Earlier this week I moved the Jenkins jobs performing our builds of master, release-1.0, and release-1.1 (which run hack/jenkins/build.sh) from the Jenkins master to a Jenkins slave.

Immediately afterwards, the scalability tests started failing:

Density [Skipped] [Performance suite] should allow starting 30 pods per node
/go/src/k8s.io/kubernetes/_output/dockerized/go/src/k8s.io/kubernetes/test/e2e/density.go:111
There should be no high-latency requests
Expected
    <int>: 1
not to be >
    <int>: 0

Load capacity [Skipped] [Performance suite] should be able to handle 30 pods per node
/go/src/k8s.io/kubernetes/_output/dockerized/go/src/k8s.io/kubernetes/test/e2e/load.go:65
Expected
    <int>: 3
not to be >
    <int>: 0

When I moved the builds back to the Jenkins master, the scalability tests started passing again.

This doesn't match my understanding of our build jobs: since they run in Docker, they should be hermetic, producing the same outputs, but apparently something is different.

cc @kubernetes/goog-testing @zmerlynn @wojtek-t