Re: [Gluster-devel] attaching tier and starting rebalance daemon in single command, (needed to see all directories)

2015-04-22 Thread Atin Mukherjee
+ gluster-devel

On 04/23/2015 09:20 AM, Atin Mukherjee wrote:
 
 
 On 04/23/2015 01:46 AM, Dan Lambright wrote:
 KP, Atin,

 Currently in tiered volumes, we attach a tier and then start the rebalance 
 daemon in separate steps.

 The problem here is if someone does readdir on the mount point before 
 starting the daemon, they won't see all of the directories. We have not yet 
 done fix-layout to make all directories live on all bricks.

 What I need to do is combine those two commands into a single one.

 What I am thinking is, at the cli level, we would first send the attach 
 command, and if it returns successfully, send the tier start command.

 The current volume status (not rebalance status) actually issues
 multiple RPCs from CLI, first it gets the volume list from glusterd and
 in issues subsequent RPCs for each volume. I guess you can have a look
 at that implementation if you find this challenging. As far as design
 goes, I don't see that should be a problem, but in that case are we
 going to disable the CLI interface for tier start? Does an admin need to
 trigger tier start explicitly in any particular case?
 
 ~Atin
 Do you think it is ok to do this at the cli level (I hope my description 
 makes sense)

 If not, do you see a better way? We can discuss if necessary..

 Thanks,
 Dan  

 

-- 
~Atin
___
Gluster-devel mailing list
Gluster-devel@gluster.org
http://www.gluster.org/mailman/listinfo/gluster-devel


Re: [Gluster-devel] attaching tier and starting rebalance daemon in single command, (needed to see all directories)

2015-04-22 Thread Krishnan Parthasarathi
  What I need to do is combine those two commands into a single one.
 
  What I am thinking is, at the cli level, we would first send the attach
  command, and if it returns successfully, send the tier start command.

Composing attach-tier and tier-start commands in cli is an acceptable 
solution.
Remember that other commands on this volume can be executed between the composed
attach-tier and tier-start command, possibly failing the tier-start command 
with
Another transaction in progress error. This could happen as the volume 
transaction
lock is released at the end of a command.

Alternately, you could start the promotion/demotion daemon on nodes that have
'relevant' bricks in the attach-tier command. You could choose depending on 
which
approach is maintainable for you.

HTH,
KP
___
Gluster-devel mailing list
Gluster-devel@gluster.org
http://www.gluster.org/mailman/listinfo/gluster-devel