Linux bridges are automatically set UP when
any parent interface is UP.
But for OVS bridges it doesn't work.
For dpdk and non-dpdk bridges, always create a
config file in /etc/network/interfaces.u/ and
bring the bridge interface up after it is configured,
even if it doesn't have IP address.
Change-Id: I92888ce0d373e412dfb7ed2e2398c0d4d008e301
Closes-Bug: https://mirantis.jira.com/browse/PROD-24343
Issue description:
PXE interfaces, which are used by salt should not be restarted
during salt calls, otherwise communication between salt master and
salt minion would be interrupted.
Therefore it is possible to specify "noifupdown: True" in pillars
for this interface or group of interfaces, which are used for PXE
network.
This pillar structure will remain until one removes it manualy.
It is not possible to remove it during deploy and enforce network
state without touching the model.
It is possible to override pillars from CLI like:
# salt ctl01* state.apply linux.network.interface \
pillar='{"linux":{"network":{"interface":{"ens3":{"noifupdown":True}}}}}'
However it is not easy/possible to predict all interfaces for PXE
network.
Solution:
Provide global noifupdown pillar value check.
If it exists, noifupdown will take effect and not otherwise.
So our deployment would have next steps:
- Execute: linux.network.interface pillar='{"linux":{"network":{"noifupdown":True}}}'
- Reboot node to enable kernel params like hugepages etc.
- Execute: linux.network.interface with no params to ensure PXE.
Pipelines may pass this parameter to control noifupdown behavior.
Change-Id: I8863f972c7805e4bf4f9e104d6c0ddf055c39cb1
Previous implementation was not able to add port 'dpdk0' to bridge
'br-dpdk0' since both matches 'grep' condition. To fix this we need to
look for port in a particular bridge
Change-Id: Ie83cebc3ab73c45a48f68fae2d6f474743215908
* Add TODO-proper fix for state - native salt fun.
But due bug[1] in saltstack - we can't enable
proper solution now
[1] 74599bbdfc
Related-PROD: PROD-20730
Change-Id: I11b6d81ae0f9a7864518f638e8fc423e4e087285
Wait for dpdk bond interfaces to come up.
linux.network.dpdk state fails to update a port within for loop
when this port does not exist yet.
Dependency will require interfaces to be added before
Prod-Related: PROD-19696
Closes-Bug: PROD-19696
Change-Id: Ia83218a76dd6e86664e7f9498a76341717eb5b80
The patch adds ability to configure variety logging options
for OVS. In order to configure OVS log options the below pillar
sctructure should be used:
linux:
network:
openvswitch:
enabled: true
logging:
enabled: true
ovsdb:
console: emer
syslog: err
file: info
facility: local0
vswitchd:
console: emer
syslog: err
file: info
facility: local1
Change-Id: I59ef0636447a974215d872259a26beb858495cfb
Related-PROD: PROD-19980
The patch makes IP address flush for external network
interface where IP address is assigned automatically
however an IP address from the same IP range is hardcoded in
the model.
Change-Id: I4220a635a96a031ad74dca1034c917e7d87d4b11
Related-PROD: PROD-19504
Issue: First time you configure dpdk ovs switch it would stuck on
answering salt-minion because kernel options, such as
intel_iommu,iommu,isolcpus, are not set and ovs would
exhaust all hugepages and fail to apply options on the fly.
Fix: Configure ovs switch without waiting for an answer and
reboot the node afterall as we do this all the time before
starting automated pipeline.
Change-Id: Ica27a6cc47312bcc0762cddde049a0abf771f9fb
Previously there were no dependency and as result we tried to add port to
non existent bridge.
Change-Id: I69ad6a6faecf399185a72650e8dbeb019b6fbc05
Related-Prod: PROD-18112
By default, the alternative installed by `openvswitch-switch`
packages shadows (higher prio) `openvswitch-switch-dpdk` version,
e.g. for UCA Pike packages:
$ update-alternatives --query ovs-vswitchd
Alternative: /usr/lib/openvswitch-switch-dpdk/ovs-vswitchd-dpdk
Priority: 50
Alternative: /usr/lib/openvswitch-switch/ovs-vswitchd
Priority: 100
To avoid confusion, when DPDK is enabled and ovs-vswitchd-dpdk
should be used, remove the alternative for classic ovs-vswitchd
(which can easily be re-eneabled later if needed).
Signed-off-by: Alexandru Avadanii <Alexandru.Avadanii@enea.com>
When resolv.conf file is managed by salt, resolvconf service should be disabled
because it overwriting resolv.conf after reboot.
Change-Id: I7ec85872ddbd5f90232d5d41a8171169c779a1c5
Previously, setting up routes did not allow passing 'require_reboot',
so each route change would lead to a networking service restart,
rendering interface configuration options like 'noifupdown' useless.
