Triage wrt 1.0? What happens if we set on this for a few weeks and merge
after 1.0?

On Wed, Jul 1, 2015 at 10:31 PM, Justin Santa Barbara <
[email protected]> wrote:

Fix bug I introduced when I renamed AWS nodes so that their names
were their AWS instance ids, instance of a resolvable DNS name.

Fix #10612

#10612

You can view, comment on, or merge this pull request online at:

#10663
Commit Summary

• Don't assume that NodeName == Node host name

File Changes

• M pkg/master/master.go
  https://github.com/GoogleCloudPlatform/kubernetes/pull/10663/files#diff-0
  (15)
• M pkg/registry/minion/registry.go
  https://github.com/GoogleCloudPlatform/kubernetes/pull/10663/files#diff-1
  (5)
• M pkg/registry/pod/etcd/etcd.go
  https://github.com/GoogleCloudPlatform/kubernetes/pull/10663/files#diff-2
  (30)
• M pkg/registry/pod/rest.go
  https://github.com/GoogleCloudPlatform/kubernetes/pull/10663/files#diff-3
  (32)

Patch Links:

• https://github.com/GoogleCloudPlatform/kubernetes/pull/10663.patch
• https://github.com/GoogleCloudPlatform/kubernetes/pull/10663.diff

—
Reply to this email directly or view it on GitHub
#10663.

GCE e2e build/test failed for commit f8a9211.

• Build Log
• Test Artifacts
• Internal Jenkins Results

So I definitely screwed up here. If we want logs/proxy/exec to work on AWS in 1.0, we should either merge this (or something like it), or revert #9728. I'm OK either way (and I'm so sorry; I don't understand how I didn't catch this earlier - I think I've been focused on the systemd work and incorrectly attributing some failures to that).

This PR still needs a little more work because the 'proxy' command is currently using PodIP rather than the NodeName, but once this is complete I think this is the "more correct" approach. I'm working on that right now. But I think for 1.0 we can revert #9728 if that is less risky, though then there will be additional work needed in the AWS code to map names -> instances again.

Another option would be to try to have this code only be triggered on AWS, although that feels like we're adding complexity.

Long-term I like this PR because we might imagine a world where one day we have to use a tunnel to contact a node, or sometimes we can go direct via an internal IP, and sometimes we have to use an external IP.

For the short-term (1.0) though, I don't think this is too risky (although I note that tests are currently failing), but I understand if we choose to revert #9728 instead.

@justinsb: The pull request builder thinks you broke port forwarding and exec on GCE with this, so that's definitely your first gate.

(Driveby) This seems somewhat risky to take at this point.

I think the roll-forward is less risky than the roll-back at this point, but I am not sure it's worth the risk either way. I want second opinions, but I am inclined to just document that v1 is broken in some regards on AWS and that v1.0.1 will fix it. @quinton-hoole because we discussed the importance of AWS support.

@bgrant0607 we should make a decision on this ASAP.

I am going to vote against my own PR here. Although this PR fixes logs, it looks like it does not fix exec / proxy, because they fall foul of the SSL certificate (which is by node-name). Fixing that would be very invasive I think.

I am preparing a patch that rolls back just the AWS renaming portion of #9728 (i.e. the minimal rollback). It will have also to map names to instance ids in a few places.

I think this will be confined to AWS. Sorry for the mess. Hopefully we can clean up node-names in 1.1

Thanks Justin. Reasonable as always.

On Thu, Jul 2, 2015 at 9:25 AM, Justin Santa Barbara <
[email protected]> wrote:

I am going to vote against my own PR here. Although this PR fixes logs, it
looks like it does not fix exec / proxy, because they fall foul of the SSL
certificate (which is by node-name). Fixing that would be very invasive I
think.

I am preparing a patch that rolls back just the AWS renaming portion of
#9728 #9728 (i.e.
the minimal rollback). It will have also to map names to instance ids in a
few places.

I think this will be confined to AWS. Sorry for the mess. Hopefully we can
clean up node-names in 1.1

—
Reply to this email directly or view it on GitHub
#10663 (comment)
.

I'm with @justinsb on this. The minimal roll-back seems the most sensible option right now.

Will hang tight for a minimal rollback..

On Thu, Jul 2, 2015 at 11:06 AM, Nikhil Jindal [email protected]
wrote:

Assigned #10663
#10663 to @thockin
https://github.com/thockin.

—
Reply to this email directly or view it on GitHub
#10663 (comment)
.

We merged #10699 instead (thanks!). Closing this one.