public inbox for isar-users@googlegroups.com
 help / color / mirror / Atom feed
From: "MOESSBAUER, Felix" <felix.moessbauer@siemens.com>
To: "isar-users@googlegroups.com" <isar-users@googlegroups.com>,
	"Kiszka, Jan" <jan.kiszka@siemens.com>
Cc: "Su, Bao Cheng" <baocheng.su@siemens.com>
Subject: Re: [RFC 1/1] delay creation of initrd until end of rootfs install
Date: Fri, 8 Dec 2023 10:46:42 +0000	[thread overview]
Message-ID: <8ccae41f0016f01665a40ada6ab6051f1b7415dc.camel@siemens.com> (raw)
In-Reply-To: <596d1dad-1b26-4357-b970-ba30230a34eb@siemens.com>

On Wed, 2023-12-06 at 09:04 +0900, Jan Kiszka wrote:
> On 04.12.23 10:36, Felix Moessbauer wrote:
> > This patch solves major performance issues around the initramfs
> > creation by ensuring that the initrd is only created once. This is
> > implemented by stubbing the update-initramfs call during the
> > package
> > installing. After all apt operations are completed, we manually
> > trigger the initrd creation. In case a custom initramfs is used,
> > the
> > creation is completely skipped in the image rootfs, as this would
> > anyways not be used.
> > 
> > Before that, each package install that made a initrd relevant
> > change
> > triggered the update of the initrd. As we have multiple apt calls
> > during
> > the build, this step was sometimes executed multiple times. In
> > addition,
> > the apt install step is emulated, further slowing down the initrd
> > generation. On some layers on non native architecutes, this summed
> > up to
> > over 10 minutes of initrd generation time.
> > 
> > Signed-off-by: Felix Moessbauer <felix.moessbauer@siemens.com>
> > ---
> >  meta/classes/image.bbclass                    |  4 +++
> >  meta/classes/rootfs.bbclass                   | 32
> > +++++++++++++++++++
> >  .../isar-bootstrap/isar-bootstrap.inc         |  2 ++
> >  3 files changed, 38 insertions(+)
> > 
> > diff --git a/meta/classes/image.bbclass
> > b/meta/classes/image.bbclass
> > index 73f1d52c..39addc59 100644
> > --- a/meta/classes/image.bbclass
> > +++ b/meta/classes/image.bbclass
> > @@ -72,6 +72,8 @@ inherit essential
> >  
> >  ROOTFSDIR = "${IMAGE_ROOTFS}"
> >  ROOTFS_FEATURES += "clean-package-cache clean-pycache generate-
> > manifest export-dpkg-status clean-log-files clean-debconf-cache"
> > +# when using a custom initrd, do not generate one as part of the
> > image rootfs
> > +ROOTFS_FEATURES += "${@ '' if d.getVar('INITRD_IMAGE') == '' else
> > 'no-generate-initrd'}"
> >  ROOTFS_PACKAGES += "${IMAGE_PREINSTALL}
> > ${@isar_multiarch_packages('IMAGE_INSTALL', d)}"
> >  ROOTFS_MANIFEST_DEPLOY_DIR ?= "${DEPLOY_DIR_IMAGE}"
> >  ROOTFS_DPKGSTATUS_DEPLOY_DIR ?= "${DEPLOY_DIR_IMAGE}"
> > @@ -488,6 +490,8 @@ do_rootfs_quality_check() {
> >      args="${ROOTFS_QA_FIND_ARGS}"
> >      # rootfs_finalize chroot-setup.sh
> >      args="${args} ! -path ${ROOTFSDIR}/var/lib/dpkg/diversions"
> > +    # initramfs is generated outside of the image rootfs
> > +    args="${args} ! -path ${ROOTFSDIR}/boot/initrd.img*"
> >      for cmd in ${ROOTFS_POSTPROCESS_COMMAND}; do
> >          case "${cmd}" in
> >             image_postprocess_mark)
> > diff --git a/meta/classes/rootfs.bbclass
> > b/meta/classes/rootfs.bbclass
> > index 1b95115a..69e38dac 100644
> > --- a/meta/classes/rootfs.bbclass
> > +++ b/meta/classes/rootfs.bbclass
> > @@ -14,6 +14,7 @@ ROOTFS_BASE_DISTRO ?= "${BASE_DISTRO}"
> >  # 'generate-manifest' - generate a package manifest of the rootfs
> > into ${ROOTFS_MANIFEST_DEPLOY_DIR}
> >  # 'export-dpkg-status' - exports /var/lib/dpkg/status file to
> > ${ROOTFS_DPKGSTATUS_DEPLOY_DIR}
> >  # 'clean-log-files' - delete log files that are not owned by
> > packages
> > +# 'no-generate-initrd' - do not generate debian default initrd
> >  ROOTFS_FEATURES ?= ""
> >  
> >  ROOTFS_APT_ARGS="install --yes -o Debug::pkgProblemResolver=yes"
> > @@ -117,6 +118,16 @@ rootfs_configure_apt() {
