Support Questions

Find answers, ask questions, and share your expertise
Announcements
Celebrating as our community reaches 100,000 members! Thank you!

HA for Name Node

avatar

I am planning to enable HA for my test cluster using Cloudera Manager.

Cloudera Manager is helping to enable HA for Name Node in Easy way


But The Question here is,  
What method is using for Name Node edits, It means, " is it using Quorum Journal or NFS share edit ?"

 

Best Regards,

Bommuraj

1 ACCEPTED SOLUTION

avatar
Mentor
Use the Quorum Journal Manager mode. I believe the Wizard also
recommends use of this mode clearly.

The NFS mode was an early way of doing it, but requires the additional
HA-NFS hardware and the maintenance associated with it and has not
been popular.

View solution in original post

3 REPLIES 3

avatar
Mentor
Use the Quorum Journal Manager mode. I believe the Wizard also
recommends use of this mode clearly.

The NFS mode was an early way of doing it, but requires the additional
HA-NFS hardware and the maintenance associated with it and has not
been popular.

avatar
Guru

I'd also highly recommend the usage of the Quorum Journal.

In addition to the requirement of having additional NFS infrastructure, NFS itself isn't that reliable and we had several issues with its availability in the past.

avatar
New Contributor

Regarding where to host the 3 journal nodes using QJM:

 

Apache docs (and similarly in Cloudera HA guide) state: The JournalNode daemon is relatively lightweight, so these daemons may reasonably be collocated on machines with other Hadoop daemons, for example NameNodes, the JobTracker, or the YARN ResourceManager

 

My 2 NameNode machines are not good candidates for hosting 2 of the 3 daemons due the recommendation that these have a dedicated disk drive. So with those out of the picture, the question circulating around my team is whether we can put these 3 daemons on nodes running data node daemons OR if we should dedicate 3 small virtual machines to running them. To me, 3 new VMs is overkill. Note that we have YARN RM on a datanode as well due to the small size of the cluster. Thanks for your input and if there are better suggestions, please let me know.