Age | Commit message (Collapse) | Author | Files | Lines |
|
There is a new bug when deploying master/rocky where the OS of the
undercloud/overcloud is now upgraded to CentOS 7.6. When the undercloud
install runs it fails to configure the hostname using hostnamectl. This
is because systemd-hostnamed is not running and fails to start. Simply
reloading dbus seems to fix the issue. In the dbus logs there are odd
error messages like:
dbus-daemon[3230]: Unknown username "root" in message bus configuration
file
Disabling selinux seems to fix this. This patch also moves to use
podman instead of docker for container management and invokes a script
in Ansible which updates NIC templates as new variables are added
upstream. Furthermore, with the new patches for routed networks in OOO,
it is now required that the MTU is set in network-data, as well as
adding the External network to the Compute role. Now the External
network is removed by default from the Compute role.
Change-Id: Ie8b86f6f28d69bda11b1f7a430df882970ac3cb9
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
Previously if a service mapped to a network which was disabled, the
disabled network would fallback to using the ctlplane network. With the
recent change in THT, this is no longer the case:
https://review.openstack.org/#/c/614457/
With the above change, now any service pointing to a disabled network
now results in an empty string value being given for network variables.
This patch sets the service netmap in network-enviornment.yaml
appropriately based on which networks are enabled in apex.
Change-Id: Idf2919935aa707da6ca48968a04cf6653923d19d
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
|
|
Right now RDO defaults to an old version of OVS. In Apex for non-sfc
scenarios we were pulling a prebuilt RPM of OVS 2.9.0 that included a
group tables fix needed for ODL. For SFC, we now just build OVS along
with the kernel module from native OVS because NSH support exists
upstream in that version.
We can move to always doing this operation for all OVS based scenarios,
which will bump our OVS and ensure the package we build at runtime will
work with our kernel.
Change-Id: If38d0b858878f1af66990d987612a8f8f6fa2718
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
No point in using NFS when we only have one node.
Change-Id: Ic80f75913bf31c2e4a875146731d9c0d3483ae1b
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
Issues still persist where sometimes instances fail to start due to a
failure with os.utime to read the file path. This could be some bad race
condition between qemu/nova while copying images on the NFS. This patch
adds more ports to open in firewall, and changes initial directory owner
to nfsnobody.
Also, includes a patch to fix an apparent race condition when nova sends
a remote call to the privsep helper daemon to modify the time of the
base file owned by qemu:
https://review.openstack.org/#/c/625741/
Includes another fix for patching container images where the docker
image was not being detected correctly because the full gerrit project
name including 'openstack/' prefix was being used to search tripleo
docker images. Additionally, there were more bugs around patching
openstack python containers where the patch was not being applied
correctly.
JIRA: APEX-654
Change-Id: I1d011035486298d5906038922e69d478c383c3f7
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
There are problems with Nova launching instantces due to permissions
with nova being able to read/write certain directories on the NFS. The
permissions are right on the NFS and the folders the NFS mounts to, but
there still seems to be issues. The cause may be using a directory under
/root as the NFS mount. This patch moves the NFS mounts to be individual
folders under /. The patch also restart nova_compute docker container as
NFS problems still persist unless this is done.
JIRA: APEX-654
Change-Id: I25eee98c1a6516dfa44c686c2e614f6dc7000d98
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
There is an issue when nova user in the nova compute container tries to
write a file a in the nova instance path when using NFS. It looks to be
resolved by fixing permissions on the NFS directory.
JIRA: APEX-654
Change-Id: Ifbf3b17a8c13938fd1e7c2dabf3086322be82c69
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
Fixes a bug where oxygen was always being used even if a different ODL
version was specified in the deploy settings. Also updates the valid ODL
versions with fluorine and neon.
JIRA: APEX-648
Change-Id: Ie49c28d1757cd7e57802d67e48904ce04864feaf
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
In order to support HA as well as live migration a shared storage
backend needs to be used. This patch preps the overcloud image to start
with an NFS server sharing out of /root/nfs. If csit environment file is
used then NFS parameters will be passed to the deployment, configuring
Nova and Glance to use NFS hosted by controller0.
