[WIP] Get endpoints for service instead of pods #11

pires · 2015-05-09T11:08:43Z

This is work-in-progress to fix #9 and improve node-discovery efficiency.

Need guidance on how to test this
Need a new release of fabric8

pires · 2015-05-16T08:40:07Z

@jimmidyson sorry about the formatting changes and some cosmetic changes but I tried hard to keep up with your conventions. How does this look?

I'm going to need you help testing this with perhaps a snapshot release or something? I'm open to ideas.

jimmidyson · 2015-05-16T08:51:07Z

src/main/java/io/fabric8/elasticsearch/discovery/k8s/K8sUnicastHostsProvider.java

+                    // pod IP is a single IP Address. We need to build a TransportAddress from it
+                    final TransportAddress[] addresses;
+                    try {
+                        addresses = transportService.addressesFromString(endpointIp.concat(":9300"));


Is there no way to get the port from the endpoint list?

Yes this should be doable. See https://github.com/GoogleCloudPlatform/kubernetes/blob/master/pkg/api/types.go#L1091 ff for more details. Basically it's a Cartesian product between addresses and ports in each endpoint subset.

While I'm hardcoding port 9300 and thus limiting the user choices here, I'm concerned that a multiport service will make it harder for us to determine which port to use.

Good point. I think using port 9300 for now is fine. Almost all deployments will use that I'm sure.

jimmidyson · 2015-05-16T08:55:31Z

@pires Thanks for this. Overall looks good. I wonder if we could use fabric8 arquillian to add some integration tests for this? Happy to have that later though. For now you should be able to build locally & use in a local docker image? Don't like publishing snapshots personally & would rather not do that.

pires · 2015-05-17T21:02:25Z

I wonder if we could use fabric8 arquillian to add some integration tests for this?

Probably would be a good thing to do. Would we need something like jube to actually bring Kubernetes up?

For now you should be able to build locally & use in a local docker image? Don't like publishing snapshots personally & would rather not do that.

OK. Will do that tomorrow then.

jimmidyson · 2015-05-17T21:25:34Z

No - fabric8 arquillian can spin up pods in a kubernetes cluster for integration testing. See http://fabric8.io/guide/testing.html for more details. It's pretty awesome! We're using it in some CD scenarios & it works great. But like I said, this can come later - we will merge this once you're happy with it.

Let me know how your testing goes.

mulloymorrow · 2015-05-26T18:23:56Z

+1

pires · 2015-05-26T21:31:56Z

I'm sorry, I have been overwhelmed with unrelated work.

For now you should be able to build locally & use in a local docker image?

Yes, will do this for now.
Unfortunately, I think I won't be able to look at Arquillian in the near future.

mulloymorrow · 2015-05-26T21:33:14Z

I'll try and look at Arquillian and test this out locally.

pires · 2015-05-26T21:33:42Z

@mulloymorrow that would be A-W-E-S-O-M-E!

mulloymorrow · 2015-05-27T22:42:45Z

Can we merge this pull request with a new branch other than master? I'd like to include these commits in a fork for testing.

jimmidyson · 2015-05-28T06:58:59Z

@mullowmorrow Perhaps @pires could add you as a collaborator on his fork?

mulloymorrow · 2015-05-28T18:47:15Z

src/main/java/io/fabric8/elasticsearch/discovery/k8s/K8sUnicastHostsProvider.java

-                    // We can ignore it in the list of DiscoveryNode
-                    logger.trace("current node found. Ignoring {} - {}", pod.getId(), podIp);
+        // get all endpoints for service
+        final Endpoints endpoints = getKubernetes().endpointsForService(this.selector, null);


IAE thrown by kubernetes-api-2.1.0::KubernetesClient.validateNamespace: Template parameter set to null.

