parted supports the "-a optimal" option to get better alignment during partition table setup.
We should use that by default, since it:
a) shouldn't cause any harm on VMs anyway and
b) might improve performance especially on SSDs with non-standard
optimal_io_size settings (as observed on a Samsung EVA 850
with a setting of 33553920, so the partition should start
at 33.6MB and not 1MiB/2048s)
NOTE: fai-setup-storage doesn't support setting the parted options
like "-a optimal" yet, reported as #812712 in Debian's BTS.
Change-Id: Ib1c49d4f2b405697f0dcac5f164473ea9f87557c
NOTE: puppet success exit codes are different for different actions:
- 'puppet agent -t' exit code is '2' if no errors happen,
- 'puppet agent --enable' exit code is '0' if no errors happen.
Jenkins should detect puppet error exit codes and fail the test job.
Change-Id: I8df33897bd647e53a4f895c79c82749b97c7313c
The "--yes" workaround for LVM2 is needed for LVM versions
newer than >=2.02.106 to avoid a hanging prompt during
partition setup. Starting with FAI 5.0 the workaround
for this is included in FAI's setup-storage and therefore
we don't need our workaround any longer, see
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=750212
So ensure that fai-setup-storage is new enough, and only
if not then check the lvm2 package version and apply the
workaround as needed.
Fixes:
| Executing: lvcreate --yes --yes -n swap -L 1002.755859375 vg0
| (STDERR) Option -y/--yes may not be repeated.
Change-Id: Ic4445e0add5aab8da0ef87b77b08e23fc0d1c9ab
The latter is a transitional dummy package that will be going away in
the near future. Switch to it now, instead of waiting for the inevitable.
Also this way we get less cruft on the installed system.
Change-Id: I6679d3eafed07548dfebf0dc8a5b2fe75a5ef58d
Sipwise systems/VMs works well because we rebuild /etc/ssh/sshd_config
from templates and enable supporting /root/.ssh/sipwise_vagrant_key
while for plain VMs we do nothing, so we have to handle it here.
Change-Id: Ib7eccda1a8b9692da6e581aef47c3ccddeb229f6
When we install a fresh squeeze system we want to avoid the
occasional "Hash Sum mismatch" errors during apt-get runs,
so let's not only cover that via upgrades but also in the
initial installation.
Also cover it for Debian lenny, since it very probably might
be affected as well (unverified though).
Change-Id: I01dbeec5ef2a727e5749be5db3ebea47818ed3b5
With the according SSL setup in place nothing should prevent
us to enable HTTPS by default.
The only issues left here are the usage of https for
our internal management web service on port 3000 and
the usage of HTTPS with approx.
Change-Id: I7c9c3d947c6b880b9145ca83a31c832049c6573a
puppet in jessie depends on ca-certificates/openssl which
cant' be removed therefore. We actually want to have
ca-certificates/openssl on all our systems anyway and
nowadays debootstrap doesn't install tcpd and xauth any longer,
so we can safely skip those package removals for jessie and newer.
The templatedir doesn't need to be set in recent puppet versions
and older versions use a sane default anyway, therefore dropping it.
puppet as present in jessie (and newer) expects 'puppet agent --enable'
to be executed to actually be enabled.
Change-Id: Iae70eaeec6a56bef0e67171aab74482aebb24181
Since lvm2 v2.02.106 its udev rule creates a
/dev/disk/by-id/lvm-pv-uuid-<PV_UUID> symlink for PVs:
https://git.fedorahosted.org/cgit/lvm2.git/commit/?id=417e52c13a8156b11c25c411d44bda8b32bf87e4
When debootstrapping Debian/wheezy to a virtio disk from a system
using lvm2 >=v2.02.106 (as present in Grml >=2014.11) this causes
the resulting GRUB installation to look like that
(`/media/ngcp-root` is the mounted logical volume with the rootfs
which also includes /boot):
| root@grml ~ # cat /media/ngcp-root/boot/grub/device.map
| (hd0) /dev/disk/by-id/lvm-pv-uuid-RNZHr8-hLlc-nles-X9Gs-RYSp-Mqo7-uZTk6N
| (hd1) /dev/vda
and failing to boot with:
> error: no such disk
> Entering rescue mode...
> grub rescue>
When manually removing the lvm-pv-uuid* symlink from the
Debian/jessie system before installing GRUB it works fine:
| root@grml ~ # rm /dev/disk/by-id/lvm-pv-uuid-RNZHr8-hLlc-nles-X9Gs-RYSp-Mqo7-uZTk6N
| root@grml ~ # ls /dev/disk/by-id
| ata-QEMU_DVD-ROM_QM00003@ dm-name-ngcp-swap@ dm-uuid-LVM-lQhnvLyYmhESObLrZcnJs4FTZq9usVDzRex3ImRtrmqhpDH9xKBj22St82lHDADh@
| dm-name-ngcp-root@ dm-uuid-LVM-lQhnvLyYmhESObLrZcnJs4FTZq9usVDzrcMaDfk8O0RionEYRxdjnYxsY6eb0Esi@
| root@grml ~ # grml-chroot /media/ngcp-root/ grub-install --recheck /dev/vda # `grml-chroot` makes sure /proc, /sysfs and /dev are properly mounted in the chroot
| Writing /etc/debian_chroot ...
| Installation finished. No error reported.
| root@grml ~ # cat /media/ngcp-root/boot/grub/device.map
| (hd0) /dev/vda
NOTE: This problem seems to exist only for virtio disks, when
using a disk on the SATA bus the resulting device.map of the
Debian/wheezy system looks like that and boots fine:
| root@grml ~ # cat /media/vg0-root/boot/grub/device.map
| (hd0) /dev/disk/by-id/ata-QEMU_HARDDISK_QM00005
| (hd1) /dev/disk/by-id/lvm-pv-uuid-lJJr4E-M96R-3eIS-b6YO-5usn-bH5o-osQV2B
As a result check for Debian wheezy (or older) and if the disk is
a virtio one (AKA /dev/vda), if so then remove the symlink of the
live system before installing Debian incl. GRUB via
grml-debootstrap.