> >  EOSUDO
> >  }
> >  
> > +ROOTFS_CONFIGURE_COMMAND += "rootfs_disable_initrd_generation"
> > +rootfs_disable_initrd_generation[weight] = "1"
> > +rootfs_disable_initrd_generation() {
> > +    # fully disable initrd generation
> > +    echo "replace update-initramfs with stub"
> > +    sudo mv "${ROOTFSDIR}/usr/sbin/update-initramfs" \
> > +            "${ROOTFSDIR}/usr/sbin/update-initramfs.isar"
> > +    sudo chroot "${ROOTFSDIR}" ln -s "/usr/bin/true"
> > "/usr/sbin/update-initramfs"
> > +}
> > +
> >  
> >  ROOTFS_INSTALL_COMMAND += "rootfs_install_pkgs_update"
> >  rootfs_install_pkgs_update[weight] = "5"
> > @@ -310,6 +321,27 @@ rootfs_cleanup_isar_apt() {
> >  EOSUDO
> >  }
> >  
> > +ROOTFS_POSTPROCESS_COMMAND += "rootfs_restore_initrd_tooling"
> > +rootfs_generate_initrd[weight] = "1"
> > +rootfs_restore_initrd_tooling() {
> > +    if [ -e "${ROOTFSDIR}/usr/sbin/update-initramfs.isar" ]; then
> > +        sudo mv -f "${ROOTFSDIR}/usr/sbin/update-initramfs.isar" \
> > +            "${ROOTFSDIR}/usr/sbin/update-initramfs"
> > +    fi
> > +}
> > +
> > +ROOTFS_POSTPROCESS_COMMAND +=
> > "${@bb.utils.contains('ROOTFS_FEATURES', 'no-generate-initrd', '',
> > 'rootfs_generate_initrd', d)}"
> > +rootfs_generate_initrd[weight] = "10"
> > +rootfs_generate_initrd() {
> > +    if [ -n "$(sudo find '${ROOTFSDIR}/boot' -type f -name
> > 'vmlinu[xz]*')" ]; then
> > +        sudo -E chroot "${ROOTFSDIR}" sh -c '\
> > +            export kernel_version=$(basename /boot/vmlinu[xz]* |
> > cut -d'-' -f2-); \
> > +            update-initramfs -u -v -k "$kernel_version";'
> > +    else
> > +        echo "no kernel in this rootfs, do not generate initrd"
> > +    fi
> > +}
> > +
> >  do_rootfs_postprocess[vardeps] = "${ROOTFS_POSTPROCESS_COMMAND}"
> >  do_rootfs_postprocess[network] = "${TASK_USE_SUDO}"
> >  python do_rootfs_postprocess() {
> > diff --git a/meta/recipes-core/isar-bootstrap/isar-bootstrap.inc
> > b/meta/recipes-core/isar-bootstrap/isar-bootstrap.inc
> > index 3477c2fb..4c6011bc 100644
> > --- a/meta/recipes-core/isar-bootstrap/isar-bootstrap.inc
> > +++ b/meta/recipes-core/isar-bootstrap/isar-bootstrap.inc
> > @@ -24,6 +24,8 @@ DISTRO_BOOTSTRAP_KEYFILES = ""
> >  THIRD_PARTY_APT_KEYFILES = ""
> >  DEPLOY_ISAR_BOOTSTRAP ?= ""
> >  DISTRO_BOOTSTRAP_BASE_PACKAGES = "locales"
> > +# install early, so we can stub the update-initramfs script before
> > rootfs install
> > +DISTRO_BOOTSTRAP_BASE_PACKAGES:append = ",initramfs-tools"
> >  DISTRO_BOOTSTRAP_BASE_PACKAGES:append:gnupg = ",gnupg"
> >  DISTRO_BOOTSTRAP_BASE_PACKAGES:append:https-support = ",ca-
> > certificates"
> >  DISTRO_VARS_PREFIX ?= "${@'HOST_' if
> > d.getVar('BOOTSTRAP_FOR_HOST') == '1' else ''}"
> 
> The idea is quite nice for us in Isar, though I'd also eventually
> like
> to improve Debian itself in this regard. The building blocks are
> there
> in upstream but they do not fully work.

Yes, definitely. I reported this on the debian-kernel ML as well:
https://lists.debian.org/debian-kernel/2023/12/msg00097.html

Anyways, are there more objections against this patch?

Felix

> 
> Jan
> 


  reply	other threads:[~2023-12-08 10:46 UTC|newest]

Thread overview: 8+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2023-12-04  2:36 Felix Moessbauer
2023-12-06  0:04 ` Jan Kiszka
2023-12-08 10:46   ` MOESSBAUER, Felix [this message]
2024-01-18 16:44     ` MOESSBAUER, Felix
2024-01-18 17:06       ` Jan Kiszka
2024-03-25 16:22         ` Gylstorff Quirin
2024-03-27  5:46           ` Uladzimir Bely
2024-04-02  6:47             ` MOESSBAUER, Felix

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=8ccae41f0016f01665a40ada6ab6051f1b7415dc.camel@siemens.com \
    --to=felix.moessbauer@siemens.com \
    --cc=baocheng.su@siemens.com \
    --cc=isar-users@googlegroups.com \
    --cc=jan.kiszka@siemens.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox