container for 0.17 is almost twice the size of 0.16 ??

[aleks-mariusz]

So with about 4.7gb of free space on my HAOS vm

# df -h /var/lib/docker
Filesystem                Size      Used Available Use% Mounted on
/dev/sda8                30.8G     25.1G      4.4G  85% /mnt/data

...i was happily running 0.16.4 for a while (which btw was already one of my largest containers on my system):

ghcr.io/blakeblackshear/frigate:0.16.4                       538165620071       2.94GB             0B   U

then i tried to update to 0.17 and initially i didn't think anything of it until i saw it fail the two times..

Failed to perform the action update/install. Error updating Frigate: An unknown error occurred with addon ccabaaf_frigate. Check supervisor logs for details (check with 'ha supervisor logs')

and i soon realized it was because:

2026-03-03 19:55:23.775 ERROR (MainThread) [supervisor.docker.manager] No space left on disk
2026-03-03 19:55:23.778 ERROR (MainThread) [supervisor.addons.addon] Could not pull image to update addon ccab4aaf_frigate: No space left on disk

i would think the supervisor should know a.) how much space is on the filesystem and b.) how big the new container will be (but i guess this is not really a question for the frigate owner, just they're the one who lead to triggered this suboptimal approach).

but what was shocking is seems that the 0.17 container is almost twice as big?!

ghcr.io/blakeblackshear/frigate:0.17.0                                 a17b8436f26a       5.46GB             0B    U

so my question - is it intended for such a massive size change going from 0.16.4 to 0.17.0 ??

[NickM-27]

it is generally expected to be larger as tensorflow is included for the classification model feature, so an uncompressed size of 5.46GB is correct. Compressed it only went up a few hundred MB

[aleks-mariusz]

it seems to not be stored compressed on the filesystem however, which is where the impact of such a drastic container size change hits hardest..

is there any options to not have so much have to be stored ? there was no warning in the changelog about such a largely increased image size (other than mention of new features, but those usually are expected to be of a few kilobytes to megabytes more usage, not nearly 200% size)