Reported as https://bugs.debian.org/782998 in Debian's BTS.
Change-Id: I4b2cf9e39c88521ab7eac7799531508e658cc30d
During the refactoring in 43c9f159 we missed the point that
/etc/debootstrap/pre-scripts/install-sipwise-key.sh should be
executed on CE install CD too even if we do NOT download key.
Otherwise we are getting:
> Reading package lists...
> W: GPG error: http://debian.sipwise.com wheezy Release: The following signatures couldn't be verified because the public key is not available: NO_PUBKEY 4EB7765DA42C4F2A
> W: GPG error: http://debian.sipwise.com wheezy-security Release: The following signatures couldn't be verified because the public key is not available: NO_PUBKEY 4EB7765DA42C4F2A
> W: GPG error: http://debian.sipwise.com wheezy-updates Release: The following signatures couldn't be verified because the public key is not available: NO_PUBKEY 4EB7765DA42C4F2A
and CE installation failed as unauthorized packages cannot be installed.
Change-Id: I3ccc8b8801ba2be829c4eea84507f11c841b7658
It allows to test scenarios where ngcp-installer is comming from one PPA
while other packages (like ngcp-templates) are comming from another PPA.
By default NGCP_PPA_INSTALLER is equal to NGCP_PPA,
so we still can define one boot option only:
> ngcpppa=gerrit_MT12345_review777
while for some cases it is useful to define:
> ngcpppa=gerrit_MT12345_review777 ngcpppainstaller=gerrit_MT54321_review778
Change-Id: Icf2fee70db1beaae08e9d9c3eba87c5a836661b7
Also removed sipwise user from 'plain' VM boxes,
it is not necessary there as we use user root in Vagrant.
Change-Id: I84c8ab6b1cd6e9dd7ba94f6871b79bafb153bd4e
Also, we did s/RETRIEVE_MGMT_CONFIG/CARRIER_EDITION/g because /etc/hosts
and /etc/network/interfaces are fully configured for Carrier by ngcp-installer.
So, no need to touch them for Carrier at all now
(see commits to installer.git through this ticket).
Change-Id: Ie16038d1f6c194564bb91c52faab75e86e3a0c11
1) Previously we were installing keys everywhere,
this produces a mess, so lets install them once on start.
2) GRML bypass /etc/apt/trusted.gpg.d/sipwise.gpg to NGCP CHROOT via install-sipwise-key.sh.
So we have /etc/apt/trusted.gpg.d/sipwise.gpg on every installed VM for wheezy/squeezy/jessie.
3) /var/log/deployment-installer-debug.log on all plain and NGCP VMs contains:
> echo "Sipwise Debian mirror key is already present."
So, we can safetely clean it from vagrant_configuration, IMHO.
Change-Id: I873f87faed24df4a93474cf17f41eb19e8a14ab1
We already have a function which take care about the keys,
we should use it everywhere!
Moreover the public keys contains both release and autobuild keys already:
>> http://deb.sipwise.com/spce/sipwise.gpg
>> http://deb.sipwise.com/sppro/sipwise.gpg
> gpg sipwise.gpg
>> pub 1024D/A42C4F2A 2010-11-29 Sipwise GmbH (Sipwise repository key) <support@sipwise.com>
>> sub 2048g/680FBA8A 2010-11-29
>> pub 4096R/EE5E097D 2011-06-06 Sipwise autobuilder (Used to sign packages for autobuild) <development@sipwise.com>
>> sub 4096R/3A5C488C 2011-06-06
Change-Id: I8b729ff624954de0d133b803e4d220c29a4fd799
It's PITA to install an internal system which depends on vlan
tools being present to get network acess but not having vlan
tools installed. So far we installed the vlan plus related tools
only when the vlan boot option was present during deployment
time. This might be relevant also for our ngcp users and it
should be safe to always include those tools. So we decided to
always include vlan (+ related) tools on all our installations.
Change-Id: I17bc5b9aba967d13bebe2db631e9167e693b09bc
We have migrated to ngcp.inc with modern list of services to be stopped
at the end of ngcp-installer.
See commit e75b01441c for more details.
Unfortunately deployment.sh is performing 'magic' with network.yml and depends on glusterfs
for pulling/pushing configs from/to shared storage. I am planning to move this 'magic' to
ngcp-installer, but while it is here, we have to start stop glusterfs-server manually.
Change-Id: Ia0aac183df52a1e2bd51ff6f2326f8438cad5b79
Michael Prokop
Alexander Lutay
Sergii Kipot
Guillem Jover
Jenkins User
Victor Seva