VXLAN tunnel bogus error unexpected type: SAI_OBJECT_TYPE_TUNNEL

[bradh352]

When trying to debug VXLAN tunnels it is bad to have errors output in the logs that are false positives. This can lead users to go down a rabbit hole when the error has nothing to do with the issue at hand.

Here is the error sequence:

Aug 16 03:14:19.373668 Odin-ec58a ERR swss#orchagent: :- meta_sai_on_port_state_change_single: data.port_id oid:0x2a0000000009bf has unexpected type: SAI_OBJECT_TYPE_TUNNEL, expected PORT, BRIDGE_PORT or LAG
Aug 16 03:14:19.375495 Odin-ec58a NOTICE swss#orchagent: :- doTask: Get port state change notification id:2a0000000009bf status:1
Aug 16 03:14:19.375495 Odin-ec58a ERR swss#orchagent: :- doTask: Failed to get port object for port id 0x2a0000000009bf

In this case, as reported in sonic-net/sonic-buildimage#10004 the fix is to simply add the SAI_OBJECT_TYPE_TUNNEL to the switch statement.

This is a trivial correction for an incorrect log entry. It doesn't appear to alter behavior in any way.

It has been confirmed via my own Dell S5248F switches to resolve this false-positive error message.

Signed-off-by: Brad House (@bradh352)

[kcudnik]

tunnel object is not expected there: https://github.com/opencomputeproject/SAI/blob/master/inc/saiport.h#L137
according to latest SAI on master only 3 objects are expected at that point, if tunnel is als oexpected teher, please bring that issue on SAI comunity meeting and post PR on SAI repo to update that metadata

[kcudnik]

according to comments

[kcudnik]

this code should be based on sai metadata, and was written when there was no metadata for those objecs yet, if you want to fix it, please add it using metadata checks

[kcudnik]

i fixed it here: #1469

[bradh352]

@kcudnik

I don't intend on adding a new metadata port type to track by any means. I was just trying to silence that warning as it inhibits the ability of someone attempting to debug VXLAN issues. I guess I didn't fully evaluate what that warning was about.

So it sounds like the better thing to do is to find where that event originates and filter out SAI_OBJECT_TYPE_TUNNEL so it never tries to send a port metadata update using that object type.

Any idea off the top of your head where to do that? I've only been playing with SONiC for about 2 weeks, so I'm still quite green.

[kcudnik]

if onject type tunnel is send in notification and by current SAI headers this is not allowed, then this seems like vendor bug, otherwise SAI metadat should be updated, but not in this PR, im meking generic fix for this and i will post it shortly

[kcudnik]

if this iss issue for you you probably can post new issue on SAI https://github.com/opencomputeproject/SAI/issues and explain the origin

[bradh352]

call chain appears to be Sai::handle_notification() -> RedisRemoteSaiInterface::syncProcessNotification() -> NotificationPortStateChange::processMetadata() -> Meta::meta_sai_on_port_state_change() -> Meta::meta_sai_on_port_state_change_single()

Sai::apiInitialize() makes a context which calls this notification callback.

The chain is a little harder to follow upstream from there. I'm assuming its being called from a different project, maybe swss->orchagent->portsarch ?

[bradh352]

ah, i see you said you'll make a generic fix for that. Thanks.

And I also see you say its probably a vendor bug, so I guess that's bcm sai sending something it shouldn't is what you're saying.

[kcudnik]

maybe tunnel is also valid for that notification and metadata was not updated, that i dont know

fixed this part to use metadata: #1469

[kcudnik]

i fixed it here: #1469

[kcudnik]

will not merge this becaue of above fix