Since metadata server is accessed via dhcp namespace, dhcp relies on
conf.OVS.datapath_type for [0] logic to disable checksum offloading
that is not supported with ovs-dpdk, making metadata server not available.
[0] https://opendev.org/openstack/neutron/src/branch/stable/train/neutron/agent/linux/interface.py#L444-L446
Change-Id: I382af9d9e83b39fd9a616351e7cd5a752a603e77
Signed-off-by: Andrii Ostapenko <andrii.ostapenko@att.com>
Updated neutron to use an Nginx sidecar to terminate internal TLS rather
than using Apache with a separate RPC servers. Multiple RPC servers (in
sidecar) causes communication issues with RabbitMQ causing expected
errors.
Change-Id: Iaa6d3d64b730a54b1b85a338517bcb5be1842bda
Signed-off-by: Tin Lam <tin@irrational.io>
This commit enables user to update dpdk bond config
only when required using the flag UPDATE_DPDK_BOND_CONFIG.
This helps in avoiding disruption in ovs dpdk during ovs
agent restart or when there is a update in bond config.
Setting UPDATE_DPDK_BOND_CONFIG to true will have default
behavior.
Setting UPDATE_DPDK_BOND_CONFIG to false will disable
configuration of bonds after the first run and can be
set to true when required.
Change-Id: I4c8ec145c8f1c1c1b5f7a5201f792e040fdd89a2
This commit removes 'del-port' commands in neutron ovs
agent init script. Since, 'add-bond' command allows
modification of the existing options, the commands
being removed are not required. This also allows to
avoid disruptions in ovs caused by deleting ports on
restart.
Change-Id: I7201b87da7b20e1ca8efedf5d86a23123ccfa150
Passing config files from other agents result in
incorrent loading of extensions.
Value metadata_proxy_socket should be set in neutron.conf
Change-Id: If22168ccc77c918f7cfcb5d9d0d548f02d2a09e9
This patch set puts in the missing certificates to the test-pod.
It also corrects the path of sriov_agent.ini file
Change-Id: Ice2124f92a36d545726243fff60db25fbb2ea9c0
OVS_CTL file is required by other parts of the init
script.
This reverts commit a9693843d77525decefc8a0ba0528624f9e11350.
Change-Id: Ia11dc18e0b13d5fe01918a4c7febb82b19303527
The charts changes are required for deployment
of various clouds based on Tungsten Fabric SDN.
Right now it's tested for Airship-in-a-bottle.
The code cannot be tested currently in
OpenStack Helm project because of absence of
tests and platform for that.
This patchset doesn't have Heat-related changes,
they'll be added later.
Change-Id: I73f2ced2b09dbb93146334b59fe4571fa13dbfb0
Depends-On: https://review.opendev.org/#/c/734635/
The current copyright refers to a non-existent group
"openstack helm authors" with often out-of-date references that
are confusing when adding a new file to the repo.
This change removes all references to this copyright by the
non-existent group and any blank lines underneath.
Change-Id: Ia035037e000f1bf95202fc07b8cd1ad0fc019094
This patch adds ability to unhardcode readiness/
liveness probes timings for those probes which still
were hardcoded. Moreover it introduces
RPC_PROBE_TIMEOUT and RPC_PROBE_RETRIES variables
which are passed to health probe script and
allow to unhardcode RPCtest timeout and number of
retries
Change-Id: I2e48eed26abb82208a4ac4ae596d27ca8db99c90
There is a bug with the Intel i40e driver version 2.11.21 or earlier
where the interface VF spoof check must be toggle on and off after bond
is created (TODO: insert release note or bug here)
Change-Id: I9723e52fc87291f5e90df29a154c04180cbfe955
This reverts commit 09d6a3e68c3d95e27ce7a7395735d058062e9d17.
It causes failures during brown field upgrade.
Change-Id: Ie619358bd0ae046bfb94b9e9d811251b54d2897e
For OVS-DPDK deployments, where tunnel interface is bound to DPDK, there
should be support to transport the tunnel traffic over a VLAN network.
Change-Id: I1e63c9a6eb03a3f78a8592244d7c4b4928164fa5
Changed Nova and Neutron health-probe script to exit if previous
probe process is still running.