JIRA: APEX-654
Change-Id: I049d84bfd454013fc7ba5048939ee17100197624
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
When adding upstream packages to undercloud image, there is a conflict
between a new dependency python2-docker and already installed
python-docker. This patch removes the latter before trying to install
upstream packages.
Also, updates nic templates based on:
https://review.openstack.org/#/c/604943/
Change-Id: Ia7db0a3b37652d066bbe80478689f4d1ebaea560
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
|
|
|
|
Commit 1817e62 introduced a side effect for aarch64.
The undercloud and overcloud images are not the same for aarch64
due to differences in the partition table and they are not provided by RH
or tripleo. Helper scripts to build them are in this repository.
For aarch64 undercloud.qcow2,overcloud.tar and ironic-python-agent.tar
Are required.
Change-Id: Ia79de801e3cb33bbcc78cdc714cb1745c1b57d6b
Signed-off-by: Charalampos Kominos <Charalampos.Kominos@enea.com>
|
|
RDO builds packages which are aarch64 compatible but some configuration
is needed to succesfully deploy.
This change:
- Prepares the aarch64 docker.io repo as the source for Kolla Containers
- Configures VM sizing for aarch64 undercloud.
- Configures VM sizing for aarch64 virtual deploy targets.
Vms need to be larger on aarch64 compared to x86 to avoid
starvation of resources. (MYSQL)
- Configures vda2 as the location of the Linux Kernel in aarch64 in
an UEFI system
- Configures the vNICs to be on the pci-bus instead of the virtio-mmio
bus.This will enalbe the Nics to come up in the same order as the
x86 ones, so the extra configuration in ansible is not needed
- Configures apex to use a stable version of the ceph:daemon container
- Configure apex for containerized undercloud in Rocky
- Add extra ansible.cfg file for aarch64 which increases waiting
times in ansible for aarch64
- Provide helper scripts for DIB to create aarch64 UEFI images
Known limitations:
- Selinux is interfering with DHCP requests in ironic and ssh
so it must be disabled before the deploy command is ran.
- The aarch64 containers are frozen for in this commit:
https://trunk.rdoproject.org/centos7-rocky/f3/18/f3180de6439333a2813119ad4b00ef897fcd596f_70883030
- The 600s timeout defined in :
https://bugs.launchpad.net/tripleo/+bug/1789680 is not enough for
aarch64. A value of 1200s is recommended
JIRA: APEX-619
Change-Id: Ia3f067821e12bba44939bbf8c0e4676f2da70239
Signed-off-by: Charalampos Kominos <Charalampos.Kominos@enea.com>
Signed-off-by: ting wu <ting.wu@enea.com>
|
|
When deploying rocky, python2-tripleo-repos fails to install because of
a bad yum cache even after updating the repo. This patch just clears the
yum cache after updating the repo.
Change-Id: I94a68611faf1c3ff89d57c3d877d3336574ce8bb
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
OOO team is removing the undercloud disk image as it is no longer needed
for containerized undercloud deployments. Instead, we can just use the
overcloud image as the undercloud image.
Additionally, OOO team has recommended we use current-tripleo instead of
current-tripleo-rdo. current-tripleo-rdo was previously thought to be
more stable with more promotion checks, but now it seems that it is
older and current-tripleo now has the same stability/checks.
This patch also bumps the undercloud RAM from 8GB to 10GB. With the new
containerized undercloud there is more RAM consumption during
deployment.
Change-Id: I9e6bb2260dbe9f8796ee54d20527c0aad96476ec
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
Branch was being passed with the 'stable' prefix which was breaking the
docker URL.
Change-Id: I802492d407bc18ec2b47212edfe4bb4fde13002a
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
The neutron_driver for opendaylight needs to be set to 'odl' instead of
'opendaylight'
Change-Id: Ib9345d8b9fc8be5b7a49ac961a2499220c1c1bd6
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
The API has changed to create/upload the docker container images to be
used for deployment. In the past the prepare commands would read the THT
environment files passed, to determine which docker images to render
into an environment file. The new behavior uses a new
"containers-prepare-parameter.yaml" format (included in this patch),
which Apex will now configure for deployment. By default docker images
will be rendered for all TripleO services identified in the
roles_data.yaml file. Therefore we must use several excludes patterns to
only pull the docker images needed for a default deployment.
