Hi there, I'm a maintainer of a linux mirror site, and we use btrfs snapshots provide better service. The workflow of our mirror job is : 1. create a working snapshot of current mirror 2. rsync from upstream to the working snapshot 3. after rsync done, remove the original subvolume and mv working snapshot to production subvolume Everything was fine until several days ago, every `btrfs subvolume snapshot` command takes more than 1 minute to finish, and during this 1 minitue, every write to the disk is extremly slow (< 1MB/s). Also, when the disk is totally idel, if I tried to `dd if=/dev/zero of=test bs=1M count=100 conv=fdatasync`, this `dd` just hangs, and iostat told me the disk is reading at speed of ~5MB/s, not a bit was writtien to the disk. Smartctl showed the disk is fine, and btrfs device stat reports no error. This server is a KVM virtual machine and btrfs is on two 4T qcow2 images, with virtio cache=none and io=native. Uname -a is Linux debian-opensource 3.16.0-0.bpo.4-amd64 #1 SMP Debian 3.16.7-ckt9-3~deb8u1~bpo70+1 (2015-04-27) x86_64 GNU/Linux Btrfs fi show is Label: none uuid: 179cb1ff-fe45-4e02-996a-a9ad02f766b9 Total devices 2 FS bytes used 5.46TiB devid 1 size 4.00TiB used 3.08TiB path /dev/vdb devid 2 size 4.00TiB used 2.41TiB path /dev/vdc Btrfs v3.14.1 SysRq+w reports: http://pastebin.com/ZRPwZiUQ Mount Options: (rw,noatime,space_cache,autodefrag) No dmesg errors. So I wanna know: 1. Why is it slow? 2. Is it bad to use btrfs inside KVM with qcow2? 3. Everything was fast before, can I somehow "reset" the filesystem without losing data? Cheers, -- Justin Wong Blog: https://bigeagle.me/ Fingerprint: 15CC 6A61 738B 1599 0095 E256 CB67 DA7A 865B AC3A
Attachment:
signature.asc
Description: PGP signature