The health-probe has RPC call timeout of 60 seconds and has 2
retries. In worst case scenario the probe process can run a little
over 180 seconds. Changing the periodSeconds so that probe starts
after previous one is complete. Also changing timeoutSeconds value
a little to give little more extra time for the probe to finish.
Increasing the liveness probe periods as they are not do critical
which will reduce the resource usage for the probes.
Co-authored-by: Randeep Jalli <rj2083@att.com>
Change-Id: Ife1c381d663c1e271a5099bdc6d0dfefb00d8d73
Removes stale DHCP and L3 namespaces. The cron runs once in 12 hours.
Network namespace cleanup is implemented as a daemonset as Kubernetes
does not have a cronjob that works like daemonset-cronjob.
Network namespace cleanup should run on all nodes where DHCP and L3
agents run.
Change-Id: I7525e493067669026e0d57889a3e3238a2bd1308
The 'options' keyword for setting mtu in 'set interface' does not
set mtu and it seems to ignore/fail the request silently.
Change-Id: Icec98c5166611a8c538f93e6326cf7d20b545ecd
This ps update neutron ovs agent to support properly:
- setting mtu on dpdk bond and nic interface port
- setting vhost-iommu-support on dpdk bond and nic interface port
- setting n_txq values on dpdk bond and nic interface port
Change-Id: I422fa21a622642ecb7c49914fef04073e4f984bc
A recently introduced readiness probe for neutron-ovs-agent makes use of
an OVSDB table entry 'dpdk_initialized' which does not exist in OVS
versions preceeding v2.10.0. This patch changes the readiness probe to
exit successfully if this table entry does not exit. Thereby it does not
give any guarantees for older versions of OVS, but at least allows the
readiness probe to pass.
Change-Id: Ic77c6bdd60730c1a7c5e55fdb4afc6db938f0ddb
This patch set is one of many to migrate existing code/script to be
python-3 compatible as python-2 is sunsetting in January of 2020.
Change-Id: I337069203a3273e9aba6a37294ee3c25e5b4870a
Signed-off-by: Tin Lam <tin@irrational.io>
The current configuration expects VF & bonding info by default. This patch
set removes the need to configure them for every deployment.
Change-Id: Id546c113b2d3c42591a0326ee8cd442cccc73578
Python psutil library has not been consistent in behavior
a. gives trucated process names at times
b. the truncated names sometimes contain path to Python instead
of the program name Python runs
Change-Id: I99b77a4c28761a2187e59be4e562d5893ef3caa9
This PS updates the test project purge script to target specificly
the desired project by its id.
Change-Id: I54bfaa7727fdad781bdecc31251c1fe53f912c18
Signed-off-by: Pete Birley <pete@port.direct>
This PS updates the os purge of the test project to be an optional
operator driven choice, as they will also need to ensure
the project is unique to neutron testing.
Additionally this updates the purge image to be driven by the
charts values.yaml, as with every other image in OSH.
Change-Id: I46807f7c4922a1b411386641eddbd8957ab56f05
Signed-off-by: Pete Birley <pete@port.direct>
This commit adds readiness checks to neutron ovs agent
to check if the ovs and dpdk configurations are working
without errors.
Change-Id: I48277bdbd91ec8121e5fec300aeb646a80a65d29
Rally usually cleans up all its resources in normal executions - normal
test success cases and normal test failure cases. But the generic cleanup
does not work well for out of the system failures like process
interruptions, pod failures, disaster cleanup etc.
This is a known issue in rally-openstack. -
"Current generic mechanism is nice but it doesn't work enough well in real life.
And in cases of existing users, persistence context
and disaster cleanups it doesn't work well."
Hence, if we shall face above such issues, it is becoming impossible
to run "helm test neutron" again because of the stale data
and different quota limits mentioned in the values.yaml.
Hence we need to purge the stale data from the "test"
project as well as reset the quota limit for such scenarios.
For the normal executions, this patch has to do nothing,
but for unexpected failures, this patch will purge the stale data
from test project and reset the quota as defined in
values.yaml for the next run.
Change-Id: I3f6851582e2ac1aa1d375fcd13c07f4f57f45dc8
