[Gluster-devel] [bug #19257] read-ahead with writebehind locks up under heavy RW load

[A. D Auria]

URL:
  <http://savannah.nongnu.org/bugs/?19257>

                 Summary: read-ahead with writebehind locks up under heavy RW
load
                 Project: Gluster
            Submitted by: adauria
            Submitted on: Friday 03/09/2007 at 11:43
                Category: GlusterFS
                Severity: 3 - Normal
                Priority: 5 - Normal
              Item Group: Improper behaviour
                  Status: None
                 Privacy: Public
             Assigned to: None
             Open/Closed: Open
         Discussion Lock: Any
        Operating System: GNU/Linux

    _______________________________________________________

Details:

[configuration...]
Using latest TLA version (09-march-2007)
...
volume writebehind
  type performance/write-behind
  option aggregate-size 131072 
  subvolumes bricks
end-volume

volume readahead
  type performance/read-ahead
  option page-size 131072    
  option page-count 16   
  subvolumes writebehind
end-volume

[Observation...]
In the above configuration, our writing process was able to run for an
extended period of time (several hours).  As soon as we started our reading
process however, all reading and writing processes stopped responding.  No
crashes, no log reports, nothing.  No further traffic was observed.  

For us to recover the client, we had 'umount -f' the gluster mount point and
then 'kill -9 glusterfs'.

After removing the 'volume readahead' section and restarting the client,
everything worked properly (without readahead).  Reading and writing with
high loads works properly with writebehind only.








    _______________________________________________________

Reply to this item at:

  <http://savannah.nongnu.org/bugs/?19257>

_______________________________________________
  Message sent via/by Savannah
  http://savannah.nongnu.org/