Open auroraanon38 opened 8 months ago
Something that has helped me was schedtool -D ionice -c 3 duperemove ...
which puts duperemove under CPU and IO idle priority policies.
If you're deduping HDDs then limiting it to 1 IO thread might also help because parallelism tends to make access more random and HDDs prefer more sequential IO patterns.
Another option is to use cgroups to set io.max
or cpu.max
policies
Hello @auroraanon38 Your issue is quite strange to me
Could you reproduce the issue with the latest code, without using the sync
mount option ?
If you do reproduce, could you give me your kernel version and the btrfs-tools version used to create the filesystem ?
Thank you
This is most likely not the right place to mention this, but I didn't manage to find the mailing list.
I've found that mounting the filesystem to deduplicate with the
sync
(synchronous IO) option can help prevent hanging and kernel timeouts which I've seen mentioned in none of the other issues about hanging.Has the thrashing aspect of async IO been looked at as a factor so far?