Override the host header when health checking the kubelet #7115
Conversation
The kubelet checks the host header, to make sure the master has the right name for the minion. This lets us separate the network-address from the node name.
|
With #7114, this should let us choose node names arbitrarily. |
|
You are faster than me on picking up this item. Thanks! Reading ... |
There was a problem hiding this comment.
I am confused here, node.Name and nodeAdd are the same, and you are initializing both Server.Addr and Server.HostHeader to the same value. Is that what you want?
There was a problem hiding this comment.
In #7114 I change nodeAddr to something derived from the NodeAddresses, which is (in general) going to be different from the node.Name. Probably should have done this as one PR.
|
Actually this PR is not required, please see #7092 for detail. Thanks! |
|
Do we know when #7092 is going to land? I believe AWS is totally broken right now which isn't a great place to be. If it's going to be a while, I could do a "temporary fix"... |
|
@dchen1107 I don't know! Certainly this is no longer required to get AWS working again. But it sounds like @cjcullen may also want to avoid the requirement that node names be resolvable (#7114 (comment)). It would make the AWS code nicer, and it sounds like it would be helpful for GCE also. I'm not entirely sure - if we get #7092 done quickly then great. Or I could make any changes required to this and #7114, and then we aren't blocked on #7092. For AWS at least, this is now just a "nice to have" which unblocks some further cleanups without waiting on #7092. |
|
Replacing the host header seems like we're abusing the Host field for source detection - I can't tell whether it's an ugly hack or an elegant hack but it was a "surprising" use of Host. |
|
I believe we have now worked around the need for this, so closing. |
The kubelet checks the host header, to make sure the master has the
right name for the minion.
This lets us separate the network-address from the node name.