Details
-
Bug
-
Resolution: Fixed
-
Major
-
Lustre 2.3.0, Lustre 2.1.3, Lustre 2.1.6
-
b2_1 g636ddbf
-
3
-
4236
Description
I have a smallish filesystem to which I only allocated a 5GB MDT since the overall dataset was always intended to be very small. This filesystem is simply being used to add and remove files in a loop with something along the lines of:
while true; do cp -a /lib /mnt/lustre/foo rm -rf /mnt/lustre/foo done
It seems in doing this I have filled up my MDT with an "oi.16" file that is now 94% of the space of the MDT:
# stat /mnt/lustre/mdt/oi.16 File: `/mnt/lustre/mdt/oi.16' Size: 4733702144 Blocks: 9254568 IO Block: 4096 regular file Device: fd05h/64773d Inode: 13 Links: 1 Access: (0644/-rw-r--r--) Uid: ( 0/ root) Gid: ( 0/ root) Access: 2012-05-27 11:55:00.175323551 +0000 Modify: 2012-05-27 11:55:00.175323551 +0000 Change: 2012-05-27 11:55:00.175323551 +0000 # df -k /mnt/lustre/mdt/ Filesystem 1K-blocks Used Available Use% Mounted on /dev/mapper/LustreVG-mdt0 5240128 5240128 0 100% /mnt/lustre/mdt # ls -ls /mnt/lustre/mdt/oi.16 4627284 -rw-r--r-- 1 root root 4733702144 May 27 11:55 /mnt/lustre/mdt/oi.16
It seems the OI is leaking and not being reaped when files are removed.