=== cpaelzer_ is now known as cpaelzer === EugenMayer597 is now known as EugenMayer59 === hexa- is now known as h === h is now known as hexa- [21:04] related to the e2fsprogs investigation. I've confirmed the failure path. It's multiple calls to resize2fs /dev/root on the first boot of an instance in Azure lunar development images. daily lunar Images dated 202303160 (from /etc/cloud/build.info) proposed suggest "no changes needed" if `resize2fs /dev/root` is called manually after cloud-init has completed resize operations. [21:06] Any Azure lunar daily image 202303170 or later (containing e2fsprogrs 1.47.0) seems to have that 4K offset checksum mismatch issue [21:06] $ sudo resize2fs /dev/root [21:06] resize2fs 1.47.0 (5-Feb-2023) [21:06] resize2fs: Superblock checksum does not match superblock while trying to open /dev/root [21:06] Couldn't find valid filesystem superblock. [21:07] still poking around at the root cause though. a big delta between 1.46.1.rc1 and 1.47. but likely the discussion you've already had in backlog here. w/ minimal. [21:08] after any subsequent boots of the system, this problem goes away. and resize2fs reports The filesystem is already 7836155 (4k) blocks long. Nothing to do! [21:16] testing 1.46.6_rc1-1.1 locally in lxd, I see: [21:16] root@lunar-daily:~# resize2fs --version [21:16] resize2fs 1.46.6-rc1 (12-Sep-2022) [21:16] root@lunar-daily:~# resize2fs /dev/root [21:16] resize2fs 1.46.6-rc1 (12-Sep-2022) [21:16] resize2fs: Superblock checksum does not match superblock while trying to open /dev/root [21:16] Couldn't find valid filesystem superblock. [21:18] (this can be reproduced in a vm on lxd by increasing the root disk size) [22:21] holmanb: just to clarify, is this with metadata_csum_seed option disabled for the ext4 fs? [22:51] minimal: looks like it to me if I check 1.46.6~rc1.1 in source `misc/mke2fs.conf.in` shows ext4 = { [22:51] features = has_journal,extent,huge_file,flex_bg,metadata_csum,64bit,dir_nlink,extra_isize [22:51] } [22:51] so it has metadata_csum set but not metadata_csum_seed [22:52] and 1.47.0 has a patch that specifically removes `metadata_csum_seed` config