24.10.2018 3:36, Marc MERLIN пишет: > Normally when btrfs fi show will show lost space because > your trees aren't balanced. > Balance usually reclaims that space, or most of it. > In this case, not so much. > > kernel 4.17.6: > > saruman:/mnt/btrfs_pool1# btrfs fi show . > Label: 'btrfs_pool1' uuid: fda628bc-1ca4-49c5-91c2-4260fe967a23 > Total devices 1 FS bytes used 186.89GiB > devid 1 size 228.67GiB used 207.60GiB path /dev/mapper/pool1 > > Ok, I have 21GB between used by FS and used in block layer. > > saruman:/mnt/btrfs_pool1# btrfs balance start -dusage=40 -v . > Dumping filters: flags 0x1, state 0x0, force is off > DATA (flags 0x2): balancing, usage=40 > Done, had to relocate 1 out of 210 chunks > saruman:/mnt/btrfs_pool1# btrfs balance start -musage=60 -v . > Dumping filters: flags 0x6, state 0x0, force is off > METADATA (flags 0x2): balancing, usage=60 > SYSTEM (flags 0x2): balancing, usage=60 > Done, had to relocate 4 out of 209 chunks > saruman:/mnt/btrfs_pool1# btrfs fi show . > Label: 'btrfs_pool1' uuid: fda628bc-1ca4-49c5-91c2-4260fe967a23 > Total devices 1 FS bytes used 186.91GiB > devid 1 size 228.67GiB used 205.60GiB path /dev/mapper/pool1 > > That didn't help much, delta is now 19GB > > saruman:/mnt/btrfs_pool1# btrfs balance start -dusage=80 -v . > Dumping filters: flags 0x1, state 0x0, force is off > DATA (flags 0x2): balancing, usage=80 > Done, had to relocate 8 out of 207 chunks > saruman:/mnt/btrfs_pool1# btrfs fi show . > Label: 'btrfs_pool1' uuid: fda628bc-1ca4-49c5-91c2-4260fe967a23 > Total devices 1 FS bytes used 187.03GiB > devid 1 size 228.67GiB used 201.54GiB path /dev/mapper/pool1 > > Ok, now delta is 14GB > > saruman:/mnt/btrfs_pool1# btrfs balance start -musage=80 -v . > Dumping filters: flags 0x6, state 0x0, force is off > METADATA (flags 0x2): balancing, usage=80 > SYSTEM (flags 0x2): balancing, usage=80 > Done, had to relocate 5 out of 202 chunks > saruman:/mnt/btrfs_pool1# btrfs fi show . > Label: 'btrfs_pool1' uuid: fda628bc-1ca4-49c5-91c2-4260fe967a23 > Total devices 1 FS bytes used 188.24GiB > devid 1 size 228.67GiB used 203.54GiB path /dev/mapper/pool1 > > and it's back to 15GB :-/ > > How can I get 188.24 and 203.54 to converge further? Where is all that > space gone? >
Most likely this is due to partially used extents which has been explained more than once on this list. When large extent is partially overwritten, extent is not physically split - it remains allocated in full but only part of it is referenced. Balance does not change it (at least that is my understanding) - it moves extents, but here we have internal fragmentation inside of extent. Defragmentation should rewrite files, but if you have snapshots, it is unclear if there will be any gain. I wonder if there is any tool that can compute physical vs. logical space consumption (i.e. how much space in extent is actually referenced). Should be possible using python-btrfs but probably time consuming as it needs to walk each extent.
