compsize(8) btrfs compsize(8)

compsize - calculate compression ratio of a set of files on btrfs

compsize file-or-dir [ file-or-dir ... ]

compsize takes a list of files on a btrfs filesystem (recursing directories) and measures used compression types and the effective compression ratio. Besides compression, compsize shows the effect of reflinks (cp --reflink, snapshots, deduplication), and certain types of btrfs waste.

The program gives a report similar to:
Processed 90319 files.
Type Perc Disk Usage Uncompressed Referenced
TOTAL 79% 1.4G 1.8G 1.9G
none 100% 1.0G 1.0G 1.0G
lzo 53% 446M 833M 843M

The fields above are:

compression algorithm
disk usage/uncompressed (compression ratio)
blocks on the disk; this is what storing these files actually costs you (save for RAID considerations)
uncompressed extents; what you would need without compression - includes deduplication savings and pinned extent waste
apparent file sizes (sans holes); this is what a traditional filesystem that supports holes and efficient tail packing, or tar -S, would need to store these files

Let's see this on an example: a file 128K big is stored as a single extent A which compressed to a single 4K page. It then receives a write of 32K at offset 32K, which also compressed to a single 4K page, stored as extent B.

The file now appears as:
+-------+-------+---------------+
extent A | used | waste | used |
+-------+-------+---------------+
extent B | used |
+-------+

The "waste" inside extent A can't be gotten rid until the whole extent is rewritten (for example by defrag). If compressed, the whole extent needs to be read every time that part of the file is being read, thus the "waste" is still required.

In this case, we have: Disk Usage: 8KB, Uncompressed: 160K, Referenced: 128K.

Show raw byte counts rather than human-friendly sizes.
Skip files and directories on different file systems.

Displays partial data for files processed so far.

Recently written files may show as not taking any space until they're actually allocated and compressed; this happens once they're synced or on natural writeout, typically on the order of 30 seconds.

The ioctls used by this program require root.

Inline extents are considered to be always unique, even if they share the same bytes on the disk.

This program doesn't currently support filesystems above 8TB on 32-bit machines but neither do other btrfs tools.

2017-09-04 btrfs