Just a late update on this one.

Since the HA chassis group was introduced in newer versions, it caused
confusion because we saw gateway_chassis as empty . After checking the
source code,  db config and the man page it was clear that part of the new
version changes as expected.

man page https://man7.org/linux/man-pages/man5/ovn-nb.5.html

The preferred way to configure a gateway is ha_chassis_group, but

       gateway_chassis is also supported for backward compatibility.

       Only one of these should be set at a time on a given LRP, since

       they configure the same features.

ovn-sbctl list port_bind 71bd5bd5-e279-430d-82bf-81d3c0ad11d8

_uuid               : 71bd5bd5-e279-430d-82bf-81d3c0ad11d8

additional_chassis  : []

additional_encap    : []

chassis             : 9a9481bf-4f58-4d66-b5c7-70f19f884a7b

datapath            : 954cfc39-d8d6-4b0d-9df7-1fef8c2058ca

encap               : []

external_ids        : {}

gateway_chassis     : []

ha_chassis_group    : b8590651-a39a-4a34-8d3f-d61c3300303a

logical_port        : cr-lrp-slc07-dev-fr2-ic-ts

mac                 : ["aa:aa:bb:00:05:02 169.254.0.52/24"]

mirror_rules        : []

nat_addresses       : []

options             : {always-redirect="true",
distributed-port=lrp-slc07-dev-fr2-ic-ts}

parent_port         : []

port_security       : []

requested_additional_chassis: []

requested_chassis   : []

tag                 : []

tunnel_key          : 3

type                : chassisredirect

up                  : true

virtual_parent      : []

Since the upgrade doc did not explain any details on data plane impact
which we hit, once we do some dry runs to bump from 2.* to 23.*, can amend
the upgrade SOP in ovn repo later if needed.

Also part of moving some chassis to the new 23.* version, we see openflow
count  bump to almost double during dry run. So we will do some more tests.



Aliasgar


On Mon, Jul 1, 2024 at 9:58 AM aginwala <aginw...@asu.edu> wrote:

> Hello team:
>
> Part of upgrading OVN from 2.11* versions to the latest 23.09, we have
> found issues in existing north south gateways where failover doesn't work
> as expected when connected to the latest 23.09 raft control plane.
>
> OVS datapath version 2.16.0-2 on host. Tunneling protocol is stt
> ovn-controller is running in a container with hostnetwork with version
> 23.09 on the north south gateways using ovs user space 3.2.*
>
> Following the
> https://docs.ovn.org/en/latest/intro/install/ovn-upgrades.html#fail-safe-upgrade
> , even if setting northd version on the gateways, when one of the gateway's
> ovn controller is stopped, there is no failover triggered.
>
> For chassis, things work as expected where we see updates in
> ovn controller logs for schema upgrade and data path works fine as expected
> for the workloads scheduled on that chassis.
>
> For n-s gateways, we don't see ovn-controller claiming lports  when
> failover is triggered.
> Also if there are any suggestions for interconnection gateway upgrade
> procedure too, would be great.
>
> Let us know for any suggestions/recommendations as we continue to debug.
>
>
> Regards,
> Aliasgar
>
_______________________________________________
discuss mailing list
disc...@openvswitch.org
https://mail.openvswitch.org/mailman/listinfo/ovs-discuss

Reply via email to