Add timeout and retry to CLI log/describe commands in e2e tests

tests/e2e/lib/backup_cli.go

@@ -5,6 +5,7 @@ import (
 	"fmt"
 	"log"
 	"strings"
+	"time"
 
 	velero "github.com/vmware-tanzu/velero/pkg/apis/velero/v1"
 	"k8s.io/apimachinery/pkg/util/wait"
@@ -191,37 +192,73 @@ func IsBackupCompletedSuccessfullyViaCLI(name string) (bool, error) {
 	)
 }
 
-// DescribeBackupViaCLI describes backup using the OADP CLI
+// DescribeBackupViaCLI describes backup using the OADP CLI with default timeout and retry.
+// The timeout prevents the command from hanging when retrieving backup details from object storage.
 func DescribeBackupViaCLI(name string) (backupDescription string) {
+	return DescribeBackupViaCLIWithOptions(name, DefaultCLITimeout, DefaultCLIRetries, DefaultCLIRetryDelay)
+}
+
+// DescribeBackupViaCLIWithOptions describes backup using the OADP CLI with specified timeout and retry options.
+func DescribeBackupViaCLIWithOptions(name string, timeout time.Duration, maxRetries int, retryDelay time.Duration) (backupDescription string) {
 	// Use CLI to describe backup
 	cmd := &CLICommand{
 		Resource: "backup",
 		Action:   "describe",
 		Name:     name,
 		Options:  []string{"--details"},
 	}
-	output, err := cmd.Execute()
+
+	var output []byte
+	var err error
+
+	if maxRetries > 1 {
+		output, err = cmd.ExecuteWithTimeoutAndRetry(timeout, maxRetries, retryDelay)
+	} else {
+		output, err = cmd.ExecuteWithTimeout(timeout)
+	}
+
 	if err != nil {
 		return fmt.Sprintf("could not describe backup via CLI: %v, output: %s", err, string(output))
 	}
 
 	return string(output)
 }
 
-// BackupLogsViaCLI gets backup logs using the OADP CLI
+// BackupLogsViaCLI gets backup logs using the OADP CLI with default timeout and retry.
+// The timeout prevents the command from hanging indefinitely when streaming logs from object storage.
+// Retry logic helps handle transient network issues.
 func BackupLogsViaCLI(name string) (backupLogs string, err error) {
+	return BackupLogsViaCLIWithOptions(name, DefaultCLITimeout, DefaultCLIRetries, DefaultCLIRetryDelay)
+}
+
+// BackupLogsViaCLIWithTimeout gets backup logs using the OADP CLI with a specified timeout (no retry).
+func BackupLogsViaCLIWithTimeout(name string, timeout time.Duration) (backupLogs string, err error) {
+	return BackupLogsViaCLIWithOptions(name, timeout, 1, 0)
+}
+
+// BackupLogsViaCLIWithOptions gets backup logs using the OADP CLI with specified timeout and retry options.
+func BackupLogsViaCLIWithOptions(name string, timeout time.Duration, maxRetries int, retryDelay time.Duration) (backupLogs string, err error) {
 	if name == "" {
 		return "", fmt.Errorf("backup name cannot be empty")
 	}
 
-	// Use CLI to get backup logs
+	// Use CLI to get backup logs with timeout and retry to prevent hanging
 	cmd := &CLICommand{
 		Resource: "backup",
 		Action:   "logs",
 		Name:     name,
 		Options:  []string{},
 	}
-	output, cmdErr := cmd.ExecuteOutput()
+
+	var output []byte
+	var cmdErr error
+
+	if maxRetries > 1 {
+		output, cmdErr = cmd.ExecuteOutputWithTimeoutAndRetry(timeout, maxRetries, retryDelay)
+	} else {
+		output, cmdErr = cmd.ExecuteOutputWithTimeout(timeout)
+	}
+
 	if cmdErr != nil {
 		return "", fmt.Errorf("failed to get backup logs via CLI: %v", cmdErr)
 	}

tests/e2e/lib/cli_common.go

@@ -1,19 +1,31 @@
 package lib
 
 import (
+	"context"
 	"fmt"
 	"log"
 	"os"
 	"os/exec"
 	"path/filepath"
 	"strings"
+	"time"
+)
+
+// Default timeout for CLI commands that may hang (e.g., log streaming)
+const DefaultCLITimeout = 5 * time.Minute
+
+// Default retry settings for CLI commands
+const (
+	DefaultCLIRetries    = 3
+	DefaultCLIRetryDelay = 10 * time.Second
 )
 
