diskimage-builder/elements/yum-minimal/root.d/08-yum-chroot
Ian Wienand 6f622fb7c9 yum-minimal: clear our rpm/dnf/yum data in chroot
I have seem some occasional odd failures coming from the "dnf -y
update" done by elements/base/install.d/00-up-to-date.

dnf seems to sometimes think a package is not installed when it really
is.  It then seems to try and re-install them, but notices they are
installed, and then bails with a failure exit [1].  The packages that
seem to cause this vary, but the common thread is that they seem to
have all been installed during the initial phase of installing the
package manager in the chroot.

I suspect that when we are building the chroot, we do our initial
install with the "external" yum & rpm.  Then we start using the
dnf/yum in the chroot, but we're actually using meta-data created by
the *external* tools -- which could be vastly different versions or
who-knows-what.  While I honestly I don't have an exact root cause,
empirically I've found rebuilding the rpm db always seems to fix
things up.

So this change takes care to rebuild the rpm db with the chroot
version of rpm, and clear out the package metadata for a refresh with
"update".  This should hopefully put us in a consistent state.

[1] http://paste.openstack.org/show/487356/

Change-Id: I565df23897ae511356c4861fdbe63823fa6b6ff9
2016-03-08 14:53:52 +11:00

225 lines
8.8 KiB
Bash
Executable File

#!/bin/bash
#
# Copyright 2015 Hewlett-Packard Development Company, L.P.
#
# Licensed under the Apache License, Version 2.0 (the "License"); you may
# not use this file except in compliance with the License. You may obtain
# a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
# License for the specific language governing permissions and limitations
# under the License.
#
if [ "${DIB_DEBUG_TRACE:-0}" -gt 0 ]; then
set -x
fi
set -eu
set -o pipefail
if [ -f ${TARGET_ROOT}/.extra_settings ] ; then
. ${TARGET_ROOT}/.extra_settings
fi
ARCH=${ARCH:-x86_64}
if [ $ARCH = amd64 ]; then
ARCH=x86_64
fi
# Calling elements will need to set DISTRO_NAME and DIB_RELEASE
DIB_YUMCHROOT_EXTRA_ARGS=${DIB_YUMCHROOT_EXTRA_ARGS:-}
YUMCHROOT_TARBALL=$DIB_IMAGE_CACHE/yumchroot-${DISTRO_NAME}-${DIB_RELEASE}-${ARCH}.tar.gz
# TODO Maybe deal with DIB_DISTRIBUTION_MIRROR
http_proxy=${http_proxy:-}
YUM=${YUM:-yum}
WORKING=$(mktemp --tmpdir=${TMP_DIR:-/tmp} -d)
EACTION="rm -r $WORKING"
trap "$EACTION" EXIT
YUM_CACHE=$DIB_IMAGE_CACHE/yum
mkdir -p $YUM_CACHE
# Note, on Debian/Ubuntu, %_dbpath is set in the RPM macros as
# ${HOME}/.rpmdb/ -- this makes sense as RPM isn't the system
# packager. This path is relative to the "--root" argument
_RPM="rpm --dbpath=/var/lib/rpm"
# install the [fedora|centos]-[release|repo] packages inside the
# chroot, which are needed to bootstrap yum/dnf
#
# note this runs outside the chroot, where we're assuming the platform
# has yum/yumdownloader
function _install_repos {
local packages
# pre-install the base system packages via rpm. We previously
# just left it up to yum to drag these in when we "yum install
# yum" in the chroot in _install_pkg_manager. This raised a small
# problem that inside the empty chroot yum went ahead and did a
# mkdir for /var/run to put some pid file in, which then messed up
# the "filesystem" package making /var/run a symlink to /run
# ... which leads to odd issues with a running system.
#
# TODO: these packages still have some small %posttrans stuff that
# depends on other packages (see rhbz#1306489) ... maybe the idea
# is that they are only installed in one big transaction with the
# rest of the system? but we don't want to use yum to do this
# (see above) so ...
packages="basesystem filesystem setup "
packages+="${DISTRO_NAME}-release "
# after fedora21, this is split into into a separate -repos
# package
if [ $DISTRO_NAME = fedora ]; then
packages+="${DISTRO_NAME}-repos "
fi
yumdownloader \
--releasever=$DIB_RELEASE \
--setopt=reposdir=$TMP_HOOKS_PATH/yum.repos.d \
--destdir=$WORKING \
${packages}
# --nodeps works around these wanting /bin/sh in some fedora
# releases, see rhbz#1265873
sudo $_RPM --root $TARGET_ROOT --nodeps -ivh $WORKING/*rpm
}
# _install_pkg_manager packages...
#
# install the package manager packages. This is done outside the chroot
# and with yum from the build system.
# TODO: one day build systems will be dnf only, but we don't handle
# that right now
function _install_pkg_manager {
# Install into the chroot, using the gpg keys from the release
# rpm's installed in the chroot
sudo sed -i "s,/etc/pki/rpm-gpg,$TARGET_ROOT/etc/pki/rpm-gpg,g" \
$TARGET_ROOT/etc/yum.repos.d/*repo
