Merge pull request spegel-org#496 from spegel-org/fix/bootstrap-old-l…

…ease

Fix p2p bootstrap to run on failed readiness check

CHANGELOG.md

@@ -34,6 +34,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - [#460](https://github.com/spegel-org/spegel/pull/460) Fix environment variable for http-bootstrap-addr flag.
 - [#471](https://github.com/spegel-org/spegel/pull/471) Fix handler key in request logging.
 - [#491](https://github.com/spegel-org/spegel/pull/491) Fix so that resolve timeout does not cancel mirroring attempts.
+- [#496](https://github.com/spegel-org/spegel/pull/496) Fix p2p bootstrap to run on failed readiness check.
 
 ### Security
 

charts/spegel/templates/daemonset.yaml

@@ -120,7 +120,7 @@ spec:
         # Startup may take a bit longer on bootsrap as Pods need to find each other.
         # This is why the startup proben is a bit more forgiving, while hitting the endpoint more often.
         startupProbe:
-          periodSeconds: 1
+          periodSeconds: 3
           failureThreshold: 60
           httpGet:
             path: /healthz

pkg/registry/registry.go

@@ -153,8 +153,8 @@ func (r *Registry) handle(rw mux.ResponseWriter, req *http.Request) {
 	rw.WriteHeader(http.StatusNotFound)
 }
 
-func (r *Registry) readyHandler(rw mux.ResponseWriter, _ *http.Request) {
-	ok, err := r.router.Ready()
+func (r *Registry) readyHandler(rw mux.ResponseWriter, req *http.Request) {
+	ok, err := r.router.Ready(req.Context())
 	if err != nil {
 		rw.WriteError(http.StatusInternalServerError, fmt.Errorf("could not determine router readiness: %w", err))
 		return

pkg/routing/mock.go

@@ -19,7 +19,7 @@ func NewMockRouter(resolver map[string][]netip.AddrPort, self netip.AddrPort) *M
 	}
 }
 
-func (m *MockRouter) Ready() (bool, error) {
+func (m *MockRouter) Ready(ctx context.Context) (bool, error) {
 	m.mx.RLock()
 	defer m.mx.RUnlock()
 	return len(m.resolver) > 0, nil

pkg/routing/p2p.go

@@ -135,7 +135,7 @@ func (r *P2PRouter) Close() error {
 	return r.host.Close()
 }
 
-func (r *P2PRouter) Ready() (bool, error) {
+func (r *P2PRouter) Ready(ctx context.Context) (bool, error) {
 	addrInfo, err := r.bootstrapper.Get()
 	if err != nil {
 		return false, err
@@ -144,6 +144,10 @@ func (r *P2PRouter) Ready() (bool, error) {
 		return true, nil
 	}
 	if r.kdht.RoutingTable().Size() == 0 {
+		err := r.kdht.Bootstrap(ctx)
+		if err != nil {
+			return false, err
+		}
 		return false, nil
 	}
 	return true, nil

pkg/routing/routing.go

@@ -6,7 +6,7 @@ import (
 )
 
 type Router interface {
-	Ready() (bool, error)
+	Ready(ctx context.Context) (bool, error)
 	Resolve(ctx context.Context, key string, allowSelf bool, count int) (<-chan netip.AddrPort, error)
 	Advertise(ctx context.Context, keys []string) error
 }

test/e2e/e2e.sh

@@ -39,8 +39,6 @@ else
 	kubectl --kubeconfig $KIND_KUBECONFIG --namespace nginx delete deployments --all
 	kubectl --kubeconfig $KIND_KUBECONFIG --namespace conformance delete jobs --all
 	helm --kubeconfig $KIND_KUBECONFIG uninstall --ignore-not-found --namespace spegel spegel
-	# Delete lease due to bug causing forcing us to wait for Spegel to fails once.
-	kubectl --kubeconfig $KIND_KUBECONFIG --namespace spegel delete lease spegel-leader-election
 
 	# Delete test images from all expect one node
 	for NODE in control-plane worker2 worker3 worker4