oc scale against a dc is fragile in the presence of races

[smarterclayton]

Attempting to scale a DC while the controller is creating new RCs returns confusing and vague error messages. If the conditional create of an RC fails (because it's being updated by the DC), we should reload the RC and check that it still current (not failed or completed) and try once more to save. Even a few retries (well factored) would make this a safer experience.

pod "test-deployment-config-1-deploy" deleted
deploymentconfig "test-deployment-config" created
Started deployment #2
CONTROLLER                 CONTAINER(S)      IMAGE(S)                                      SELECTOR                                                                                           REPLICAS   AGE
test-deployment-config-2   ruby-helloworld   127.0.0.1:5001/openshift/origin-ruby-sample   deployment=test-deployment-config-2,deploymentconfig=test-deployment-config,name=test-deployment   0          0s
Scaling the controller failed with: replicationControllers "test-deployment-config-3" not found; Current resource version Unknown
!!! Error in test/cmd/deployments.sh:41
  'oc scale dc test-deployment-config --replicas=1' exited with status 1
Call stack:
  1: test/cmd/deployments.sh:41 main(...)

[smarterclayton]

@Kargakis @ironcladlou

[liggitt]

whatever we do here, we should come up with a pattern that makes sense for the label, annotate, volume commands

[smarterclayton]

Let's open an upstream issue - "Client commands tend to race against
controllers due to status changes / scheduling when used in bash" with the
fix being "client spec changes should be against spec only using generation
or "

On Tue, Aug 25, 2015 at 12:30 AM, Jordan Liggitt notifications@github.com
wrote:

whatever we do here, we should come up with a pattern that makes sense for
the label, annotate, volume commands

—
Reply to this email directly or view it on GitHub
#4327 (comment).

Clayton Coleman | Lead Engineer, OpenShift

[ironcladlou]

New/related upstream issue to lay the foundation for safe concurrent scaling/deployment: kubernetes/kubernetes#14062

[liggitt]

probably addressed by dc/scale changes

[ironcladlou]

Fixed by #5875

[smarterclayton]

Such confidence.

On Wed, Dec 9, 2015 at 9:46 AM, Dan Mace notifications@github.com wrote:

Fixed by #5875 #5875

—
Reply to this email directly or view it on GitHub
#4327 (comment).

[ironcladlou]

oc scale against a dc no longer touches RCs as of #5875, so there's no longer a way for the reported condition to arise.

[0xmichalis]

It doesn't even touch the DC.

[0xmichalis]

But the scaling endpoint