2015-05-28T18:17:16.071899300Z [2015-05-28 18:17:16,071][WARN ][io.fabric8.elasticsearch.discovery.k8s.K8sUnicastHostsProvider] [Cayman] Exception caught during discovery java.lang.IllegalArgumentException : Template parameter value is set to null 2015-05-28T18:17:16.071899300Z java.lang.IllegalArgumentException: Template parameter value is set to null 2015-05-28T18:17:16.071899300Z at org.apache.cxf.jaxrs.impl.UriBuilderImpl.doBuild(UriBuilderImpl.java:97) 2015-05-28T18:17:16.071899300Z at org.apache.cxf.jaxrs.impl.UriBuilderImpl.buildFromEncoded(UriBuilderImpl.java:230) 2015-05-28T18:17:16.071899300Z at org.apache.cxf.jaxrs.client.ClientProxyImpl.invoke(ClientProxyImpl.java:180) 2015-05-28T18:17:16.071899300Z at com.sun.proxy.$Proxy29.endpointsForService(Unknown Source) 2015-05-28T18:17:16.071899300Z at io.fabric8.elasticsearch.discovery.k8s.K8sUnicastHostsProvider.getNodesFromKubernetesSelector(K8sUnicastHostsProvider.java:119) 2015-05-28T18:17:16.071899300Z at io.fabric8.elasticsearch.discovery.k8s.K8sUnicastHostsProvider.buildDynamicNodes(K8sUnicastHostsProvider.java:102) 2015-05-28T18:17:16.071899300Z at org.elasticsearch.discovery.zen.ping.unicast.UnicastZenPing.sendPings(UnicastZenPing.java:313) 2015-05-28T18:17:16.071899300Z at org.elasticsearch.discovery.zen.ping.unicast.UnicastZenPing$2.doRun(UnicastZenPing.java:228) 2015-05-28T18:17:16.071899300Z at org.elasticsearch.common.util.concurrent.AbstractRunnable.run(AbstractRunnable.java:36) 2015-05-28T18:17:16.071899300Z at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142) 2015-05-28T18:17:16.071899300Z at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617) 2015-05-28T18:17:16.071899300Z at java.lang.Thread.run(Thread.java:745)

We probably need to set a default namespace, like default.

Testing default

jimmidyson · 2015-05-28T20:27:38Z

ok to test

fusesource-ci · 2015-05-28T20:27:39Z

Can one of the admins verify this patch?

fusesource-ci · 2015-05-28T20:27:50Z

Refer to this link for build results (access rights to CI server needed):
https://fabric8-ci.fusesource.com/job/elasticsearch-cloud-pullreq/1/
Test FAILed.

jimmidyson · 2015-05-28T20:29:32Z

retest this please

fusesource-ci · 2015-05-28T20:30:06Z

Please comment with '[merge]' to automerge this request

fusesource-ci · 2015-05-28T20:30:07Z

Refer to this link for build results (access rights to CI server needed):
https://fabric8-ci.fusesource.com/job/elasticsearch-cloud-pullreq/2/
Test FAILed.

jimmidyson · 2015-05-28T21:11:58Z

@pires @mulloymorrow I've merged this locally & made some changes to support your use case, getting endpoints from service name. I've tested & it works fine. Please can you review service-endpoints branch? If you're happy then I will merge this to master & release.

mulloymorrow · 2015-05-30T02:20:47Z

We're running in v1beta2, currently. Services in our environment can be found vi /api/v1beta2/services/<SERVICE_NAME>?namespace=default.

jimmidyson · 2015-05-30T06:17:54Z

As v1beta1 & v1beta2 has been dropped from kubernetes we've removed those api versions from the fabric8 client, only supporting v1beta3 ATM. Soon that will change to only v1. Sorry but you'll need to upgrade your cluster in order to use this plugin going forward. We're targeting the v1 API in line with kubernetes 1.0 release.

pires · 2015-05-30T10:39:42Z

I agree with Jimmi. You really need to upgrade.
On May 30, 2015 07:17, "Jimmi Dyson" notifications@github.com wrote:

As v1beta1 & v1beta2 has been dropped from kubernetes we've removed those
api versions from the fabric8 client, only supporting v1beta3 ATM. Soon
that will change to only v1. Sorry but you'll need to upgrade your cluster
in order to use this plugin going forward. We're targeting the v1 API in
line with kubernetes 1.0 release.

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

mulloymorrow · 2015-06-01T18:25:32Z

Switched over to Gcloud! Will be testing there.

pires · 2015-06-01T19:01:16Z

You mean GKE?

