e2e auth breakage it caused. The fix is to not set project/zone/kube_master
to the empty string partway through the script, which I should have
realized was a bad idea in the first place.
This will allow us greated isolation between test case runs without the
need to include randomized uuids to many of the fields in the pods,
services and rcs creation requests.
Tested:
- Ran two instances of `e2e -t TestLivenessHttp -t TestLivenessExec`
simultaneously and confirmed that pod names did not clash and that
both of them completed successfully.
This commit reimplements hack/e2e-suite/liveness.sh in Go as part of cmd/e2e.
Tested by running it on a live cluster:
$ cmd/e2e --host=https://w.x.y.z --provider=gce -t TestLivenessHttp -t TestLivenessExec
I0122 08:12:53.183298 6502 liveness.go:72] Restart count of pod liveness-exec-6f917474-a251-11e4-8cc2-d4ae52bb3eea increased from 0 to 1 during the test
I0122 08:13:23.605471 6502 liveness.go:72] Restart count of pod liveness-http-84d28569-a251-11e4-8cc2-d4ae52bb3eea increased from 0 to 1 during the test
Also ran the full e2e suite including kube-up/kube-down to confirm it works.
Jeff Lowdermilk