Project

General

Profile

Feature #16757

enable MDS replacement migration

Added by Greg Farnum about 3 years ago. Updated 6 months ago.

Status:
New
Priority:
Normal
Assignee:
-
Category:
-
Target version:
-
Start date:
07/20/2016
Due date:
% Done:

0%

Source:
Development
Tags:
Backport:
Reviewed:
Affected Versions:
Component(FS):
Labels (FS):
multimds
Pull request ID:

Description

Right now, without multi-mds the only way we have to switch MDSes is to do a failover from the current active to something that is (hopefully) a standby-replay. This still requires a client reconnect process, though.

With multi-MDS clusters, we make it possible cleanly remove an MDS from the cluster by exporting all of its data to the others.

It would be great if we could enable live MDS migration more smoothly. For instance, somebody running a single MDS a and a standby b could issue a "failover" command which brings the standby MDS b into the "cluster", smoothly (and incrementally) migrates all data from mds.a to mds.b, and then shuts down mds.a for whatever purpose. This ought to significantly reduce the client impact of MDS failover.
One could try increasing the size of the mds cluster and then removing mds.a from it, but I think that has peculiar consequences in terms of the ranks remaining in the system? And if you already have a multi-MDS system, you might or might not want the evacuation of one MDS to impact the other live daemons in the cluster.

History

#1 Updated by Patrick Donnelly 6 months ago

  • Category deleted (90)
  • Labels (FS) multimds added

Also available in: Atom PDF