packaging/rpm-ostree.spec: Update to sync with rawhide #5047

jmarrero · 2024-08-14T15:46:52Z

The specs between upstream, fedora and centos are out of sync. This attempts to sync it to restart using it as the source of truth.

jlebon · 2024-08-14T18:36:34Z

packaging/rpm-ostree.spec

@@ -132,12 +138,11 @@ Requires:       librepo%{?_isa} >= %{librepo_version}
 # rpm-ostree wraps more of ostree (such as `ostree admin unlock` etc.)
 Requires: ostree
 Requires: bubblewrap
-Requires: fuse
+Requires: fuse3


This seems to come from https://src.fedoraproject.org/rpms/rpm-ostree/c/3c602a23787fd2df873c0b18df3133c9fec4b66a, but the reasoning there is not sufficient.

OSTree already lists its fuse dependencies as needed. git archeology shows the fuse here is because we call fusermount -u: #443.

fuse3 ships fusermount3 and I'm not sure if they're supposed to be interchangeable for unmount purposes (though I would assume it mostly boils down to a umount() syscall) but clearly using fusermount -u to unmount a FUSE mount from rofiles-fuse (compiled with libfuse3) has been working for a while, ever since we started compiling ostree in Fedora against libfuse3.

Hmm, wonder if this is the root cause of coreos/coreos-assembler#3848 somehow; maybe that mismatch stopped working recently?

Anyway, I think ideally there'd be an rofiles-fuse -u which would just proxy to the right fusermount vs fusermount3 based on compiled options. Barring that, in practice I think this is fine, but we should probably change our code to use fusermount3 -u also.

Hmm, wonder if this is the root cause of coreos/coreos-assembler#3848 somehow; maybe that mismatch stopped working recently?

At least testing locally with latest fuse bits, it seems to work fine to unmount with fusermount -u an rofiles-fuse mount, so maybe not.

But still, we need the Requires here to agree with the actual binary we use in the code.

cc also fedora-silverblue/issue-tracker#589

Yeah, definitely a mess here...a perfect example of the pitfalls of trying to maintain spec files upstream.

A larger fix of course is for us to just stop using rofiles-fuse (and go with overlayfs, but...)

Won't this break if we explicitly call fusermount as the fuse3 package does not provide such a binary?

We should likely revert that change until we've moved the logic in rpm-ostree to try both fuse2 and fuse3 (or add a wrapper).

Won't this break if we explicitly call fusermount as the fuse3 package does not provide such a binary?

Yup, this is what I meant higher up with:

we need the Requires here to agree with the actual binary we use in the code.

I think rather than reverting, it makes more sense to just plow forward and move over to the v3 version. Pushed a commit for that.

Yeah, definitely a mess here...a perfect example of the pitfalls of trying to maintain spec files upstream.

Yeah, I'd be OK revisiting that. Historically, the canonical version lived here because it often required changes along with upstream changes (e.g. bundled libdnf deps). Since now things are calmer, it might make sense to just move it back to dist-git.

rofiles-fuse is now built against libfuse3 on all the platforms we care about in rpm-ostree. Accordingly also switch over the fusermount utility we use to the v3 version to match. This also allows composes to drop out libfuse2 if nothing else pulls it in. See also: coreos#5047 (comment)

travier · 2024-08-20T13:29:08Z

packaging/rpm-ostree.spec

@@ -268,5 +273,4 @@ fi
 %files devel -f files.devel

 %changelog
-* Thu Nov 17 2022 Colin Walters <walters@verbum.org> - 2022.15-3
- Dummy change to satisfy rpm timestamp clamping
+%autochangelog


I vaguely remember that breaking some things with copr repos. Might have to look if this works with our continuous copr builds.

