Retry the initial boot using p-retry

Author: matt-aitken

apps/webapp/app/services/dataStores/organizationDataStoresRegistry.server.ts

@@ -13,7 +13,12 @@ export class OrganizationDataStoresRegistry {
   private _loaded = false;
   private _readyResolve!: () => void;
 
-  /** Resolves once the initial `loadFromDatabase()` completes successfully. */
+  /**
+   * Resolves once the initial `loadFromDatabase()` completes successfully.
+   * At process startup the singleton loads the registry with unbounded retries
+   * (exponential backoff, capped delay) until Postgres is reachable; until then
+   * this promise stays pending and callers that await readiness will block.
+   */
   readonly isReady: Promise<void>;
 
   constructor(prisma: PrismaClient | PrismaReplicaClient) {

apps/webapp/app/services/dataStores/organizationDataStoresRegistryInstance.server.ts

@@ -1,14 +1,32 @@
+import pRetry from "p-retry";
 import { $replica } from "~/db.server";
 import { env } from "~/env.server";
+import { logger } from "~/services/logger.server";
 import { signalsEmitter } from "~/services/signals.server";
 import { singleton } from "~/utils/singleton";
 import { OrganizationDataStoresRegistry } from "./organizationDataStoresRegistry.server";
 
 export const organizationDataStoresRegistry = singleton("organizationDataStoresRegistry", () => {
   const registry = new OrganizationDataStoresRegistry($replica);
 
-  registry.loadFromDatabase().catch((err) => {
-    console.error("[OrganizationDataStoresRegistry] Failed to initialize", err);
+  // Runs as soon as this singleton is created (first import of this module). The
+  // registry’s `isReady` promise resolves when this eventually succeeds.
+  const startupLoadPromise = pRetry(() => registry.loadFromDatabase(), {
+    forever: true,
+    retries: 10,
+    minTimeout: 1_000,
+    maxTimeout: 60_000,
+    factor: 2,
+    onFailedAttempt: (error) => {
+      logger.warn("[OrganizationDataStoresRegistry] Startup load failed, retrying", {
+        attemptNumber: error.attemptNumber,
+        retriesLeft: error.retriesLeft,
+        error: error.message,
+      });
+    },
+  });
+  startupLoadPromise.catch((err) => {
+    console.error("[OrganizationDataStoresRegistry] Unexpected startup load failure", err);
   });
 
   const interval = setInterval(() => {