Support Questions

Find answers, ask questions, and share your expertise

Inadequate control over roles in Installation Path A ?

avatar
Explorer

I have gone through a Cloudera Manager installation several times today.

It is a beautiful installer ... yet I am left very puzzled by something.

 

I have not been able to find a way to prevent my "master" NameNode+JobTracker machine from also being given DataNode & TaskTracker roles. Clearly this is undesirable, yet I have been able to figure out how to achieve it cleanly through the Installation Path A.

 

I reinstallled (essentially from scratch) several times, thinking I had missed something.

 

I ended up decommissioning & deleting my "master" from the undesirable storage/computation roles.

 

Not surprisingly, my disk config is different. Therefore, the install created special groups for it. 

Even though my master no longer has those roles, the config groups still exist as remnants ... with no member machines.

 

Q: Did I miss something? Isn't there a way to prevent your NameNode & JobTracker from also being assigned DataNode and TaskTracker roles?

 

Q: Is there a way to delete a "configuration group" that has no members?

 

Thanks,

Michael

 

1 ACCEPTED SOLUTION

avatar
Hi Michael,

During the wizard, there's a button to Inspect Role Assignments. This will let you change which hosts have which roles.

In the next release of CM5, we changed this behavior to never recommend co-located worker and master nodes, except for single node clusters. Previous versions we would recommend co-locating slaves and masters for clusters up to and including size 20.

Thanks,
Darren

View solution in original post

3 REPLIES 3

avatar
Explorer

I figured out how to delete a role group.
I am still interested in understanding why the install doesn't *require* one to treat their NameNode, JobTracker, etc. as "special"

avatar
Hi Michael,

During the wizard, there's a button to Inspect Role Assignments. This will let you change which hosts have which roles.

In the next release of CM5, we changed this behavior to never recommend co-located worker and master nodes, except for single node clusters. Previous versions we would recommend co-locating slaves and masters for clusters up to and including size 20.

Thanks,
Darren

avatar
Explorer

Thank you *very* much for the prompt response.

 

I'll go through it again and check it out.

 

> we changed this behavior to never recommend co-located worker and master nodes, except for single node clusters.

 

very good