# See notes on $_RPM variable -- we need to override the
# $HOME-based dbpath set on debian/ubuntu here. Unfortunately,
# yum does not have a way to override rpm macros from the command
# line. So we modify the user's ~/.rpmmacros to set %_dbpath back
# to "/var/lib/rpm" (note, this is taken relative to the
# --installroot).
#
# Also note, we only want this done around this call -- this is
# the only place we are using yum outside the chroot, and hence
# picking up the base-system's default rpm macros. For example,
# the yumdownloader calls above in _install_repos want to use
# ~/.rpmdb/ ... there is nothing in the build-system /var/lib/rpm!
#
# Another issue we hit is having to set --releasedir here. yum
# determines $releasevar based on (more or less) "rpm -q
# --whatprovides $distroverpkg". By default, this is
# "redhat-release" (fedora-release provides redhat-release) but
# some platforms like CentOS override it in /etc/yum.conf (to
# centos-release in their case). You can't override this (see
# [1]), but setting --releasever works around this.
#
# [1] https://bugzilla.redhat.com/show_bug.cgi?id=1287333
(
flock -w 1200 9 || die "Can not lock .rpmmacros"
echo "%_dbpath /var/lib/rpm" >> $HOME/.rpmmacros
sudo -E yum -y \
--setopt=cachedir=$YUM_CACHE/$ARCH/$DIB_RELEASE \
--setopt=reposdir=$TARGET_ROOT/etc/yum.repos.d \
--releasever=$DIB_RELEASE \
--installroot $TARGET_ROOT \
install $@
sed -i '$ d' $HOME/.rpmmacros
) 9>$HOME/.rpmmacros.dib.lock
rm $HOME/.rpmmacros.dib.lock
# Set gpg path back because subsequent actions will take place in
# the chroot
sudo sed -i "s,$TARGET_ROOT/etc/pki/rpm-gpg,/etc/pki/rpm-gpg,g" \
$TARGET_ROOT/etc/yum.repos.d/*repo
}
if [ -n "$DIB_OFFLINE" -o -n "${DIB_YUMCHROOT_USE_CACHE:-}" ] && [ -f $YUMCHROOT_TARBALL ] ; then
echo $YUMCHROOT_TARBALL found in cache. Using.
sudo tar -C $TARGET_ROOT --numeric-owner -xzf $YUMCHROOT_TARBALL
else
# Note this is not usually done for root.d elements (see
# lib/common-functions:mount_proc_dev_sys) but it's important that
# we have things like /dev/urandom around inside the chroot for
# the rpm [pre|post]inst scripts within the packages.
sudo mkdir -p $TARGET_ROOT/proc $TARGET_ROOT/dev $TARGET_ROOT/sys
sudo mount -t proc none $TARGET_ROOT/proc
sudo mount --bind /dev $TARGET_ROOT/dev
sudo mount --bind /dev/pts $TARGET_ROOT/dev/pts
sudo mount -t sysfs none $TARGET_ROOT/sys
# initalize rpmdb
sudo mkdir -p $TARGET_ROOT/var/lib/rpm
sudo $_RPM --root $TARGET_ROOT --initdb
# this makes sure that running yum/dnf in the chroot it can get
# out to download stuff
sudo mkdir $TARGET_ROOT/etc
sudo cp /etc/resolv.conf $TARGET_ROOT/etc/resolv.conf
# Bind mount the external yum cache inside the chroot. Same logic
# as in the yum element to provide for yum caching copied here
# because the sequencing is wrong otherwise
sudo mkdir -p $TMP_MOUNT_PATH/tmp/yum
sudo mount --bind $YUM_CACHE $TMP_MOUNT_PATH/tmp/yum
_install_repos
if [ $DIB_RELEASE -ge 22 ]; then
# install dnf for >= f22
_install_pkg_manager dnf dnf-plugins-core yum
else
_install_pkg_manager yum
fi
# we just installed yum/dnf with "outside" tools (yum/rpm) which
# might have created /var/lib/[yum|rpm] (etc) that are slighlty
# incompatible. Refresh everything with the in-chroot tools
sudo -E chroot $TARGET_ROOT rpm --rebuilddb
sudo -E chroot $TARGET_ROOT ${YUM} clean all
# bootstrap the environment within the chroot; bring in new
# metadata with an update and install some base packages we need.
sudo -E chroot $TARGET_ROOT ${YUM} -y update
sudo -E chroot $TARGET_ROOT ${YUM} -y \
--setopt=cachedir=/tmp/yum/$ARCH/$DIB_RELEASE \
install passwd findutils sudo util-linux-ng
# Put in a dummy /etc/resolv.conf over the temporary one we used
# to bootstrap. systemd has a bug/feature [1] that it will assume
# you want systemd-networkd as the network manager and create a
# broken symlink to /run/... if the base image doesn't have one.
# This broken link confuses things like dhclient.
# [1] https://bugzilla.redhat.com/show_bug.cgi?id=1197204
echo -e "# This file intentionally left blank\n" | \
sudo tee $TARGET_ROOT/etc/resolv.conf
# cleanup
# TODO : move this into a exit trap; and reconsider how
# this integrates with the global exit cleanup path.
sudo umount $TMP_MOUNT_PATH/tmp/yum
sudo umount $TARGET_ROOT/proc
sudo umount $TARGET_ROOT/dev/pts
sudo umount $TARGET_ROOT/dev
sudo umount $TARGET_ROOT/sys
# RPM doesn't know whether files have been changed since install
# At this point though, we know for certain that we have changed no
# config files, so anything marked .rpmnew is just a bug.
for newfile in $(sudo find $TARGET_ROOT -type f -name '*rpmnew') ; do
sudo mv $newfile $(echo $newfile | sed 's/.rpmnew$//')
done
echo Caching result in $YUMCHROOT_TARBALL
sudo tar --numeric-owner \
-C $TARGET_ROOT \
-zcf $YUMCHROOT_TARBALL --exclude='./tmp/*' .
fi
sudo rm -f ${TARGET_ROOT}/.extra_settings