Age | Commit message (Collapse) | Author | Files | Lines |
|
This patch creates a new installer tree in xci that uses
openstack-helm to deploy openstack on a kubernetes cluster.
USAGE: Export INSTALLER_TYPE=osh, DEPLOY_SCENARIO=k8-calico-nofeature
and XCI_FLAVOR=noha or mini and run xci-deploy.sh as in documentation.
deploy-scenario:k8-calico-nofeature
installer-type:osh
Change-Id: I212f70eb51c2a38c798c11367d2ebb8bf5f4a1de
Signed-off-by: nikoskarandreas <nick@intracom-telecom.com>
|
|
When working with an interface that uses vlan, the vlan is created when
executing the ifup command. As we previously execute the command:
/sbin/ip addr flush dev {{ item }}
it will not return code 0 because the interface does not exist yet.
Consequently, the second part of the command (ifup) is never executed
and the new network config is never triggered.
This patch removes the requirement to get a rc=0 in the first command to
run the second one.
Note that /sbin/ip addr flush dev {{ item }} must be run before ifup
when working with non-vlan interfaces that already have an ip, otherwise
ifup fails in Ubuntu
Change-Id: Ieac273a49b78104f24d4c1ba81c263b450a40a9f
Signed-off-by: Manuel Buil <mbuil@suse.com>
|
|
The IDF files contain DNS information so we should respect that when
we configure the various XCI nodes. The DNS information is also a
list instead of a string so treat it as such.
Change-Id: I1c4d5eb600baaca35b2838dcafa7a75e59bf6783
Signed-off-by: Markos Chandras <mchandras@suse.de>
|
|
When deploying baremetal, the traffic should be segmented using the
different interfaces instead of through vlans. All the config is done
based on idf and pdf information.
When doing non-baremetal, opnfv and nodes get the same config.
When doing baremetal, opnfv and nodes get a different network config
Apart from that, if vlan_id is defined in the name, there is no need
for VLAN_ID in the interface descriptor. This simplifies things
Change-Id: Iddbb90af807b43e247e5ee11fe735df9e823d4bf
Signed-off-by: Manuel Buil <mbuil@suse.com>
|
|
When deploying baremetal, the traffic should be segmented using the
different interfaces instead of through vlans. All the config is done
based on idf and pdf information.
When doing non-baremetal, opnfv and nodes get the same config. When
doing baremetal, opnfv and nodes get a different network config
Change-Id: I23aa576bc782c7c69d511a5558827110c37b558a
Signed-off-by: Manuel Buil <mbuil@suse.com>
|
|
According to OpenStack admin is the network for pxe boot and mgmt is the
network for OpenStack services to communicate. We were using both in XCI
indistinctly
Change-Id: I3959e767098ac2be7161a5e84735fde9ab129784
Signed-off-by: Manuel Buil <mbuil@suse.com>
|
|
We are configuring static IPs in the various nodes but we don't do
anything for DNS assuming that DNS is being configured by another
entity. However, the IDF file already contains DNS information for us
so we should use that instead. Moreover, we update the IDF file to use
the gateway as DNS instead of the Google one in order to make it more
usable on restricted networks.
Change-Id: Ieba58ec9558080a1296e204c4f99bae859e9daef
Signed-off-by: Markos Chandras <mchandras@suse.de>
|
|
|
|
The bootstrap role configures NTP and networking on hosts so we
should use it on k8s deployments as well.
installer-type:kubespray
deploy-scenario:k8-nosdn-nofeature
Change-Id: I04bd1e1c2c325baabfb836bd8cca60c5f59344c7
Signed-off-by: Markos Chandras <mchandras@suse.de>
|
|
The IDF file contains the netmask for every network so we should use
that information instead of using hardcoded values.
Change-Id: Ie798cb49563bdb72fdfb7b6e9e269692bf1f7bc9
Signed-off-by: Markos Chandras <mchandras@suse.de>
|
|
The PDF and IDF files contain all the information we need for the
virtual XCI deployment, so we can use it to create a dynamic inventory
and get rid of all the static ones which could easily get outdated
as PDF and IDF files evolve over time.
This inital version of the dynamic inventory contains a lot of
unnecessary generated information but we do that in order to ease
the migration from static files to the dynamic inventory. The dynamic
inventory will be improved in the future as we consume more and more
information from the PDF and IDF files.
Change-Id: Id9f07a61c67a5cffcbc18079a341e5d395020a27
Signed-off-by: Markos Chandras <mchandras@suse.de>
|
|
We split the networking task on distro specific files to make it
easier to read. Moreover, the debian network configuration has been
improved by simply sharing a common file across all nodes and also
use the 'source' facility in the main /etc/network/interfaces file
to use one configuration file per interface.
Change-Id: Ic822fe6dc197227e70c0ba7cee812629df287d82
Signed-off-by: Markos Chandras <mchandras@suse.de>
|
|
The jinja templates that are used for networking setup are based on the
openstack-ansible needs, those needs can differ for another installers.
This change propose to make the network configuration depending on the
installer.
Signed-off-by: Victor Morales <victor.morales@intel.com>
Change-Id: Ie805c3c7716393377d4dfcb32ed794cc1039d515
|
|
When we run XCI for the first time, Ansible picks the first active
interface as the default one. However, after we configure all the XCI
bridges etc, and we try to run this role again, Ansible may have changed
its mind about what interface is active and it could default to one of
the bridges. This forces the role to redo the network configuration but
this time the bridges are being attached to bridges so everything goes
terribly wrong after that. The way to solve this would be to add a local
fact about what interface should be considered as the 'real' default one
so subsequent calls to this role to not destroy the network.
This also drops the task which removed the network configuration files
on SUSE platforms since Ansible is smart enough to not touch them if
they are configured properly.
Change-Id: Ic0525e934b1934a40d69e6cf977615ab9b3dac6d
Signed-off-by: Markos Chandras <mchandras@suse.de>
|
|
In preparation for adding support for the 'ansible-lint' tool we fix
various problems in our playbooks to make the tool happy before we make
it mandatory.
Some of the problems that are fixed here are
- [ANSIBLE0011] All tasks should be named
- [ANSIBLE0012] Commands should not change things if nothing needs doing
- [ANSIBLE0013] Use shell only when shell functionality is required
- [ANSIBLE0010] Package installs should not use latest
installer-type:osa
deploy-scenario:os-nosdn-nofeature
Change-Id: I66c759d3932a414b81b2846393d2d98ce80c0b6d
Signed-off-by: Markos Chandras <mchandras@suse.de>
|
|
In some cases the XCI development environment can be located behind a
corporate proxy resulting in a additional layer to consider to
configure. These changes pretend to include proxy support for all
linux distros in all the posible flavors.
Change-Id: Iab469268809ac471d09e244bb3ccd83de1a41b88
Signed-off-by: Victor Morales <victor.morales@intel.com>
|
|
Those roles were too small and only used as part of the host
bootstrapping process. As such, we merge them into a common
'bootstrap-host' role which can be used to prepare the hosts
after the initial deployment
Change-Id: Ifc84cf40b98ced91b31aae699dc28e9642380550
Signed-off-by: Markos Chandras <mchandras@suse.de>
|