Open karuboniru opened 10 months ago
Hmmm... so /usr/share/rpm/rpmdb.sqlite-shm
is a somewhat special file I think because it may be memory mapped. I wonder if we're somehow leaking writes to it through rofiles-fuse or so?
How reproducible is this?
Not quite, I only noticed this when doing the rebase for the first time and once after playing with rpm-ostree compose
. I will try to do simliar steps in virtual machine when had time.
I just ran into this same exact issue with Fedora Silverblue 40 and various overrides applied. I went from a rpm-ostree repository to a container native image with rpm-ostree rebase
.
Describe the bug
I am trying to switch a silverblue system from layering to container image, and this end up with a broken commit
Reproduction steps
Start with a system with some overrrides:
Rebase to a container image, with simliar package list
Ends up with broken commit,
ostree fsck
complains about broken files. Andbtrfs scrub
don't report any corrupted file.Expected behavior
It should not break the internal status.
Actual behavior
Got corrupted file.
System details
Additional information
Doing a reset before rebasing will workaround this issue. If rebasing from dirty image is not supported I believe a warning before the actual execution is needed to avoid confusion.
This is not a filesystem error since
btrfs scrub
return no error.And the file being broken is
Also, I somehow break a working system without doing layering, maybe during playing with
rpm-ostree compose
, but that is hard to reproduce. I will update in this issue if found any pattern behind this.refs: https://github.com/fedora-silverblue/issue-tracker/issues/528