JIRA: APEX-642
Change-Id: Iab00fcb874554bb98540dc9a4c3051e58ea68a3b
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
This change allows downloading of files larger than 20GB.
Change-Id: I881ea35316e67980d4144e46bfbd41501e80f497
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
The delorean path does not have images for rocky, however they do exist
under the rdo_trunk path. Comparing the images for queens between the 2
paths looks the same, so switching to this URL.
Change-Id: I5d762af69b984c0fed1c79b73f2b4d61341298fa
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
|
|
Master code only supports containerized undercloud now, so this
migration is needed.
- Containerized services in undercloud
We can still apply patches to THT and other non-docker services, but
we will need to add support for patching openstack services on
undercloud.
Change-Id: I1ca4c6108f144efef7b5889503af265ef0fff8b2
Signed-off-by: Ricardo Noriega <rnoriega@redhat.com>
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
For now we need to retain the ability to deploy odl-queens for CSIT.
Change-Id: Idea550ff5e7aeb230aa93d3575de11e5e9656302
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
As of Queens only HA OVN deployments are supported.
Change-Id: I184c5a096fec9cbc3cf2ec06218700138ea3ed57
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
Change-Id: Ibfbd08dc2fa5fca95668fd0590707cfebd92099f
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
New arguments are added to allow snapshot deployment:
--snapshot, --snap-cache
The previous tripleo-quickstart code has been removed/replaced
with the snapshot option.
Snapshot deployments are supported on CentOS and Fedora, and snapshot
artifacts use a similar caching system as the standard deployment.
Snapshots are produced daily by Apex, and include latest as well as n-1
OpenStack versions. The os-odl-nofeature scenario is used for the
snapshots. Additionally multiple topology verions of Snapshots are
available. The Snapshot pulled at deploy time depends on the
deploy-settings and number of virtual-computes used at deploy time.
Since there is only one network used with snapshot deployments (admin),
there is no reason to pass in network settings for snapshot deployments.
That argument is now optional. Previously we required even in Standard
virtual deployments that the network settings be provided. However that
is also unnecessary, as we can default to the virtual network settings.
Includes minor fix to the tox.ini to allow specifying test cases
to run (useful for developers writing tests). Default behavior of
tox is unchanged.
JIRA: APEX-548
Change-Id: I1e08c4e54eac5aae99921f61ab7f69693ed12b47
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
Need this fix to make VLAN tenant networks
to work.
Change-Id: Ie2dea9f0145c576e1b36280e6340ed2cc0460179
Signed-off-by: Ricardo Noriega <rnoriega@redhat.com>
|
|
|
|
- This patch will install OVS 2.9.2 including
its kernel module which allows native NSH
headers.
- Fix Custom OVS due to bug:
https://bugzilla.redhat.com/show_bug.cgi?id=1544892
- Tacker is disable for the time being, tacker-conductor
needs to be enabled.
JIRA: APEX-630
Change-Id: Ia410309fd7053602ce78eae919839d0f57c9742a
Signed-off-by: Ricardo Noriega <rnoriega@redhat.com>
|
|
Adds mocking some file opens introduced by:
https://gerrit.opnfv.org/gerrit/#/c/60991/
Change-Id: I013b9c96a7c3f678c6e227f98cd58f9547d79ecb
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
|
|
JIRA: APEX-635
Change-Id: I488dd5d9b114aba5f4bb4591a0cdd8c19b692ce9
Signed-off-by: Ricardo Noriega <rnoriega@redhat.com>
|
|
- fix ipv6 enable command on bridge interfaces
- fix external subnet creation command
- enable NAT when admin is ipv4
- fix network settings file for ipv6 for br-ex interface type
Change-Id: Ibd2dbe1d0c8e8df72ad1b0b3713a19f8bbc56eef
Signed-off-by: Feng Pan <fpan@redhat.com>
|
|
The common patch design doesn't really work across branches very well.
This patch makes it so the common patches are specified in the same
file, but on a per branch basis. Making it easier to manage common
patches per branch.
Also, includes a fix and adds test coverage to the prep_image method
where we were referencing 'undercloud_admin_ip' before assignment in a
case where we were not using ODL and had patches to overcloud.
Change-Id: I7672947afd826fdc0042361a0139e22d6d5dd864
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
This patch adds allowing for common patches that should be applied to
every scenario to be included. It by default pulls in a file in the
deploy directory 'common-patches.yaml', but can optionally be
overridden.
