Wait out a failed lookup of the registry's name in a build, and say it was the resolver (issue 1748)

internal/builder/registry_wait_test.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"crypto/sha256"
 	"encoding/hex"
+	"errors"
 	"net"
 	"net/http"
 	"strings"
@@ -152,3 +153,67 @@ func TestAnImagePushWaitsForARegistryHeldStill(t *testing.T) {
 type errorString string
 
 func (e errorString) Error() string { return string(e) }
+
+// A resolver that does not answer for a moment (novox/hq issue 1748): a lookup of the registry's name that
+// timed out failed the build at once, though the same name resolved a second later.
+
+const lookupTimeout = "Head \"http://novox.internal:5100/v2/x/blobs/sha256:1\": dial tcp: lookup novox.internal on 10.10.0.3:53: " +
+	"read udp 10.10.0.3:44790->10.10.0.3:53: i/o timeout"
+
+type lookupTimedOut struct{}
+
+func (lookupTimedOut) Error() string { return lookupTimeout }
+
+func TestABuildWaitsOutAResolverThatDoesNotAnswerAndGoesOn(t *testing.T) { // 1748/AC1
+	waitingFor(t, 5*time.Second)
+	said := saying(t)
+	calls := 0
+	err := waitForRegistry(context.Background(), "novox.internal:5100", "HEAD /v2/x", func() error {
+		calls++
+		if calls < 3 {
+			return lookupTimedOut{}
+		}
+		return nil
+	})
+	if err != nil {
+		t.Fatalf("a lookup timing out twice failed the build: %v", err)
+	}
+	if calls != 3 {
+		t.Errorf("tried %d times, want 3", calls)
+	}
+	log := strings.Join(said(), "\n")
+	if !strings.Contains(log, "resolver did not answer") || !strings.Contains(log, "answers again") {
+		t.Errorf("the build's log does not say the resolver was waited for, and that it came back:\n%s", log)
+	}
+}
+
+func TestABuildFailsNamingTheResolverWhenItDoesNotAnswerPastTheBound(t *testing.T) { // 1748/AC2
+	waitingFor(t, 100*time.Millisecond)
+	saying(t)
+	started := time.Now()
+	err := waitForRegistry(context.Background(), "novox.internal:5100", "HEAD /v2/x", func() error { return lookupTimedOut{} })
+	if err == nil {
+		t.Fatal("a resolver that never answered let the build go on")
+	}
+	if !strings.Contains(err.Error(), "resolver did not answer") || !strings.Contains(err.Error(), "i/o timeout") {
+		t.Errorf("the failure does not name the resolver: %v", err)
+	}
+	if strings.Contains(err.Error(), "refused every connection") {
+		t.Errorf("the failure blames a refusal: %v", err)
+	}
+	if waited := time.Since(started); waited < 100*time.Millisecond {
+		t.Errorf("failed after %s, before the bound", waited)
+	}
+}
+
+func TestAnErrorThatIsNeitherARefusalNorALookupFailsAtOnce(t *testing.T) { // 1748/AC3
+	waitingFor(t, 5*time.Second)
+	calls := 0
+	err := waitForRegistry(context.Background(), "novox.internal:5100", "HEAD /v2/x", func() error {
+		calls++
+		return errors.New("401 unauthorized")
+	})
+	if err == nil || calls != 1 {
+		t.Fatalf("err %v after %d calls, want a failure at once", err, calls)
+	}
+}