Created attachment 1471504 [details]
example app 200/503 response rate

Description of problem:
Users reported seemingly random, but quite frequent 503's when accessing applications over routes on starter-ca-central-1 (v3.10.14). The issue may have starter occurring after the recent upgrade to 3.10.14 on July 25th; v3.10.9 Starter clusters do not seem to be affected. The application itself seems to be running properly in all cases - indicating the issue could be caused by the router.
I have induced this by deploying the node.js example app and trying to access the default page every two seconds:

Version-Release number of selected component (if applicable):
Server https://api.starter-ca-central-1.openshift.com:443
openshift v3.10.14
kubernetes v1.10.0+b81c8f8

How reproducible:
appears to be consistently reproducible

Steps to Reproduce:
1. Deploy a new app on starter-ca-central-1 (or use an already existing one)
2. Expose a service and wait for the route to be admitted by the router
3. Hit the route repeatedly

Actual results:
At least some 503s while the app is running properly

Expected results:
Consistent response, same as when hitting the service from within the cluster