Create a standalone k8s binary, capable of running a full cluster by brendandburns · Pull Request #2121 · kubernetes/kubernetes

brendandburns · 2014-11-02T05:45:51Z

smarterclayton · 2014-11-02T17:58:56Z

IIRC version can race on startup because it loads earlier than other attributes. Can't find the issue now, but we switched to reading the root key in a few places because of it.

Yeah - /version returns 200s before things are up and really running. See: #963 (comment)

I think that's ok. Nothing crashes if etcd isn't fully there, and if there
is a /version there, it will get there eventually...

I can switch to hitting the root key, if we really want to, but unlike the
integration tests, I don't immediately start hammering the server and
expecting correct responses.

--brendan

On Sun, Nov 2, 2014 at 10:11 AM, Joe Beda notifications@github.com wrote:

In pkg/tools/etcd_tools.go:

@@ -358,3 +364,42 @@ func (h *EtcdHelper) AtomicUpdate(key string, ptrToType runtime.Object, tryUpdat
return err
}
}
+
+func checkEtcd(host string) error {

response, err := http.Get(host + "/version")

Yeah - /version returns 200s before things are up and really running.
See: #963 (comment)
#963 (comment)

—
Reply to this email directly or view it on GitHub
https://github.com/GoogleCloudPlatform/kubernetes/pull/2121/files#r19713556
.

derekwaynecarr · 2014-11-03T03:02:52Z

Is there a plan to add an option in the future to just start as go-routines what would typically go in the kubernetes-master, and an option to just start what would go on the kubernetes-minions using this single top-level binary instead of the separate daemons? I like this as it more aligns with what we started doing in OpenShift Origin v3, and makes the CLI setup much simpler.

brendandburns · 2014-11-03T03:31:36Z

Yes, I'd like to achieve this. So there's a single binary for everything,
and you pass in --kubelet or --master or whatever.

There's more work to do to achieve this, mostly around networking, but yes,
that's my goal.

--brendan

On Sun, Nov 2, 2014 at 7:03 PM, Derek Carr notifications@github.com wrote:

Is there a plan to add an option in the future to just start as
go-routines what would typically go in the kubernetes-master, and an option
to just start what would go on the kubernetes-minions using this single
top-level binary instead of the separate daemons? I like this as it more
aligns with what we started doing in OpenShift Origin v3, and makes the CLI
setup much simpler.

—
Reply to this email directly or view it on GitHub
#2121 (comment)
.

smarterclayton · 2014-11-03T03:50:12Z

Structurally, we went with a config -> Run pattern for each of the routines. https://github.com/openshift/origin/blob/master/pkg/cmd/server/kubernetes/master.go has some examples of how we launch the master, although some of this we have to change as the authorizer changes @erictune is working on.

erictune · 2014-11-03T16:48:09Z

Would you either change this PR or file an issue so that cmd/apiserver calls RunApiServer
and likewise for kubelet, scheduler, and controller?
Otherwise, if I want to write a test that tests whether an apiserver is constructed with the right config, I will have to test two binaries instead of one. And keeping them in sync is a pain.

brendandburns · 2014-11-03T16:50:32Z

Agree that's the long term goal. I'll file an issue.

On Mon, Nov 3, 2014 at 8:48 AM, Eric Tune notifications@github.com wrote:

Would you either change this PR or file an issue so that cmd/apiserver
calls RunApiServer
and likewise for kubelet, scheduler, and controller?
Otherwise, if I want to write a test that tests whether an apiserver is
constructed with the right config, I will have to test two binaries instead
of one. And keeping them in sync is a pain.

—
Reply to this email directly or view it on GitHub
#2121 (comment)
.

erictune · 2014-11-03T19:44:59Z

So, is your long term goal to get rid of the standalone binaries?

dchen1107 · 2014-11-03T19:45:31Z

What is the motivation behind this? For testing? or other stuff? I couldn't figure out by reading the PR.

dchen1107 · 2014-11-03T19:49:10Z

ok, @bgrant0607 just pointed #108 to me.

brendandburns · 2014-11-03T19:54:20Z

The goal of this is to both reduce the size of our download (since each
individual binary contains duplicate code) and simplify our install
experience (since you only need to install one binary instead of N)

Also to make local development/experimentation easier, since you only need
to run a single binary locally to have a cluster to play around with.

--brendan

On Mon, Nov 3, 2014 at 11:49 AM, Dawn Chen notifications@github.com wrote:

ok, @bgrant0607 https://github.com/bgrant0607 just pointed #108
#108 to me.

—
Reply to this email directly or view it on GitHub
#2121 (comment)
.

dchen1107 · 2014-11-03T23:26:24Z

Ok, with this PR, are all cmd/$binary should be removed by this single binary?

brendandburns · 2014-11-10T19:27:49Z

Eventually I will remove binaries, but for now, I want to just check in the standalone version.

Rebased, please take another look.

Thanks!
--brendan

smarterclayton · 2014-11-10T22:20:39Z

When you move on to the next step I want to sync about standardizing a run in a Docker container config for the Kubelet - I got the first steps working in OpenShift re: running in a container (https://lists.openshift.redhat.com/openshift-archives/dev/2014-November/msg00042.html) but I want to figure out what security implications and patterns are going to work best for that and make sure we match.

