Coder

the question is:

what happens then?

does the failed Namenode becomes a standby node -- TRUE

https://stackoverflow.com/questions/33311585/how-does-hadoop-namenode-failover-process-works

avatar
New Contributor

Thank you for the answer,

the answer (TRUE) was expected actually, but I needed an official statement for it

I found it in the Jira issue mentioned above in the state machine diagram

Thank you anyway