On 19 December 2017 at 00:11, Stefan Solbrig <stefan.solbrig@xxxxx> wrote:
Hi all,
with the option "cluster.min-free-disk" set, glusterfs avoids placing files bricks that are "too full".
I'd like to understand when the free space on the bricks is calculated. It seems to me that this does not happen for every write call (naturally) but at some interval or that some other event triggers this.
This is checked as part of the file create and mkdir calls.
i.e, if I write two files quickly (that together would fill a brick) I'd get a error message:
dd if=/dev/zero of=a bs=1k count=15000 && sleep 1 && dd if=/dev/zero of=aa bs=1k count=15000
#yiels: dd: error writing ‘aa’: No space left on device
#(brick1 is full, but glusterd still tries to place file "aa" on the same brick
dd if=/dev/zero of=a bs=1k count=15000 && sleep 60 && dd if=/dev/zero of=aa bs=1k count=15000
#this works....
#link file on brick1 to "aa" and the real "aa"-file on brick2
Are these being done in parallel? If yes, you can hit this problem as the value is calculated before the create at which point there is sufficient disk space.
Regards,
Nithya
when is statfs (or a simlar call) evaluated?
Thank you!
best wishes,
Stefan
_______________________________________________
Gluster-users mailing list
Gluster-users@xxxxxxxxxxx
http://lists.gluster.org/mailman/listinfo/gluster-users
_______________________________________________ Gluster-users mailing list Gluster-users@xxxxxxxxxxx http://lists.gluster.org/mailman/listinfo/gluster-users
