mbox series

[v8,0/3] pkg-database and systemd-sysext image

Message ID 20240512063159.1249522-1-johannes.schneider@leica-geosystems.com
Headers show
Series pkg-database and systemd-sysext image | expand

Message

SCHNEIDER Johannes May 12, 2024, 6:31 a.m. UTC
systemd-sysext allows to overlay another image (or multiple) ontop of
a "base-image" = the current rootfs, via the use of overlayfs; to add
tools and features meant for development purposes.

To quote the documentation on systemd-sysext:
" ...addition in order to make debugging/development easier). System
extension images should not be misunderstood as a generic software
packaging framework, ..."

To build a lean image, that only holds packages that are not already
part of the base-image, a snapshot of the package-database is taken
after the installation of the base-rootfs is done, and picked up again
when collecting the rootfs of such a extension image.

with all this in place an example usage could look like this:
some-core-image.bb
  inherit core-image
  IMAGE_GEN_PKGDBFS = "1"

extending-image.bb
  inherit image-sysext
  IMAGE_FSTYPES = "squashfs"
  IMAGE_BASE_PKGDB = "some-core-image"
  # the above pointing at a package-db similar to:
  # build/deploy/images/$MACHINE/some-core-image-$MACHINE-20240210172305-pkgdb.rootfs.tar.gz

then on the device, running some-core-image, with the extension image placed at FN:
$> ln -s "$FN" /run/extensions/$(basename $FN).raw
$> systemd-sysext list
$> SYSTEMD_LOG_LEVEL=debug systemd-sysext merge

As long as the VERSION_ID of the extension image matches the os-release
in the base image, the above commands return sucessfully;
for details on the compativility check see the docs for systemd-sysext.

=========

changes with v2:
        rebase from 'kirkstone' onto 'master'

changes with v3:
	incorporate review suggestions for simplification
	add task dependency handling
	add oe-selftest for the pkgdb handling
	add variable documentation and
	some more comments, and examples in the commit-msg

changes with v4:
	rebase onto 'master' => no functional changes
	fixed patchtest findings

changes with v5:
        rebase onto 'master'
	add '.sysext' to the deployed symlink name
        sidenote on the tests and autobuilder failure: run locally they succeed, e.g.:
          #> oe-selftest --verbose -r imagefeatures.ImageFeatures.test_image_gen_pkgdbfs

changes with v6:
	tests: restructure to call 'bitbake' only once in the testcase itself
	  (in hopes of solving the autobuilder problem; local test runs succeed)

changes with v7:
        tests: undo the restructuring of done in v6, in favour of explicitly forcing steps:
	  '-c rootfs -f'.  since concurrency of the autobuilder, and reusing artifacts/sstate-caching is probably the issue?

changes with v8:
        tests: fixed missing feature=PACKAGE_CLASSES during 'test_image_gen_pkgdbfs'
	   which is now split into three, one for each of the three available packagemanager: ipk, deb, rpm

Comments

Alexandre Belloni May 15, 2024, 7:09 p.m. UTC | #1
Hello,

It seems that this reliably fails oe-selftest-armhost. I didn't really
pay attention until now beause we had other issues but in this build,
those are the only failing tests:

https://autobuilder.yoctoproject.org/typhoon/#/builders/127/builds/3322/steps/15/logs/stdio

On 12/05/2024 08:31:56+0200, Johannes Schneider via lists.openembedded.org wrote:
> systemd-sysext allows to overlay another image (or multiple) ontop of
> a "base-image" = the current rootfs, via the use of overlayfs; to add
> tools and features meant for development purposes.
> 
> To quote the documentation on systemd-sysext:
> " ...addition in order to make debugging/development easier). System
> extension images should not be misunderstood as a generic software
> packaging framework, ..."
> 
> To build a lean image, that only holds packages that are not already
> part of the base-image, a snapshot of the package-database is taken
> after the installation of the base-rootfs is done, and picked up again
> when collecting the rootfs of such a extension image.
> 
> with all this in place an example usage could look like this:
> some-core-image.bb
>   inherit core-image
>   IMAGE_GEN_PKGDBFS = "1"
> 
> extending-image.bb
>   inherit image-sysext
>   IMAGE_FSTYPES = "squashfs"
>   IMAGE_BASE_PKGDB = "some-core-image"
>   # the above pointing at a package-db similar to:
>   # build/deploy/images/$MACHINE/some-core-image-$MACHINE-20240210172305-pkgdb.rootfs.tar.gz
> 
> then on the device, running some-core-image, with the extension image placed at FN:
> $> ln -s "$FN" /run/extensions/$(basename $FN).raw
> $> systemd-sysext list
> $> SYSTEMD_LOG_LEVEL=debug systemd-sysext merge
> 
> As long as the VERSION_ID of the extension image matches the os-release
> in the base image, the above commands return sucessfully;
> for details on the compativility check see the docs for systemd-sysext.
> 
> =========
> 
> changes with v2:
>         rebase from 'kirkstone' onto 'master'
> 
> changes with v3:
> 	incorporate review suggestions for simplification
> 	add task dependency handling
> 	add oe-selftest for the pkgdb handling
> 	add variable documentation and
> 	some more comments, and examples in the commit-msg
> 
> changes with v4:
> 	rebase onto 'master' => no functional changes
> 	fixed patchtest findings
> 
> changes with v5:
>         rebase onto 'master'
> 	add '.sysext' to the deployed symlink name
>         sidenote on the tests and autobuilder failure: run locally they succeed, e.g.:
>           #> oe-selftest --verbose -r imagefeatures.ImageFeatures.test_image_gen_pkgdbfs
> 
> changes with v6:
> 	tests: restructure to call 'bitbake' only once in the testcase itself
> 	  (in hopes of solving the autobuilder problem; local test runs succeed)
> 
> changes with v7:
>         tests: undo the restructuring of done in v6, in favour of explicitly forcing steps:
> 	  '-c rootfs -f'.  since concurrency of the autobuilder, and reusing artifacts/sstate-caching is probably the issue?
> 
> changes with v8:
>         tests: fixed missing feature=PACKAGE_CLASSES during 'test_image_gen_pkgdbfs'
> 	   which is now split into three, one for each of the three available packagemanager: ipk, deb, rpm

> 
> -=-=-=-=-=-=-=-=-=-=-=-
> Links: You receive all messages sent to this group.
> View/Reply Online (#199209): https://lists.openembedded.org/g/openembedded-core/message/199209
> Mute This Topic: https://lists.openembedded.org/mt/106051545/3617179
> Group Owner: openembedded-core+owner@lists.openembedded.org
> Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub [alexandre.belloni@bootlin.com]
> -=-=-=-=-=-=-=-=-=-=-=-
>
SCHNEIDER Johannes May 15, 2024, 10:42 p.m. UTC | #2
Hoi Alexandre,


and thanks again for patiently pointing out the autobuilder problems :-)

"only" the test_image_gen_pkgdbfs_{ipk,deb,rpm} fail, the test_image_use_pkgdbfs succeeds

picking apart the yocto-autobuilder-helper/config.json showed the difference: basically all oe-selftest-* use
 "MACHINE" : "qemux86-64"
**but** oe-selftest-armhost sets:
"MACHINE": "qemuarm64"

the test_image_gen_pkgdbfs_ checks to sanity check the packagamanager state looked for a "kernel-image" package in the listing... but for quemuarm64 "packagegroup-core-boot" replaces that

v9 of the patchstack adapts that particular assert(s) to look for either the kernel-image or core-boot depending on the set MACHINE


gruß
Johannes