On Nov 10, 2014, at 2:28 PM, Brendan Burns notifications@github.com wrote:

Eventually I will remove binaries, but for now, I want to just check in the standalone version.

Rebased, please take another look.

Thanks!
--brendan

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

brendandburns · 2014-11-11T05:31:32Z

Travis break was a flake. I can have mergez?

dchen1107 · 2014-11-11T06:04:25Z

LGTM

Create a standalone k8s binary, capable of running a full cluster

lavalamp · 2014-11-13T00:44:05Z

This apiserver setup code looks wrong. https://github.com/GoogleCloudPlatform/kubernetes/blob/master/cmd/apiserver/apiserver.go is quite different.

Should extract code into a library-- perhaps in pkg/master/cmd/ or something.

Not much point to an all-in-one binary if it doesn't behave the same as our current binaries.

jbeda · 2014-11-13T00:46:24Z

@lavalamp I'm going to be picking this stuff up. I'll continue to refactor this and will make sure you are the reviews. Thanks.

bgrant0607 · 2014-11-13T01:09:10Z

@jbeda Please hold off on this particular part. I'm merging #2282 with these changes at this moment.

bgrant0607 · 2014-11-13T01:16:17Z

Yeah, I have to fix this. standalone.go is doing things that now require access to private members of the master.

lavalamp · 2014-11-13T01:18:28Z

I think this was forked off of cmd/apiserver.go before @erictune's refactoring. Otherwise I'm not sure how it got like this.

smarterclayton · 2014-11-13T01:18:48Z

Anything standalone go needs Openshift will need :)

On Nov 12, 2014, at 8:16 PM, bgrant0607 notifications@github.com wrote:

Yeah, I have to fix this. standalone.go is doing things that now require access to private members of the master.

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

bgrant0607 · 2014-11-13T01:26:31Z

I think I can just delete the mux and calls to NewAPIGroup. All of that is already done by master.New.

dchen1107 · 2014-11-13T01:45:06Z

Ahh, just realized you guys were talking about this PR which I merged two days ago. Yes, this was merged before @erictune's refactoring and auth stuff in. Now I believe this shouldn't work with the rest of system with newly PRs for auth. My bad. I should insist on cleaning / consolidate two copies of apiserver, or at least add a TODO in cmd/apiserver/apiserver.go.

bgrant0607 · 2014-11-13T01:46:43Z

Pleases nobody touch this now

brendandburns force-pushed the standalone branch 3 times, most recently from eff91f1 to d953a94 Compare November 2, 2014 05:56

smarterclayton reviewed Nov 2, 2014
View reviewed changes

dchen1107 self-assigned this Nov 3, 2014

brendandburns force-pushed the standalone branch from d953a94 to c268d49 Compare November 10, 2014 19:17

Make a standalone binary.

2c12218

brendandburns force-pushed the standalone branch from c268d49 to 2c12218 Compare November 10, 2014 21:34

dchen1107 added a commit that referenced this pull request Nov 11, 2014

Merge pull request #2121 from brendandburns/standalone

30fcf24

Create a standalone k8s binary, capable of running a full cluster

dchen1107 merged commit 30fcf24 into kubernetes:master Nov 11, 2014

jbeda mentioned this pull request Nov 11, 2014

Dramatically simplify Kubernetes deployment #2303

Closed

22 tasks

lavalamp reviewed Nov 13, 2014
View reviewed changes

Conversation

brendandburns commented Nov 2, 2014

Uh oh!

smarterclayton Nov 2, 2014

Choose a reason for hiding this comment

Uh oh!

jbeda Nov 2, 2014

Choose a reason for hiding this comment

Uh oh!

brendandburns Nov 2, 2014

Choose a reason for hiding this comment

Uh oh!

derekwaynecarr commented Nov 3, 2014

Uh oh!

brendandburns commented Nov 3, 2014

Uh oh!

smarterclayton commented Nov 3, 2014

Uh oh!

erictune commented Nov 3, 2014

Uh oh!

brendandburns commented Nov 3, 2014

Uh oh!

erictune commented Nov 3, 2014

Uh oh!

dchen1107 commented Nov 3, 2014

Uh oh!

dchen1107 commented Nov 3, 2014

Uh oh!

brendandburns commented Nov 3, 2014

Uh oh!

dchen1107 commented Nov 3, 2014

Uh oh!

brendandburns commented Nov 10, 2014

Uh oh!

smarterclayton commented Nov 10, 2014

Uh oh!

brendandburns commented Nov 11, 2014

Uh oh!

dchen1107 commented Nov 11, 2014

Uh oh!

lavalamp Nov 13, 2014

Choose a reason for hiding this comment

Uh oh!

jbeda commented Nov 13, 2014

Uh oh!

bgrant0607 commented Nov 13, 2014

Uh oh!

bgrant0607 commented Nov 13, 2014

Uh oh!

lavalamp commented Nov 13, 2014

Uh oh!

smarterclayton commented Nov 13, 2014

Uh oh!

bgrant0607 commented Nov 13, 2014

Uh oh!

dchen1107 commented Nov 13, 2014

Uh oh!

bgrant0607 commented Nov 13, 2014

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

8 participants