On 2018-10-27 04:19 PM, Marc MERLIN wrote: > Thanks for confirming. Because I always have snapshots for btrfs > send/receive, defrag will duplicate as you say, but once the older > snapshots get freed up, the duplicate blocks should go away, correct? > > Back to usage, thanks for pointing out that command: > saruman:/mnt/btrfs_pool1# btrfs fi usage . > Overall: > Device size: 228.67GiB > Device allocated: 203.54GiB > Device unallocated: 25.13GiB > Device missing: 0.00B > Used: 192.01GiB > Free (estimated): 32.44GiB (min: 19.88GiB) > Data ratio: 1.00 > Metadata ratio: 2.00 > Global reserve: 512.00MiB (used: 0.00B) > > Data,single: Size:192.48GiB, Used:185.16GiB > /dev/mapper/pool1 192.48GiB > > Metadata,DUP: Size:5.50GiB, Used:3.42GiB > /dev/mapper/pool1 11.00GiB > > System,DUP: Size:32.00MiB, Used:48.00KiB > /dev/mapper/pool1 64.00MiB > > Unallocated: > /dev/mapper/pool1 25.13GiB > > > I'm still seing that I'm using 192GB, but 203GB allocated. > Do I have 25GB usable: > Device unallocated: 25.13GiB > > Or 35GB usable? > Device size: 228.67GiB > - > Used: 192.01GiB > = 36GB ? >
The answer is somewhere between the two. (BTRFS's estimate of 32.44 Free is probably as close as you'll get to predicting.) So you have 7.32GB allocated but still free space for data, and 25GB of completely unallocated disk space. However, as you add more data, or create more snapshots and create metadata duplication, some of that 25GB will be allocated for Metadata. Remember that Metadata is Duplicated (so that 3.42GB of Metadata you are using now is actually using 6.84GB of disk space, out of the allocated 11GB You want to be careful that unallocated space doesn't run out. If the system runs out of usable space for metadata, it can be tricky to get yourself out of the corner. That is why a large discrepency between Data Size and Used would be a concern. If those 25GB of space were allocated to data, your would get out of space errors even if the 25GB was still unused. On that note, you seem to have a rather high metadata to data ratio.. (at least, compared to my limited experience.). Are you using noatime on your filesystems? without it, snapshots will end up causing duplicated metadata when atime updates.
<<attachment: remi.vcf>>
