Add configurable exponential backoff to Agent auto-auth (#10964)

changelog/10964.txt

@@ -0,0 +1,5 @@
+```release-note:changes
+agent: Failed auto-auth attempts are now throttled by an exponential backoff instead of the
+~2 second retry delay. The maximum backoff may be configured with the new `max_backoff` parameter,
+which defaults to 5 minutes.
+```

command/agent.go

@@ -575,6 +575,7 @@ func (c *AgentCommand) Run(args []string) int {
 			Logger:                       c.logger.Named("auth.handler"),
 			Client:                       c.client,
 			WrapTTL:                      config.AutoAuth.Method.WrapTTL,
+			MaxBackoff:                   config.AutoAuth.Method.MaxBackoff,
 			EnableReauthOnNewCredentials: config.AutoAuth.EnableReauthOnNewCredentials,
 			EnableTemplateTokenCh:        enableTokenCh,
 		})

command/agent/auth/auth.go

@@ -8,11 +8,16 @@ import (
 	"net/http"
 	"time"
 
-	hclog "github.com/hashicorp/go-hclog"
+	"github.com/hashicorp/go-hclog"
 	"github.com/hashicorp/vault/api"
 	"github.com/hashicorp/vault/sdk/helper/jsonutil"
 )
 
+const (
+	initialBackoff    = 1 * time.Second
+	defaultMaxBackoff = 5 * time.Minute
+)
+
 // AuthMethod is the interface that auto-auth methods implement for the agent
 // to use.
 type AuthMethod interface {
@@ -48,6 +53,7 @@ type AuthHandler struct {
 	client                       *api.Client
 	random                       *rand.Rand
 	wrapTTL                      time.Duration
+	maxBackoff                   time.Duration
 	enableReauthOnNewCredentials bool
 	enableTemplateTokenCh        bool
 }
@@ -56,6 +62,7 @@ type AuthHandlerConfig struct {
 	Logger                       hclog.Logger
 	Client                       *api.Client
 	WrapTTL                      time.Duration
+	MaxBackoff                   time.Duration
 	Token                        string
 	EnableReauthOnNewCredentials bool
 	EnableTemplateTokenCh        bool
@@ -72,6 +79,7 @@ func NewAuthHandler(conf *AuthHandlerConfig) *AuthHandler {
 		client:                       conf.Client,
 		random:                       rand.New(rand.NewSource(int64(time.Now().Nanosecond()))),
 		wrapTTL:                      conf.WrapTTL,
+		maxBackoff:                   conf.MaxBackoff,
 		enableReauthOnNewCredentials: conf.EnableReauthOnNewCredentials,
 		enableTemplateTokenCh:        conf.EnableTemplateTokenCh,
 	}
@@ -91,6 +99,13 @@ func (ah *AuthHandler) Run(ctx context.Context, am AuthMethod) error {
 		return errors.New("auth handler: nil auth method")
 	}
 
+	backoff := initialBackoff
+	maxBackoff := defaultMaxBackoff
+
+	if ah.maxBackoff > 0 {
+		maxBackoff = ah.maxBackoff
+	}
+
 	ah.logger.Info("starting auth handler")
 	defer func() {
 		am.Shutdown()
@@ -130,8 +145,7 @@ func (ah *AuthHandler) Run(ctx context.Context, am AuthMethod) error {
 		default:
 		}
 
-		// Create a fresh backoff value
-		backoff := 2*time.Second + time.Duration(ah.random.Int63()%int64(time.Second*2)-int64(time.Second))
+		backoff = calculateBackoff(backoff, maxBackoff)
 
 		var clientToUse *api.Client
 		var err error
@@ -311,3 +325,16 @@ func (ah *AuthHandler) Run(ctx context.Context, am AuthMethod) error {
 		}
 	}
 }
+
+// calculateBackoff determines a new backoff duration that is roughly twice
+// the previous value, capped to a max value, with a measure of randomness.
+func calculateBackoff(previous, max time.Duration) time.Duration {
+	maxBackoff := 2 * previous
+	if maxBackoff > max {
+		maxBackoff = max
+	}
+
+	// Trim a random amount (0-25%) off the doubled duration
+	trim := rand.Int63n(int64(maxBackoff) / 4)
+	return maxBackoff - time.Duration(trim)
+}

command/agent/auth/auth_test.go

@@ -6,7 +6,7 @@ import (
 	"testing"
 	"time"
 
