Skaffold: port-forwarding in `skaffold dev` is basically broken

[dgageot]

Now that we track for port collision, skaffold dev will use a new port each time a pod is redeployed.

For example:

• The first time I deploy a pod that's on port 8080, it will be exposed on local port 8080
• When I make a file change, it's redeployed and exposed on port 4503
• Then 4504...

[cmoad]

Also experiencing this after trying 0.22.0. Old behavior would maintain the exposed service port. Now that only happens on the first deploy and all subsequent deploys change to a different port as described above.

[priyawadhwa]

This issue should have been fixed by #1616 -- we'd recommend using skaffold built from master, or using v0.21.0 until v0.23.0 comes out next week with the fix!

[lsowen]

This behaviour seems to be happening again in v0.25.0.

[botwhytho]

I can confirm this behavior is still around in v0.26.0. I reverted to v.0.23.0 and the problem is indeed fixed on that version. Adding some tests around this would be great so this regression doesn't occur again.

Port forwarding during skaffold dev is a great value add for developers and can't really get more adoption from developer teams if this feature is broken. Too bad I'm having to stay on an older version for this to work at the moment.

[arshbot]

@priyawadhwa any idea on when a fix could make it's way into master? Just tested v0.27.0 and it still has this issue

[priyawadhwa]

Hey @arshbot -- I suspect this is happening because "pod name" was re-added to the port-forwarding key in #1780, so every time a pod is recreated it's assigned to a new port.

I'm not sure if removing it will work because I don't want to re-introduce the race condition that PR was meant to fix. I'll go ahead and reopen this issue, our team is going to look at fixing port-forwarding soon & we will definitely take a look at this.

cc @nkubala just wanted to confirm, pod name is necessary in the key, right?

[nkubala]

@priyawadhwa pod name is not necessary, I should have thought about it a bit more. this will be addressed with some of the upcoming port forwarding improvements.

[avloss]

Thank you for taking note of this issue - I just want to make sure you understand the magnitude of this error for independent developers. This issue makes local development of Microservices with tools like Postman practically unfeasible - I've reverted to 0.23.0 for local development. While much praised "Cloud Code" needs a more recent version. So currently I have TWO versions of skaffold locally.

[tejal29]

Related to #1815

[priyawadhwa]

I just removed the podname from the key in #2047 -- could someone please confirm if this fixes their issue? You can find instructions for installing the latest version of skaffold at HEAD here under Latest bleeding edge binary

[arshbot]

I can confirm that at first glance Latest bleeding edge binary works great on my machine. I'll continue using it and post any issues here.

If I could make a suggestion, do you think it'd be possible to add in some tests for this feature ( and hopefully others like this ) to prevent regression in the future? Tests are really useful in detecting subtle behavior changes that occur outside the scope of a newly implemented feature.

[priyawadhwa]

Cool, thanks @arshbot. Added a unit test in #2059.

I'll go ahead and close this issue then, and if anyone sees more errors please comment here and I'll open it up again.