This patch also includes a patch upstream to fix OSCLI not working
anymore due to breakage with the Cinder version in the overcloudrc.
Change-Id: I97b9efb937deff07e085b9ef75b9799fb65bfc57
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
This patch adds the ability to deploy all in one single nodes (Control
+ Compute). To enable this functionality do the following for each
deployment type:
- Baremetal: do not tag any nodes as compute in the inventory file
- Virtual: use argument '--virtual-computes 0'
JIRA: APEX-548
Change-Id: I22525c9eb21d331129c819449316c26a6fcf522d
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
We are seeing queens baremetal deployments are all failing in step 5 due
to timeout. Not sure why the sudden increase in deployment time, but we
can try increasing the deploy timeout to see if it fixes it.
Also includes a revert for a commit that broke OOO master:
https://bugs.launchpad.net/tripleo/+bug/1785872
Change-Id: I0bcfd5b3ca49b176f901c13b8b432fa29c899188
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
- Injection of Quagga tarball via overcloud builder.
- Extraction and installation of all related packages.
- It uses SDNVPN artifact repository to download Quagga
tarball, so there is only one source to test.
- Modifies bgpvpn scenario files to use OS master branch,
ODL master branch and containers.
JIRA: APEX-627
Change-Id: Icdbc2853d9531048e23fd6d5e444bd68208d18fc
Signed-off-by: Ricardo Noriega <rnoriega@redhat.com>
|
|
|
|
Change-Id: Ib3779733ac27183def6a59f3404cfbe9a3e9a211
Signed-off-by: Ricardo Noriega <rnoriega@redhat.com>
|
|
This bug fix is not present in the RDO versions of OVS:
https://bugzilla.redhat.com/show_bug.cgi?id=1544892
The impact of this is that traffic sent through a patch port (like from
br-int to br-ex) will not be sent correctly with OF groups (which ODL
uses). This would cause tenant vlan traffic not to work in our case.
Change-Id: Ia477b174118dcd2d7fb24189c2d093218eb97053
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
|
|
This patch removes the logic to use an specific tag for
Ceph containers. We will use whatever docker image TripleO
upstream uses. For aarch64, an ansible task will replace
the tag to pull the proper container image.
This patch also refactors the preparation of the local
registry. In Queens, there is no need to execute twice
the overcloud container image prepare command.
JIRA: APEX-622
Change-Id: I947d931609e58505675bb460a59d08c1d10d1d0b
Signed-off-by: Ricardo Noriega <rnoriega@redhat.com>
|
|
With deploying snapshots with a new ODL, we currently bring down the
docker container and bring up the tar.gz distro of ODL on the Overcloud
host itself (not rebuilding/using container). Therefore we need java
installed so that ODL can run on the host. In the future this may
change, but it works well and keeps things simple for now.
Additionally, there was a change upstream to make the opendaylight
container docker restart policy "unless-stopped" which means it will
no longer restart automatically when docker is stopped/started.
Therefore on first snapshot bring up (without the previously mentioned
ODL reinstallation) the container does not start, and snapshot
deployment fails. This patch includes a change to the restart policy to
always restart it.
Change-Id: Icc712ba147e578a28e371313154ae3190676f0dc
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
Introduces tenant vlan support through configuration of network settings file.
JIRA: APEX-272
Change-Id: Ie8e20a92ca4ebb9e0de4c53b5e03856d19ff660c
Signed-off-by: Feng Pan <fpan@redhat.com>
|
|
Recent changes upstream have removed the default 'admin' ODL password
and now password is randomly generated:
https://review.openstack.org/#/c/578505/
So in OPNFV we now store the password in overcloudrc as
SDN_CONTROLLER_PASSWORD variable.
Also includes minor fixes to unittests.
Change-Id: Iabe7e4f902442c80af99ba1603a3927cf13d0393
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
Also includes minor fix to patching where stable was not being removed
from the short hand branch name correctly.
Includes fix where path was changed in queens for env files:
https://review.openstack.org/#/c/567810/
Change-Id: Ia3a6769be70ea97c9bec9d5539c9ab0373fed30a
Signed-off-by: Tim Rozet <trozet@redhat.com>
|
|
|