Allow disabling network restart per-interface using the existing
'noifupdown' option.
Signed-off-by: Alexandru Avadanii <Alexandru.Avadanii@enea.com>
Fixed:
* The udev-rules template is not tested
* Wrong unicode character in the template leads to udev ignoring
the rule completely
* The template is unable to be rendered due to absent import
* udev is not retrigerred with new rules
Change-Id: I134b5e49b883afcc5e34feaaa561d7ca70192796
Closes-Bug: PROD-16649
Currently when OVS-DPDK is enabled, instances fail to spawn
due to permissions mismatch, see nova bug for details:
https://bugs.launchpad.net/nova/+bug/1670950
This patch configures OVS to use a separate dir, created by
nova formula [1].
[1] https://gerrit.mcp.mirantis.net/11213
Related-PROD: PROD-14413
Change-Id: Ia86658b0967b4d35f063c7f6f53e26e78a1fbd39
Current logic does not apply n_rxq, pmd_rxq_affinity and mtu
for a dpdk port in case it is attached directly to a bridge
(not bonded).
This patch makes formula apply parameters in both cases.
Change-Id: I5b19999b1ac8a80d1e31b266fede7dad0d7fd0d4
This workaround is required until salt with the patch [1] is not
used.
- use 'ifenslave' tool to enslave necessary interfaces only if bond
interface has been changed
- install package that provides 'ifenslave' tool
[1] https://github.com/saltstack/salt/pull/39912
Change-Id: I65b607f26cf7319efb60f154951855d1334e1640
Non-destructive patch, won't affect anything because this
option is not used in any model yet.
Official guide for network.managed state provides an example:
bond0:
network.managed:
- type: bond
- slaves: eth2 eth3
- require:
- network: eth2
- network: eth3
...
without parameter: '- use:'
* add a new option 'require_interfaces' that allows to set
'- require:' not only for 'bridge' interfaces
* using 'require_interfaces' allows to fix the order for states
that create bonds (do not try to create bond0 before slave
interfaces are configured!)
* using 'require_interfaces' allows to fix the order for states
that create VLANs (do not try to create tagged interfaces before
parent interface is created, it locks the parent interface
creation!), example: bond0.2516 created before bond0
* using 'require_interfaces' allows to fix the order for states
that depend on OVS interfaces
Change-Id: Id4539a7e24f2c55fb62afe5d47e485effc9a882b
Interface name was hardcoded when checking if peer is set for
ovs interface.
This patch unhardcode interface name.
Change-Id: Id465d5f84751b5516aebc9d80716d21d14e56abc
This patch adds dependency for ovs_port_up_{{ interface_name }} on
linux_interfaces_final_include. This is needed to avoid errors like
"Unknown interface XXX"
Change-Id: I763ff8ae32a170df3802a65a547c6e76959a0137
Change of mtu is not handled by reclass. Example interface:
```
br_ctl:
enabled: true
type: bridge
proto: static
mtu: 1450
address: ${_param:single_address}
gateway: ${_param:ctl_gateway}
netmask: ${_param:control_network_netmask}
use_interfaces:
- ${_param:primary_first_nic}
```
This enables a more fine tuned dpdk for better performance.
More details on pmd-rxq-affinity config:
http://docs.openvswitch.org/en/latest/howto/dpdk/
Change-Id: I630c6af94ff733619fb175f522fc515984862ff6
We need to save our manualy managed interfaces to different directory to
workaround salt bug that causes interface deconfiguration in
/etc/network/interfaces for interfaces manualy configured in
/etc/network/interfaces.d
Upstream-Bug: https://github.com/saltstack/salt/issues/40262
Config:
linux:
network:
tap_custom_txqueuelen: 10000
in case of configuration parameter defined will create file:
/etc/udev/rules.d/60-net-txqueue.rules
with content:
KERNEL==”tap[0-9a-z\-]*", RUN+="/sbin/ip link set %k txqueuelen 10000"
Add possiblity to add ip address and mtu on dpdk ovs bridge to
be able use VXLAN as tenant segmentation.
Change-Id: I1394d30c5d9935218841b17ff7651dac7b4abefa
Epic: PROD-8957
Bruno Binet
Martin Horak
Tomas Pipota
Dennis Dmitriev
Dzmitry Stremkouski
Dmitry Ukov
azvyagintsev
Oleksii Chupryn
Oleh Hryhorov
Alexandru Avadanii
Michael Fladischer
Andrey Shestakov
Michael Polenchuk
Jiri Broulik
Denis V. Meltsaykin
Oleg Bondarev
Petr Jediný
Ales Komarek
Petr Michalec
Vasyl Saienko
mceloud
Ramon Melero
Dmitry Tyzhnenko
Oleksandr Vlasov
Andrii Petrenko
Jakub Pavlik