DLPX-69865 IP configuration is lost if a MAC address changes#23
Conversation
The prior commit 271b349 appears to only work on ESX based DE due to the fact that we don't use any cloud datasource when running on that platform. This change attempts to extend that same change, but to work properly on the Azure, Ec2, and Oracle cloud platforms.
|
|
|
I tested this on a dlpx-trunk based VM in AWS and verified the network config was correct: |
pzakha
left a comment
There was a problem hiding this comment.
Shouldn't we also do a similar change for gcp?
Can you point to the code that you're concerned about? I didn't see where we match on MAC for GCP, but I might have overlooked something. |
|
I'm not sure GCP is an issue. I had just noticed that GCP wasn't mentioned in this review, however it does have its own datasource, so I was wondering if it was being handled by the logic in the previous review. Might be worth to verify it if you haven't already. I'm also wondering if we'd want some tests on the appliance-level to make sure we are using the proper type of matching for the interface. |
|
Here's what I see on a GCP based VM from DCOA: so it looks like we're good as-is.
I don't understand.. what do you mean? |
Great!
Like how do we make sure we don't regress back to the default mac-matching behaviour on some platforms. I believe we should be fine in the future for Azure, ec2 and Oracle since we have explicit unit tests in cloud-init, but from my understanding Upstream could modify other DataSources, such as GCP, and have them implement MAC matching without us even noticing. Also, if we were to add a new DataSource, this is likely something we'd overlook unless we have an explicit test (such as a dxos-test) that would run on that platform and catch it, or perhaps some notes for gotchas to look out for when implementing support for a new DataSource. We may want to consult with Platform QA to see what are their thoughts on this. |
Yea, I see what you mean now. IMO, in an ideal situation, we'd have QA verify the cases where this has been an issue for customers, by performing the same VM operations that previously got us into the state where networking was broken. I'm open to discussing this at one of our team meetings. |
|
I've also verified this change on an Azure based VM: |
|
There are some test which add NICs to these environments and we should check to make sure that we don't see a regression in those cases. The code LGTM. |
The prior commit 271b349 appears to only work on ESX based DE due to the
fact that we don't use any cloud datasource when running on that
platform. This change attempts to extend that same change, but to work
properly on the Azure, Ec2, and Oracle cloud platforms.