add: waitForDomain func to wait for a domain to be resolvable before pinging lets-encrypt

pkg/controller/tls/certs.go

@@ -5,6 +5,7 @@ import (
 	"crypto/x509"
 	"encoding/pem"
 	"fmt"
+	"net"
 	"strings"
 	"time"
 
@@ -49,6 +50,24 @@ func RequireSecretTypeTLS(h router.Handler) router.Handler {
 	})
 }
 
+func WaitForDomain(domain string, retryInterval time.Duration, maxRetries int) error {
+	done := make(chan error)
+
+	go func() {
+		for retries := 0; retries < maxRetries; retries++ {
+			ips, err := net.LookupIP(domain)
+			if err == nil && len(ips) > 0 {
+				done <- nil // Domain is resolvable
+				return
+			}
+			time.Sleep(retryInterval)
+		}
+		done <- fmt.Errorf("domain %s is not resolvable after %d retries", domain, maxRetries) // Domain is not resolvable
+	}()
+
+	return <-done
+}
+
 // RenewCert handles the renewal of existing TLS certificates
 func RenewCert(req router.Request, resp router.Response) error {
 	sec := req.Object.(*corev1.Secret)
@@ -82,6 +101,11 @@ func RenewCert(req router.Request, resp router.Response) error {
 
 		logrus.Infof("Renewing TLS cert for %s", domain)
 
+		if err := WaitForDomain(domain, 5*time.Second, 12); err != nil {
+			logrus.Warnf("Domain %s is not resolvable, skipping certificate renewal: %v", domain, err)
+			return
+		}
+
 		// Get new certificate
 		cert, err := leUser.getCert(req.Ctx, domain)
 		if err != nil {

pkg/publish/cert.go

@@ -5,11 +5,14 @@ import (
 	"encoding/pem"
 	"fmt"
 	"sort"
+	"sync"
+	"time"
 
 	"github.com/acorn-io/baaah/pkg/name"
 	"github.com/acorn-io/baaah/pkg/router"
 	"github.com/acorn-io/baaah/pkg/typed"
 	v1 "github.com/acorn-io/runtime/pkg/apis/internal.acorn.io/v1"
+	"github.com/acorn-io/runtime/pkg/controller/tls"
 	"github.com/acorn-io/runtime/pkg/labels"
 	"github.com/acorn-io/runtime/pkg/system"
 	"github.com/sirupsen/logrus"
@@ -155,6 +158,25 @@ func setupCertsForRules(req router.Request, svc *v1.ServiceInstance, rules []net
 		ingressTLS = setupCertManager(svc.Name, rules)
 	}
 
+	// Best effort to wait for all domains to be ready, so we don't spam Let's Encrypt
+	// with requests for domains where the DNS entry was not yet propagated
+	hostsSeen := map[string]struct{}{}
+	wg := sync.WaitGroup{}
+	for _, rule := range rules {
+		if _, ok := hostsSeen[rule.Host]; ok {
+			continue
+		}
+		hostsSeen[rule.Host] = struct{}{}
+		wg.Add(1)
+		go func(host string) {
+			err := tls.WaitForDomain(host, 5*time.Second, 6)
+			if err != nil {
+				logrus.Debugln(err)
+			}
+			wg.Done()
+		}(rule.Host)
+	}
+	wg.Wait()
 	return secrets, ingressTLS, annotations, nil
 }