I see this comment:
https://fedoramagazine.org/use-rpmautospec-in-fedora-linux/#comment-550158

from a year ago hopefully now it's good 🤞

jlebon · 2024-08-28T20:04:48Z

I think this is actually needed now to unbreak CI: #5069 (comment)

jmarrero · 2024-08-28T20:25:03Z

I think this is actually needed now to unbreak CI: #5069 (comment)

I just changed it back to fuse then. Let me know if you think it needs any other changes.
or are you saying we need fuse 3?

cgwalters · 2024-08-29T11:56:00Z

Ohhh I finally see now:

warning: Found 1 packages to download in cache-only mode

OK and we're actually explicitly turning off network. Now I understand this isn't just a flake. Hmm, we should clearly make that a hard error instead, at least for the cosa case. Adding --cache-only-strict or something.

cgwalters · 2024-08-29T12:02:50Z

I'm so confused by this though still...we still Requires: fuse in f40, which is what we're trying to build here right? Ohh I think I see, the fuse requirement change actually changed in f40 sometime post-"GA" of f40. Probably because we merged git branches? Ugh...

Anyways yes I see:
https://github.com/coreos/fedora-coreos-config/blob/f5ea8ce3c5b2fcc23aca646885ceaae134936e48/manifest-lock.x86_64.json#L717

So I guess yes we just need to update to fuse3 here.

jlebon · 2024-08-29T12:29:31Z

or are you saying we need fuse 3?

Yup, I had pushed a commit to do that. I'm saying we should match the libfuse version that rofiles-fuse uses.

cgwalters · 2024-08-29T13:17:32Z

I don't see an updated commit?

cgwalters · 2024-08-29T13:49:57Z

OK though what really needs to be done here is to upstream make the spec use distribution conditionals, which is what we should have done from the start to help us maintain a single spec

cgwalters · 2024-08-29T13:52:16Z

Hmm do we really even need this dependency at all? Today the ostree package has rofiles-fuse and ostree has the dep...

cgwalters · 2024-08-29T14:03:02Z

OK my MRs-for-dist-git aren't set up, any objections to me pushing this to fedora and here?

From 92516fb809e96ed6a29f35d7aad0619add63b972 Mon Sep 17 00:00:00 2001
From: Colin Walters <walters@verbum.org>
Date: Thu, 29 Aug 2024 10:01:55 -0400
Subject: [PATCH] Drop `Requires: fuse3`

We already have this transitively through the `Requires: ostree`
and dealing with fuse-vs-fuse3 is a pain; cc
https://github.com/coreos/rpm-ostree/pull/5047
---
 rpm-ostree.spec | 1 -
 1 file changed, 1 deletion(-)

diff --git a/rpm-ostree.spec b/rpm-ostree.spec
index 10d4c79..a2d6cb5 100644
--- a/rpm-ostree.spec
+++ b/rpm-ostree.spec
@@ -136,7 +136,6 @@ Requires:       librepo%{?_isa} >= %{librepo_version}
 # rpm-ostree wraps more of ostree (such as `ostree admin unlock` etc.)
 Requires: ostree
 Requires: bubblewrap
-Requires: fuse3
 
 # For container functionality
 # https://github.com/coreos/rpm-ostree/issues/3286
-- 
2.44.0

jmarrero · 2024-08-29T14:47:15Z

lgtm, I'll update it here.

jlebon · 2024-08-29T16:25:06Z

Had a chat with @jmarrero and @cgwalters on this. There was agreement that ideally, there'd be an rofiles-fuse -u. That said for now, we will:

change the spec file to be conditional: on el9, we Requires: fuse, everywhere else, we Requires: fuse3. We also flip a build-time conditional based on that same spec condition.
change the code to use this build time conditional to know if to call fusermount or fusermount3.

cgwalters · 2024-08-29T16:48:44Z

There was agreement that ideally, there'd be an rofiles-fuse -u.

Maybe...but hold up, I think this may just be Fedora packaging fuse in a broken way. At least looking at e.g. ubuntu, they have a lib fuse for each of 2 and 3 but only a single fusermount binary in https://packages.ubuntu.com/focal/fuse and I would bet money that it's totally compatible to run fusermount (linked to v2) against a mountpoint created with libfuse3.

cgwalters · 2024-08-29T21:01:51Z

Filed https://bugzilla.redhat.com/show_bug.cgi?id=2308595

We have been building with fuse but changed to fuse3 on: https://src.fedoraproject.org/rpms/rpm-ostree/c/3c602a23787fd2df873c0b18df3133c9fec4b66a However our code is just calling fuse's fusermount. We are updating our spec and code based on the discusion on: coreos#5047 Co-authored-by: Colin Walters <walters@verbum.org>

