use public address for deployer pod environment #1226
Conversation
| @@ -609,8 +609,8 @@ func (c *MasterConfig) RunBuildImageChangeTriggerController() { | |||
| func (c *MasterConfig) RunDeploymentController() { | |||
| _, kclient := c.DeploymentControllerClients() | |||
| env := []api.EnvVar{ | |||
| {Name: "KUBERNETES_MASTER", Value: c.MasterAddr}, | |||
| {Name: "OPENSHIFT_MASTER", Value: c.MasterAddr}, | |||
| {Name: "KUBERNETES_MASTER", Value: c.MasterPublicAddr}, | |||
There was a problem hiding this comment.
Doesn't that mean --master is set incorrectly? --master should be the address in-cluster things (e.g. nodes) can use to reach the master. --public-master is the address out of cluster things (e.g. browser, osc from user's desktop) can use to reach the master.
There was a problem hiding this comment.
There are cases where the public address won't be accessible from within the cluster, so I don't think just switching to that is correct.
There was a problem hiding this comment.
Thanks @liggitt I'll take a look at how the service is starting up. If there is a use case where public address is not accessible in the cluster then this is certainly wrong
There was a problem hiding this comment.
@liggitt That really helps clarify, thanks.
|
10.0.2.15 is virtualbox endpoint and should not be used anywhere in a multi-cluster env. That is precise reason 'config.vm.network "private_network", ip: 10.245..' was created. |
|
@rajatchopra I think @liggitt 's comment is on the right track, probably the openshift-master is not starting with the right address. I'll investigate |
|
Of note, the 10.0.2.15 only appears in the master's kubeconfig. The minions all have the 10.245 address |
|
Seems like changing |
|
Closing for the correct fix, will test it out tomorrow and submit |
|
that said, if $MASTER_IP can't be used from within the master vm itself, then we sort of have 3 network boundaries: master, in-cluster, out-of-cluster. We're not equipped to deal with that at the moment (with just two specifiable addresses). Let me know if that ends up being the case, and we'll think about it some more. |
…service-catalog/' changes from ef63307bdb..ae6b643caf ae6b643caf Use oc adm instead of oadm which might not exist in various installations. 66a4eb2a2c Update instructions... will remove once documented elsewhere 1b704d1530 replace build context setup with init containers ee4df18c7f hack/lib: dedup os::util::host_platform and os::build::host_platform 1cd6dfa998 origin: Switch out owners to Red Hatters 664f4d318f Add instructions for syncing repos 2f2cdd546b origin-build: delete files with colon in them cdf8b12848 origin-build: don't build user-broker ebfede9056 origin build: add _output to .gitignore 55412c7e3d origin build: make build-go and build-cross work 68c74ff4ae origin build: modify hard coded path 3d41a217f6 origin build: add origin tooling a8fc27d Fix typos in walkthrough (openshift#1224) e77edbf openshift#1157: Limit the amount of time for reconciliations (openshift#1196) 1b1a749 temporarily disabled verify-links.sh from the verify target (openshift#1226) acf8fab Send originating identity headers in OSB requests (openshift#1162) 821ba16 new admission controller to block updates to service instance updates that (openshift#1210) d69c5e5 Minor improvement to godoc in binaries (openshift#1211) 5b81814 fix typos (openshift#1221) 836dc4a Adds how to download Helm chart (charts/catalog) (openshift#1219) 2fd0115 Fix "visit the project on github" link. (openshift#1217) 325e4b6 Add how to set $GOPATH. (openshift#1218) 68b775f Update the installation (openshift#1199) 6e3a3c1 v0.0.19 (openshift#1207) 8b69791 Removing errexit from TLS setup script (openshift#1206) 273260f Instance deletion lifecycle enhancements, issue openshift#820 (openshift#1159) c050713 fix cleaning of build output for non-root users (openshift#1205) 5995df1 Merge branch 'pr/1204' 72f4802 Remove osb prefix from example ServiceClass (openshift#1201) f9dbd4e pin all dependencies in glide to current version except for glog where we want to pick up the prior version to fix issue 1187. f148bc5 v0.0.18 (openshift#1202) b86ab8d Removing the helm install command (openshift#1185) 3cff482 Remove Alpha* prefix on all API fields for issue openshift#1180 (openshift#1184) 154b74d Fix gofmt issue (openshift#1192) 2ee894a do the clean before building an arch (openshift#1179) b4976ef Fix bad URL (openshift#1189) cd3dede Fix hrefs again (openshift#1190) f066226 Design: Instance/Binding parameters (openshift#1075) eb37682 This generated file is missing from master (openshift#1191) 28c0ae7 Use generation instead of checksum for Instances and InstanceCredentials (openshift#1151) 5cdd323 Fix bad href (openshift#1188) 8a892f0 handle lingering polling cases (openshift#1174) f5fabd6 remove TPRs from Jenkins e2e pipeline (openshift#1175) 717df78 Add godoc explaining that Instance and InstanceCredential specs are immutable (openshift#1182) REVERT: ef63307bdb origin build: add origin tooling git-subtree-dir: cmd/service-catalog/go/src/github.com/kubernetes-incubator/service-catalog git-subtree-split: ae6b643cafd3a17412f173e70ed7c1a2e39ee549
…ess-timeout-3.9 [3.9] Timeout on instances.NodeAddresses cloud provider request
@ramr noticed an issue that in the multi-machine vagrant environment his deployer was not connecting back to master. After inspecting the deployer pod it looks like it was getting the 10.0.2.15 master address which was not accessible from a minion.
We checked with @rajatchopra to see if the ip should be accessible and his thoughts were that it was probably a bug. Switching to use the public address fixed the issue.
@ironcladlou @smarterclayton PTAL