qemu-img convert is TOO slow

[Dmitry]

I have 100gb raw image with only 7gb used data in it:

$ qemu-img info from.img
image: from.img
file format: raw
virtual size: 100G (107374182400 bytes)
disk size: 100G
$ virt-df -h from.img
Filesystem                                Size       Used  Available  Use%
from.img:/dev/sda1                 98G       7.2G        89G    8%

And try to run
virt-sparsify --convert qcow2 from.img from-spars.qcow2

this create temprorary qcow2 image backed by from.img

$ qemu-img info /tmp/sparsify388744.qcow2
image: /tmp/sparsify388744.qcow2
file format: qcow2
virtual size: 100G (107374182400 bytes)
disk size: 91G
cluster_size: 65536
backing file: /vmimages/from.img
Format specific information:
    compat: 1.1
    lazy refcounts: false
    refcount bits: 16
    corrupt: false


and when virt-sparsify internally do

qemu-img convert -f qcow2 -O 'qcow2' '/tmp/sparsify388744.qcow2'
'./front-spars.qcow2'

it processes it TOO slow ( total time of convert ~2days ). If run
pidstat on this qemu-img process pidstat says that it consumes 100%
system time

$ pidstat -p  29211 1
Linux 5.4.88-1.el7.elrepo.x86_64 (v2210)        04/24/23
_x86_64_        (104 CPU)

19:56:48      UID       PID    %usr %system  %guest    %CPU   CPU  Command
19:56:49        0     29211    0.00  100.00    0.00  100.00    66  qemu-img
19:56:50        0     29211    0.00  100.00    0.00  100.00    42  qemu-img
19:56:51        0     29211    0.00  100.00    0.00  100.00    73  qemu-img

When I run perf "record -a -g -F 997 sleep 10" the most top cpu
consumer ( after idle ) is qemu-img :

            |                     cpuidle_enter_state
            |                     intel_idle
            |
             --2.78%--intel_idle

    28.90%     0.00%  qemu-img         libpthread-2.17.so
[.] 0x00007f515acbaddd
            |
            ---0x7f515acbaddd
               entry_SYSCALL_64_after_hwframe
               do_syscall_64
               __x64_sys_lseek
               |
                --28.90%--ksys_lseek
                          shmem_file_llseek
                          |
                           --28.83%--shmem_seek_hole_data
                                     |
                                     |--24.38%--find_get_entries
                                     |          |
                                     |           --0.81%--xas_find
                                     |                     |
                                     |                      --0.62%--xas_load
                                     |
                                      --3.14%--__pagevec_release
                                                |
                                                 --3.12%--release_pages



Why "qemu-img convert" do so many find_get_entries ? How can I speed
up this process?





--
  Dmitry