HuijingHei · 2024-09-03T08:45:27Z

I am confused why the CI failed with can not download fuse, but IMU we do not need it for fedora build, see #5047 (comment)

[2024-08-29T19:40:05.618Z] warning: Found 1 packages to download in cache-only mode

[2024-08-29T19:40:05.618Z] Downloading from 'fedora'...done

[2024-08-29T19:40:05.618Z] error: Installing packages: Downloading from 'fedora': Cannot download Packages/f/fuse-2.9.9-21.fc40.x86_64.rpm: All mirrors were tried; Last error: Curl error (6): Couldn't resolve host name for https://dl.fedoraproject.org/pub/fedora/linux/releases/40/Everything/x86_64/os/Packages/f/fuse-2.9.9-21.fc40.x86_64.rpm [Could not resolve host: dl.fedoraproject.org]

[2024-08-29T19:40:08.123Z] failed to execute cmd-build: exit status 1

cgwalters · 2024-09-03T15:28:52Z

packaging/rpm-ostree.spec

+# However our code is just calling fuse's fusermount.
+# We are updating our spec and code based on the discusion on:
+# https://github.com/coreos/rpm-ostree/pull/5047
+%if 0%{?el} <= 9


I don't think this conditional is right...I suspect it's evaluating to true on fedora too and that's why we're hitting this bug still.

I think it may need to be e.g.

%if 0%{?rhel} && 0%{?el} <= 9

or so?

https://github.com/search?q=+0%25%7B%3Fel%7D&type=code

I see people doing:

%if 0%{?el} >= 6 || 0%{?rhel} >= 6

ok, trying your conditional in this push.

something is still weird with the conditional:
https://gitlab.com/redhat/centos-stream/rpms/rpm-ostree/-/merge_requests/57#note_2093239477

I don't get why... maybe switching this around to:

%if 0%{?fedora} || 0%{?rhel} > 9 || 0%{?el} > 9 Requires: fuse3 %else Requires: fuse %endif

Would work... I am trying to find another rpm with an example.

Apparently 0%{?rhel} should match the family of distros based on this: https://docs.fedoraproject.org/en-US/packaging-guidelines/DistTag/#_conditionals I wonder if the el conditional is throwing it.

%if 0%{?fedora} || 0%{?rhel} > 9 || 0%{?el} > 9

Check on centos stream 9, get %{?rhel} -> 9, but get %{?el} null. IMU, the condition LGTM.

$ podman run --rm -it quay.io/centos/centos:stream9 bash [root@b7210efdfe5a /]# rpm --eval %{?rhel} 9 [root@b7210efdfe5a /]# rpm --eval %{?el}

We have been building with fuse but changed to fuse3 on: https://src.fedoraproject.org/rpms/rpm-ostree/c/3c602a23787fd2df873c0b18df3133c9fec4b66a However our code is just calling fuse's fusermount. We are updating our spec and code based on the discusion on: coreos#5047 Co-authored-by: Colin Walters <walters@verbum.org>

The specs between upstream, fedora and centos are out of sync. This attempts to sync it to restart using it as the source of truth.

We have been building with fuse but changed to fuse3 on: https://src.fedoraproject.org/rpms/rpm-ostree/c/3c602a23787fd2df873c0b18df3133c9fec4b66a However our code is just calling fuse's fusermount. We are updating our spec and code based on the discusion on: coreos#5047 Co-authored-by: Colin Walters <walters@verbum.org>

jmarrero requested a review from jlebon August 14, 2024 15:47

jlebon reviewed Aug 14, 2024

View reviewed changes

This was referenced Aug 20, 2024

[branched] Start tracking Fedora 41 coreos/fedora-coreos-config#3092

Merged

tracker: Rebase onto Fedora 41 coreos/fedora-coreos-tracker#1695

Open

jlebon force-pushed the more-spec branch from e4bec9a to 448ff92 Compare August 20, 2024 13:27

