Simple leader election using Leases.
Elector elects a leader by acquiring a Lease with OwnerReference set to the owning pod.
This ensures that only a single leader exists at any given time. As long as the leader pod exists, it will continue to be leader. This means that a leader in CrashLoopBackoff state will still be leader.
This choice is made on the basis that it is better with no leader than two leaders.
Users should make sure to have alerts to detect when a leader is stuck.
Elector has two API endpoints on the election port for getting information about the currently elected leader. The endpoints return the same information, but one is a simple JSON object and the other is a Server Sent Events stream.
The object returned looks like this:
{
"name": "pod-name",
"last_update": "timestamp of last update"
}Simple GET with immediate return of the described object.
The SSE API is a stream of server sent events that will send a message whenever there is an update. Each event will be a JSON object as described above.
Default election port is 6060 (override with --http).
Metrics are available on port 9090 (override with --metrics-address).
Probes are available on port 8080 (override with --probe-address).
Some of the tests uses envtest to simulate a kubernetes cluster.
Using the test target in make will configure envtest for you before running the tests.
If you would rather control the setup of envtest yourself, use the setup-envtest command to install and configure envtest.
go run sigs.k8s.io/controller-runtime/tools/setup-envtest list # Get list of supported versions
source <(go run sigs.k8s.io/controller-runtime/tools/setup-envtest use -p env ${SUPPORTED_K8S_VERSION}) # Activate selected version in current shell