Occassional problems with unfreeze

[Dave Chinner]

On Thu, Sep 25, 2014 at 11:37:47AM +0530, Hrishikesh Barua wrote:
> This happened again yesterday and I took the "echo w >
> /proc/sysrq-trigger".
> Output - http://pastebin.com/rpKNSYY7

How long have you been ignoring these messages for?

[30278027.798067] XFS (xvdj): xlog_space_left: head behind tail
[30278027.798069]   tail_cycle = 1085, tail_bytes = 1041920
[30278027.798070]   GH   cycle = 1085, GH   bytes = 1039296

XFS is emitting that warning several times a second. Isn't that a
pretty good sign that there's something not quite right and is worth
reporting?

Indeed, you are running an old kern (3.2.0-54) which means it won't
have this fix:

3948659 xfs: Account log unmount transaction correctly

And that's likely to be the source of all your freeze/unfreeze
problems because it will eventually hang the log when it's leaked
all it's space in the ether....

Cheers,

Dave.
-- 
Dave Chinner
david at fromorbit.com