travier reviewed Aug 20, 2024

View reviewed changes

marmijo mentioned this pull request Aug 21, 2024

tracker: package additions/removals for F41 coreos/fedora-coreos-tracker#1785

Open

jlebon mentioned this pull request Aug 28, 2024

core: Continue to write now-empty rpmostree.modules metadata #5069

Merged

jmarrero force-pushed the more-spec branch from 448ff92 to fcf367c Compare August 28, 2024 20:24

cgwalters force-pushed the more-spec branch from fcf367c to 6935480 Compare August 29, 2024 13:49

jmarrero force-pushed the more-spec branch from 6935480 to 0f847cd Compare August 29, 2024 14:48

jmarrero mentioned this pull request Aug 29, 2024

Release 2024.8 #5073

Merged

jmarrero force-pushed the more-spec branch from db361f0 to a594e67 Compare August 29, 2024 19:23

jmarrero mentioned this pull request Aug 30, 2024

rust/src/bwrap.rs: check for fuse version #5074

Closed

cgwalters reviewed Sep 3, 2024

View reviewed changes

jmarrero force-pushed the more-spec branch from a594e67 to 14205cc Compare September 3, 2024 16:44

jmarrero and others added 2 commits September 3, 2024 14:08

packaging/rpm-ostree.spec: Update to sync with rawhide

04ebd25

The specs between upstream, fedora and centos are out of sync. This attempts to sync it to restart using it as the source of truth.

jmarrero force-pushed the more-spec branch from 14205cc to 26a0107 Compare September 3, 2024 18:09

cgwalters approved these changes Sep 3, 2024

View reviewed changes

cgwalters enabled auto-merge September 3, 2024 20:08

cgwalters merged commit 3c1bcef into coreos:main Sep 3, 2024
17 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

packaging/rpm-ostree.spec: Update to sync with rawhide #5047

packaging/rpm-ostree.spec: Update to sync with rawhide #5047

jmarrero commented Aug 14, 2024

jlebon Aug 14, 2024

jlebon Aug 14, 2024

cgwalters Aug 19, 2024

cgwalters Aug 19, 2024

travier Aug 20, 2024 •

edited

Loading

jlebon Aug 20, 2024

jlebon Aug 20, 2024

travier Aug 20, 2024

jmarrero Aug 28, 2024

jlebon commented Aug 28, 2024

jmarrero commented Aug 28, 2024 •

edited

Loading

cgwalters commented Aug 29, 2024

cgwalters commented Aug 29, 2024

jlebon commented Aug 29, 2024

cgwalters commented Aug 29, 2024

cgwalters commented Aug 29, 2024

cgwalters commented Aug 29, 2024

cgwalters commented Aug 29, 2024

jmarrero commented Aug 29, 2024

jlebon commented Aug 29, 2024

cgwalters commented Aug 29, 2024

cgwalters commented Aug 29, 2024

HuijingHei commented Sep 3, 2024

cgwalters Sep 3, 2024

jmarrero Sep 3, 2024

jmarrero Sep 3, 2024

jmarrero Sep 6, 2024

jmarrero Sep 6, 2024 •

edited

Loading

HuijingHei Sep 6, 2024

packaging/rpm-ostree.spec: Update to sync with rawhide #5047

packaging/rpm-ostree.spec: Update to sync with rawhide #5047

Conversation

jmarrero commented Aug 14, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

travier Aug 20, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jlebon commented Aug 28, 2024

jmarrero commented Aug 28, 2024 • edited Loading

cgwalters commented Aug 29, 2024

cgwalters commented Aug 29, 2024

jlebon commented Aug 29, 2024

cgwalters commented Aug 29, 2024

cgwalters commented Aug 29, 2024

cgwalters commented Aug 29, 2024

cgwalters commented Aug 29, 2024

jmarrero commented Aug 29, 2024

jlebon commented Aug 29, 2024

cgwalters commented Aug 29, 2024

cgwalters commented Aug 29, 2024

HuijingHei commented Sep 3, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jmarrero Sep 6, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

travier Aug 20, 2024 •

edited

Loading

jmarrero commented Aug 28, 2024 •

edited

Loading

jmarrero Sep 6, 2024 •

edited

Loading