[ ssic-linux-Bugs-1925545 ] Processes stuck in I/O after transparent CFS failover

Bugs item #1925545, was opened at 2008-03-25 17:35
Message generated for change (Settings changed) made by rogertsang
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=405834&aid=1925545&group_id=32541

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: Filesystem
Group: v1.9.1
Status: Open
Resolution: None
Priority: 5
>Private: Yes
Submitted By: Roger Tsang (rogertsang)
Assigned to: Roger Tsang (rogertsang)
Summary: Processes stuck in I/O after transparent CFS failover

Initial Comment:
Processes or threads doing I/O during transparent CFS failover get stuck waiting in I/O.  They cannot be interrupted.  If they are part of a thread group zombies can also appear.

How to reproduce:  Start reading a large file on a CFS hard mount on the surviving OpenSSI node and force CFS to transparently failover.  The application on the surviving OpenSSI node is expected to continue uninterrupted and finish reading the entire file, but instead gets stuck waiting in I/O.

----------------------------------------------------------------------

Comment By: Roger Tsang (rogertsang)
Date: 2008-03-25 18:04

Message:
Logged In: YES 
user_id=1246761
Originator: YES

NB: This bug might not manifest in the earlier 1.9.x releases because the
CFS rebuild thread is RT before (around) 1.9.3.

----------------------------------------------------------------------

Comment By: Roger Tsang (rogertsang)
Date: 2008-03-25 17:53

Message:
Logged In: YES 
user_id=1246761
Originator: YES

This is a CFS super block rebuild flush race with CFS async code where
down requests being flushed are asynchronously pushed back into the list of
down requests.  These requests have the page lock.

Fixed in 2.0.0pre3.

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=405834&aid=1925545&group_id=32541

-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference 
Don't miss this year's exciting event. There's still time to save $100. 
Use priority code J8TL2D2. 
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone