November 7, 201411 yr unRAID OS Version: v6b10a Description: The idea behing moving docker containers into a fixed size btrfs image file mounted as a loopback was to ensure that non docker processes dont fill up the disk and kill docker i.e. only docker can write to the loopback. However this has not solve the problem as if you fill up the disk the loopback image file sits on the loopback image will start to corrupt with csum errors. How to reproduce: Create a docker image file on the cache disk. Start a bunch of relatively noisy dockers but ensure there is plenty of space left in the image. Fill the cache disk up 100% and leave the dockers running. Eventually the image file will start to corrupt. Expected results: Errors like: BTRFS info (device loop8): csum failed ino 42849 off 4096 csum 3469043092 expected csum 0 Actual results: As expected Other information: Fundamentally this all stems for me at least due to the other bug that allows the cache disk to fill up regardles of floor levels however this bug stands on its own and calls into question the loopback as a solution.
Archived
This topic is now archived and is closed to further replies.