I’m just going to guess, but you’ve got active/active OSTs, and some are 
running on the standby node.

A lustre client will contact the MGS on mount and get the list of MDTs and 
OSTs, along with their primary and secondary (etc.) locations.  The client then 
starts contacting the targets.  The thing is, that for any target, the client 
tries querying the primary location for the target, and has to wait for a 
timeout value before continuing on.  It then tries the secondary location, etc. 
 Combine that with targets not responding right away due to being busy, and you 
get the behavior you describe.

Try it on a quiescent system, where all targets are mounted on the primary 
server, and it will all happen pretty much instantaneously.

-Ben Evans

From: lustre-discuss 
<[email protected]<mailto:[email protected]>>
 on behalf of "Andrus, Brian Contractor" 
<[email protected]<mailto:[email protected]>>
Date: Tuesday, September 22, 2015 at 12:39 PM
To: "[email protected]<mailto:[email protected]>" 
<[email protected]<mailto:[email protected]>>
Subject: [lustre-discuss] client showing "inactive device" at mount time

All,

I have been seeing this pretty regularly.
I reboot a system and mount a lustre filesystem.
It mounts, but when I do ‘lfs df’, I see numerous OSTs that are listed as 
‘inactive device’
If I wait, they all eventually mount. Odd thing is that I always get some that 
mount fine and others that wait that are both served by the same OSS.

Any insight as to why OSTs can take several minutes to mount on a client? There 
doesn’t seem to be any network issues or congestion.

Brian Andrus
ITACS/Research Computing
Naval Postgraduate School
Monterey, California
voice: 831-656-6238

_______________________________________________
lustre-discuss mailing list
[email protected]
http://lists.lustre.org/listinfo.cgi/lustre-discuss-lustre.org

Reply via email to