Also, I'm sorry for not updating this, but right now there's a lot going on regarding reading the Kubernetes API that will have an impact on a lot of the stuff I'm working on and most probably this project as well.

@jimmidyson did you check the elasticsearch example from @satnam6502 on the Kubernetes tree? They're using service-accounts for allowing secure access to the API and it seems read-only on an unsecure port will be dropped soon.

mulloymorrow · 2015-06-01T19:37:55Z

@pires , Yes. The Google Container Engine.

jimmidyson · 2015-06-02T08:42:14Z

@pires I'm working on the fabric8 kubernetes API support for service accounts now, but that shouldn't stop this PR going through. Separate issue for that please.

pires · 2015-06-02T09:27:54Z

@jimmidyson point taken service-endpoints branch works for me with Kubernetes 0.17.0 and 0.18.0, so looks good. Care to merge then?

jimmidyson · 2015-06-02T09:41:06Z

@pires Cool thanks for the feedback. Btw seen the advice to use SRV records rather than API server calls?

pires · 2015-06-02T09:43:16Z

@jimmidyson yes, but see kubernetes/kubernetes#9107. We should be able to work without DNS, because it's not something that is granted in every Kubernetes cluster.

jimmidyson · 2015-06-02T09:44:16Z

@pires Agreed - was just pointing out that I prefer DNS if available.

jimmidyson · 2015-06-02T09:48:43Z

Superseded by #12

jimmidyson · 2015-06-02T09:48:53Z

Thanks @pires!

pires · 2015-06-02T09:56:54Z

Thank you @jimmidyson

mulloymorrow · 2015-06-03T03:22:26Z

Thanks @pires @jimmidyson

Moved to a cleaner way of finding pods exposing Elasticsearch transport.

86185b3

pires mentioned this pull request May 12, 2015

ES cluster cannot startup #9

Closed

Upgraded to fabric8-2.1.0 and Elasticsearch 1.5.2.

2925c66

jimmidyson reviewed May 16, 2015
View reviewed changes

mulloymorrow reviewed May 28, 2015
View reviewed changes

jimmidyson closed this Jun 2, 2015

pires deleted the 9-endpoints_vs_pods branch October 26, 2016 17:48

[WIP] Get endpoints for service instead of pods #11

[WIP] Get endpoints for service instead of pods #11

Conversation

pires commented May 9, 2015

pires commented May 16, 2015

jimmidyson May 16, 2015

Choose a reason for hiding this comment

jimmidyson May 16, 2015

Choose a reason for hiding this comment

pires May 17, 2015

Choose a reason for hiding this comment

jimmidyson May 17, 2015

Choose a reason for hiding this comment

jimmidyson commented May 16, 2015

pires commented May 17, 2015

jimmidyson commented May 17, 2015

mulloymorrow commented May 26, 2015

pires commented May 26, 2015

mulloymorrow commented May 26, 2015

pires commented May 26, 2015

mulloymorrow commented May 27, 2015

jimmidyson commented May 28, 2015

mulloymorrow May 28, 2015

Choose a reason for hiding this comment

pires May 28, 2015

Choose a reason for hiding this comment

mulloymorrow May 28, 2015

Choose a reason for hiding this comment

jimmidyson commented May 28, 2015

fusesource-ci commented May 28, 2015

fusesource-ci commented May 28, 2015

jimmidyson commented May 28, 2015

fusesource-ci commented May 28, 2015

fusesource-ci commented May 28, 2015

jimmidyson commented May 28, 2015

mulloymorrow commented May 30, 2015

jimmidyson commented May 30, 2015

pires commented May 30, 2015

mulloymorrow commented Jun 1, 2015

pires commented Jun 1, 2015

mulloymorrow commented Jun 1, 2015

jimmidyson commented Jun 2, 2015

pires commented Jun 2, 2015

jimmidyson commented Jun 2, 2015

pires commented Jun 2, 2015

jimmidyson commented Jun 2, 2015

jimmidyson commented Jun 2, 2015

jimmidyson commented Jun 2, 2015

pires commented Jun 2, 2015

mulloymorrow commented Jun 3, 2015