These scale tests are meant to be run against a large kubernetes cluster. We run
these tests in CI on GKE clusters with 100 n1-standard-8 nodes with ip
aliasing
and network policy
enabled.
After deploying an appropriate size GKE cluster, CI will deploy cf-for-k8s and push 2000 app instances (1000 apps, 2 instances per app) with 1000 routes (1 route per app).
The steady state test runs once the environment has been set up with 2000 app instances and 1000 routes. This test is "steady state" because it keeps the number of routes constant at 1000, by deleting one route every time it maps a new route. We chose to use a steady state test because we want to keep the number of routes constant to measure the control plane latency under the desired load.
For each route the test maps, the test measures the latency from when the route
is mapped until when that route is reachable. The test asserts that this latency
is under 10 seconds for 95% of the map-route requests.
Currently we run scale tests as defined in the scaling pipeline.