 type CLICommand struct {
 	Resource string // "backup" or "restore"
 	Action   string // "create", "get", "delete", etc.
 	Name     string
 	Options  []string
+	Timeout  time.Duration // Optional timeout for commands that may hang
 }
 
 func (c *CLICommand) Execute() ([]byte, error) {
@@ -40,6 +52,103 @@ func (c *CLICommand) ExecuteOutput() ([]byte, error) {
 	return cmd.Output()
 }
 
+// ExecuteWithTimeout executes the CLI command with a timeout.
+// If the timeout is exceeded, the command is killed and an error is returned.
+func (c *CLICommand) ExecuteWithTimeout(timeout time.Duration) ([]byte, error) {
+	args := []string{"oadp", c.Resource, c.Action}
+	if c.Name != "" {
+		args = append(args, c.Name)
+	}
+	args = append(args, c.Options...)
+
+	c.LogCLICommand()
+
+	ctx, cancel := context.WithTimeout(context.Background(), timeout)
+	defer cancel()
+
+	cmd := exec.CommandContext(ctx, "kubectl", args...)
+	output, err := cmd.CombinedOutput()
+
+	if ctx.Err() == context.DeadlineExceeded {
+		return output, fmt.Errorf("command timed out after %v: kubectl %s", timeout, strings.Join(args, " "))
+	}
+
+	return output, err
+}
+
+// ExecuteOutputWithTimeout executes the CLI command with a timeout and returns stdout only.
+// If the timeout is exceeded, the command is killed and an error is returned.
+func (c *CLICommand) ExecuteOutputWithTimeout(timeout time.Duration) ([]byte, error) {
+	args := []string{"oadp", c.Resource, c.Action}
+	if c.Name != "" {
+		args = append(args, c.Name)
+	}
+	args = append(args, c.Options...)
+
+	c.LogCLICommand()
+
+	ctx, cancel := context.WithTimeout(context.Background(), timeout)
+	defer cancel()
+
+	cmd := exec.CommandContext(ctx, "kubectl", args...)
+	output, err := cmd.Output()
+
+	if ctx.Err() == context.DeadlineExceeded {
+		return output, fmt.Errorf("command timed out after %v: kubectl %s", timeout, strings.Join(args, " "))
+	}
+
+	return output, err
+}
+
+// ExecuteOutputWithTimeoutAndRetry executes the CLI command with a timeout and retry logic.
+// It retries the command up to maxRetries times with a delay between attempts.
+// This is useful for commands that may fail due to transient issues (e.g., network problems).
+func (c *CLICommand) ExecuteOutputWithTimeoutAndRetry(timeout time.Duration, maxRetries int, retryDelay time.Duration) ([]byte, error) {
+	var lastErr error
+	var lastOutput []byte
+
+	for attempt := 1; attempt <= maxRetries; attempt++ {
+		output, err := c.ExecuteOutputWithTimeout(timeout)
+		if err == nil {
+			return output, nil
+		}
+
+		lastErr = err
+		lastOutput = output
+
+		if attempt < maxRetries {
+			log.Printf("CLI command failed (attempt %d/%d): %v. Retrying in %v...", attempt, maxRetries, err, retryDelay)
+			time.Sleep(retryDelay)
+		}
+	}
+
+	return lastOutput, fmt.Errorf("CLI command failed after %d attempts: %v", maxRetries, lastErr)
+}
+
+// ExecuteWithTimeoutAndRetry executes the CLI command with a timeout and retry logic.
+// It retries the command up to maxRetries times with a delay between attempts.
+func (c *CLICommand) ExecuteWithTimeoutAndRetry(timeout time.Duration, maxRetries int, retryDelay time.Duration) ([]byte, error) {
+	var lastErr error
+	var lastOutput []byte
+
+	for attempt := 1; attempt <= maxRetries; attempt++ {
+		output, err := c.ExecuteWithTimeout(timeout)
+		if err == nil {
+			return output, nil
+		}
+
+		lastErr = err
+		lastOutput = output
+
+		if attempt < maxRetries {
+			log.Printf("CLI command failed (attempt %d/%d): %v. Retrying in %v...", attempt, maxRetries, err, retryDelay)
+			time.Sleep(retryDelay)
+		}
+	}
+
+	return lastOutput, fmt.Errorf("CLI command failed after %d attempts: %v", maxRetries, lastErr)
+}
+
 func (c *CLICommand) LogCLICommand() {
 	args := []string{"kubectl", "oadp", c.Resource, c.Action}
 	if c.Name != "" {

