fix(fleet): More flaky mitigations again (#32495)

DataDog · Dec 24, 2024 · f93202d · f93202d
1 parent 00ecffe
commit f93202d
Show file tree

Hide file tree

Showing 3 changed files with 18 additions and 11 deletions.
diff --git a/pkg/fleet/installer/packages/docker.go b/pkg/fleet/installer/packages/docker.go
@@ -148,7 +148,8 @@ func (a *apmInjectorInstaller) verifyDockerRuntime(ctx context.Context) (err err
 		return nil
 	}
 
-	for i := 0; i < 3; i++ {
+	currentRuntime := ""
+	for i := 0; i < 5; i++ {
 		if i > 0 {
 			time.Sleep(time.Second)
 		}
@@ -157,17 +158,22 @@ func (a *apmInjectorInstaller) verifyDockerRuntime(ctx context.Context) (err err
 		cmd.Stdout = &outb
 		err = cmd.Run()
 		if err != nil {
-			if i < 2 {
+			if i < 5 {
 				log.Debug("failed to verify docker runtime, retrying: ", err)
 			} else {
 				log.Warn("failed to verify docker runtime: ", err)
 			}
 		}
 		if strings.TrimSpace(outb.String()) == "dd-shim" {
+			span.SetTag("retries", i)
+			span.SetTag("docker_runtime", "dd-shim")
 			return nil
 		}
+		currentRuntime = strings.TrimSpace(outb.String())
 	}
-	err = fmt.Errorf("docker default runtime has not been set to injector docker runtime")
+	span.SetTag("retries", 5)
+	span.SetTag("docker_runtime", currentRuntime)
+	err = fmt.Errorf("docker default runtime has not been set to injector docker runtime (is \"%s\")", currentRuntime)
 	return err
 }
 

diff --git a/test/new-e2e/tests/installer/host/host.go b/test/new-e2e/tests/installer/host/host.go
@@ -79,7 +79,14 @@ func (h *Host) setSystemdVersion() {
 
 // InstallDocker installs Docker on the host if it is not already installed.
 func (h *Host) InstallDocker() {
-	defer func() { h.remote.MustExecute("sudo systemctl start docker") }()
+	defer func() {
+		_, err := h.remote.Execute("sudo systemctl reset-failed docker")
+		if err != nil {
+			h.t.Logf("warn: failed to reset-failed for docker.d: %v", err)
+		}
+		_, err = h.remote.Execute("sudo systemctl start docker")
+		require.NoErrorf(h.t, err, "failed to start Docker, logs: %s", h.remote.MustExecute("sudo journalctl -xeu docker"))
+	}()
 	if _, err := h.remote.Execute("command -v docker"); err == nil {
 		return
 	}

diff --git a/test/new-e2e/tests/installer/unix/all_packages_test.go b/test/new-e2e/tests/installer/unix/all_packages_test.go
@@ -18,7 +18,6 @@ import (
 	"github.com/DataDog/test-infra-definitions/scenarios/aws/ec2"
 	"github.com/stretchr/testify/require"
 
-	"github.com/DataDog/datadog-agent/pkg/util/testutil/flake"
 	"github.com/DataDog/datadog-agent/test/new-e2e/pkg/e2e"
 	"github.com/DataDog/datadog-agent/test/new-e2e/pkg/environments"
 	awshost "github.com/DataDog/datadog-agent/test/new-e2e/pkg/provisioners/aws/host"
@@ -40,7 +39,7 @@ var (
 		e2eos.AmazonLinux2,
 		e2eos.Debian12,
 		e2eos.RedHat9,
-		e2eos.FedoraDefault,
+		// e2eos.FedoraDefault, // Skipped instead of marked as flaky to avoid useless logs
 		e2eos.CentOS7,
 		e2eos.Suse15,
 	}
@@ -108,11 +107,6 @@ func TestPackages(t *testing.T) {
 			suite := test.t(flavor, flavor.Architecture, method)
 			t.Run(suite.Name(), func(t *testing.T) {
 				t.Parallel()
-				// FIXME: Fedora currently has DNS issues
-				if flavor.Flavor == e2eos.Fedora {
-					flake.Mark(t)
-				}
-
 				opts := []awshost.ProvisionerOption{
 					awshost.WithEC2InstanceOptions(ec2.WithOSArch(flavor, flavor.Architecture)),
 					awshost.WithoutAgent(),