feat(webhooks): add exponential backoff with jitter to delivery retries

src/appConfiguration.ts

@@ -30,6 +30,7 @@
   chaosTargets: string[];
   chaosProbability: number;
   circuitBreaker: CircuitBreakerConfig;
+  webhookRetry: WebhookRetryConfig;
   /**
    * Per-provider circuit-breaker configuration for outbound webhook delivery.
    * Thresholds are intentionally separate from the RPC circuit breaker so
@@ -81,7 +82,7 @@
     .filter(Boolean);
 }
 
 function parseAssets(value: string | undefined): string[] {
   if (!value) {
     return ['USDC', 'XLM', 'BTC', 'ETH']; // Default assets
   }
@@ -117,6 +118,13 @@
       successThreshold: clamp(toNumber(env.CB_SUCCESS_THRESHOLD, 1), 1, 20),
       timeoutMs: clamp(toNumber(env.CB_TIMEOUT_MS, 30_000), 1_000, 300_000),
     },
+    webhookRetry: {
+      maxAttempts: clamp(toNumber(env.WEBHOOK_RETRY_MAX_ATTEMPTS, 5), 1, 20),
+      initialDelayMs: clamp(toNumber(env.WEBHOOK_RETRY_INITIAL_DELAY_MS, 1_000), 100, 60_000),
+      maxDelayMs: clamp(toNumber(env.WEBHOOK_RETRY_MAX_DELAY_MS, 30_000), 1_000, 600_000),
+      multiplier: clamp(toNumber(env.WEBHOOK_RETRY_MULTIPLIER, 2), 1, 10),
+      jitterFactor: clamp(toNumber(env.WEBHOOK_RETRY_JITTER_FACTOR, 0.1), 0, 1),
+    },
     webhookCircuitBreaker: {
       failureThreshold: clamp(toNumber(env.WEBHOOK_CB_FAILURE_THRESHOLD, 5), 1, 100),
       successThreshold: clamp(toNumber(env.WEBHOOK_CB_SUCCESS_THRESHOLD, 1), 1, 20),

src/webhookDelivery.test.ts

@@ -40,13 +40,28 @@ function makeRegistry() {
 function makeService(
   registry: Registry,
   breakerConfig: { failureThreshold?: number; successThreshold?: number; timeoutMs?: number } = {},
+  retryConfigOrCallback?: Partial<WebhookRetryConfig> | ((entry: DLQEntry) => Promise<void> | void),
+  dlqCallback?: (entry: DLQEntry) => Promise<void> | void,
 ) {
-  return new WebhookDeliveryService(registry, {
-    failureThreshold: 3,
-    successThreshold: 1,
-    timeoutMs: 50, // short cooldown so HALF_OPEN tests don't need real timers
-    ...breakerConfig,
-  });
+  const isRetryConfig =
+    typeof retryConfigOrCallback === 'object' &&
+    retryConfigOrCallback !== null &&
+    ('maxAttempts' in retryConfigOrCallback || 'initialDelayMs' in retryConfigOrCallback);
+
+  const retryConfig = isRetryConfig ? (retryConfigOrCallback as Partial<WebhookRetryConfig>) : undefined;
+  const callback = isRetryConfig ? dlqCallback : (retryConfigOrCallback as ((entry: DLQEntry) => Promise<void> | void) | undefined);
+
+  return new WebhookDeliveryService(
+    registry,
+    {
+      failureThreshold: 3,
+      successThreshold: 1,
+      timeoutMs: 50,
+      ...breakerConfig,
+    },
+    retryConfig,
+    callback,
+  );
 }
 
 const basePayload: DeliveryPayload = {