tests/e2e/lib/restore_cli.go

@@ -5,6 +5,7 @@ import (
 	"fmt"
 	"log"
 	"strings"
+	"time"
 
 	velero "github.com/vmware-tanzu/velero/pkg/apis/velero/v1"
 	"k8s.io/apimachinery/pkg/util/wait"
@@ -166,37 +167,73 @@ func IsRestoreCompletedSuccessfullyViaCLI(name string) (bool, error) {
 	)
 }
 
-// DescribeRestoreViaCLI describes restore using the OADP CLI
+// DescribeRestoreViaCLI describes restore using the OADP CLI with default timeout and retry.
+// The timeout prevents the command from hanging when retrieving restore details from object storage.
 func DescribeRestoreViaCLI(name string) string {
+	return DescribeRestoreViaCLIWithOptions(name, DefaultCLITimeout, DefaultCLIRetries, DefaultCLIRetryDelay)
+}
+
+// DescribeRestoreViaCLIWithOptions describes restore using the OADP CLI with specified timeout and retry options.
+func DescribeRestoreViaCLIWithOptions(name string, timeout time.Duration, maxRetries int, retryDelay time.Duration) string {
 	// Use CLI to describe restore
 	cmd := &CLICommand{
 		Resource: "restore",
 		Action:   "describe",
 		Name:     name,
 		Options:  []string{"--details"},
 	}
-	output, err := cmd.Execute()
+
+	var output []byte
+	var err error
+
+	if maxRetries > 1 {
+		output, err = cmd.ExecuteWithTimeoutAndRetry(timeout, maxRetries, retryDelay)
+	} else {
+		output, err = cmd.ExecuteWithTimeout(timeout)
+	}
+
 	if err != nil {
 		return fmt.Sprintf("could not describe restore via CLI: %v, output: %s", err, string(output))
 	}
 
 	return string(output)
 }
 
-// RestoreLogsViaCLI gets restore logs using the OADP CLI
+// RestoreLogsViaCLI gets restore logs using the OADP CLI with default timeout and retry.
+// The timeout prevents the command from hanging indefinitely when streaming logs from object storage.
+// Retry logic helps handle transient network issues.
 func RestoreLogsViaCLI(name string) (restoreLogs string, err error) {
+	return RestoreLogsViaCLIWithOptions(name, DefaultCLITimeout, DefaultCLIRetries, DefaultCLIRetryDelay)
+}
+
+// RestoreLogsViaCLIWithTimeout gets restore logs using the OADP CLI with a specified timeout (no retry).
+func RestoreLogsViaCLIWithTimeout(name string, timeout time.Duration) (restoreLogs string, err error) {
+	return RestoreLogsViaCLIWithOptions(name, timeout, 1, 0)
+}
+
+// RestoreLogsViaCLIWithOptions gets restore logs using the OADP CLI with specified timeout and retry options.
+func RestoreLogsViaCLIWithOptions(name string, timeout time.Duration, maxRetries int, retryDelay time.Duration) (restoreLogs string, err error) {
 	if name == "" {
 		return "", fmt.Errorf("restore name cannot be empty")
 	}
 
-	// Use CLI to get restore logs
+	// Use CLI to get restore logs with timeout and retry to prevent hanging
 	cmd := &CLICommand{
 		Resource: "restore",
 		Action:   "logs",
 		Name:     name,
 		Options:  []string{},
 	}
-	output, cmdErr := cmd.ExecuteOutput()
+
+	var output []byte
+	var cmdErr error
+
+	if maxRetries > 1 {
+		output, cmdErr = cmd.ExecuteOutputWithTimeoutAndRetry(timeout, maxRetries, retryDelay)
+	} else {
+		output, cmdErr = cmd.ExecuteOutputWithTimeout(timeout)
+	}
+
 	if cmdErr != nil {
 		return "", fmt.Errorf("failed to get restore logs via CLI: %v", cmdErr)
 	}