-	hclog "github.com/hashicorp/go-hclog"
+	"github.com/hashicorp/go-hclog"
 	"github.com/hashicorp/vault/api"
 	"github.com/hashicorp/vault/builtin/credential/userpass"
 	vaulthttp "github.com/hashicorp/vault/http"
@@ -106,3 +106,42 @@ consumption:
 		}
 	}
 }
+
+func TestCalculateBackoff(t *testing.T) {
+	tests := []struct {
+		previous time.Duration
+		max      time.Duration
+		expMin   time.Duration
+		expMax   time.Duration
+	}{
+		{
+			1000 * time.Millisecond,
+			60000 * time.Millisecond,
+			1500 * time.Millisecond,
+			2000 * time.Millisecond,
+		},
+		{
+			1000 * time.Millisecond,
+			5000 * time.Millisecond,
+			1500 * time.Millisecond,
+			2000 * time.Millisecond,
+		},
+		{
+			4000 * time.Millisecond,
+			5000 * time.Millisecond,
+			3750 * time.Millisecond,
+			5000 * time.Millisecond,
+		},
+	}
+
+	for _, test := range tests {
+		for i := 0; i < 100; i++ {
+			backoff := calculateBackoff(test.previous, test.max)
+
+			// Verify that the new backoff is 75-100% of 2*previous, but <= than the max
+			if backoff < test.expMin || backoff > test.expMax {
+				t.Fatalf("expected backoff in range %v to %v, got: %v", test.expMin, test.expMax, backoff)
+			}
+		}
+	}
+}

command/agent/config/config.go

@@ -66,6 +66,8 @@ type Method struct {
 	MountPath     string        `hcl:"mount_path"`
 	WrapTTLRaw    interface{}   `hcl:"wrap_ttl"`
 	WrapTTL       time.Duration `hcl:"-"`
+	MaxBackoffRaw interface{}   `hcl:"max_backoff"`
+	MaxBackoff    time.Duration `hcl:"-"`
 	Namespace     string        `hcl:"namespace"`
 	Config        map[string]interface{}
 }
@@ -358,6 +360,14 @@ func parseAutoAuth(result *Config, list *ast.ObjectList) error {
 		}
 	}
 
+	if result.AutoAuth.Method.MaxBackoffRaw != nil {
+		var err error
+		if result.AutoAuth.Method.MaxBackoff, err = parseutil.ParseDurationSecond(result.AutoAuth.Method.MaxBackoffRaw); err != nil {
+			return err
+		}
+		result.AutoAuth.Method.MaxBackoffRaw = nil
+	}
+
 	return nil
 }
 

command/agent/config/config_test.go

@@ -126,6 +126,7 @@ func TestLoadConfigFile(t *testing.T) {
 				Config: map[string]interface{}{
 					"role": "foobar",
 				},
+				MaxBackoff: 0,
 			},
 			Sinks: []*Sink{
 				{
@@ -181,6 +182,7 @@ func TestLoadConfigFile_Method_Wrapping(t *testing.T) {
 				Type:       "aws",
 				MountPath:  "auth/aws",
 				WrapTTL:    5 * time.Minute,
+				MaxBackoff: 2 * time.Minute,
 				Config: map[string]interface{}{
 					"role": "foobar",
 				},

command/agent/config/test-fixtures/config-method-wrapping.hcl

@@ -7,6 +7,7 @@ auto_auth {
 		config = {
 			role = "foobar"
 		}
+		max_backoff = "2m"
 	}
 
 	sink {

website/content/docs/agent/autoauth/index.mdx

@@ -20,9 +20,8 @@ are locations where the agent should write a token any time the current token
 value has changed.
 
 When the agent is started with Auto-Auth enabled, it will attempt to acquire a
-Vault token using the configured Method. On failure, it will back off for a
-short while (including some randomness to help prevent thundering herd
-scenarios) and retry. On success, unless the auth method is configured to wrap
+Vault token using the configured Method. On failure, it will exponentially back
+off and then retry. On success, unless the auth method is configured to wrap
 the tokens, it will keep the resulting token renewed until renewal is no longer
 allowed or fails, at which point it will attempt to reauthenticate.
 
@@ -128,6 +127,10 @@ These are common configuration values that live within the `method` block:
   structure. Values can be an integer number of seconds or a stringish value
   like `5m`.
 
+- `max_backoff` `(string or integer: "5m")` - The maximum time Agent will delay
+  before retrying after a failed auth attempt. The backoff will start at 1 second
+  and double (with some randomness) after successive failures, capped by `max_backoff.`
+
 - `config` `(object: required)` - Configuration of the method itself. See the
   sidebar for information about each method.