fix(login): replace custom SWR cache with lru-cache (#11945)

# Which Problems Are Solved

The in-memory SWR promiseCache in zitadel.ts used keys like
getBrandingSettings-${org} that had no instance identifier. In
multi-tenant mode, where a single Next.js process serves multiple
instances, an instance A's cached settings without org context
(branding, languages, settings, etc.) could be served to Instance B
request without org context. The cache now additionally is bound to 100
entries by default (configurable via maxSize in `API_CACHE_CONFIG`).
When capacity is exceeded, expired entries are swept first

It additionally replaces the hand-rolled PromiseCache (Map-based, FIFO
eviction) with lru-cache, leveraging its built-in fetchMethod for
stale-while-revalidate, request deduplication, and true LRU eviction.

# How the Problems Are Solved

Added an instanceCacheKey() helpe that prefixes every cache key with
serviceConfig.instanceHost in addition to the org context if available.

---------

Co-authored-by: Ramon <[email protected]>

apps/login/next-env-vars.d.ts

@@ -85,12 +85,13 @@ declare namespace NodeJS {
     API_CACHE_ENABLED?: string;
 
     /**
-     * Optional: JSON string to configure the cache TTLs (in minutes) for specific backend API routes or global fallbacks.
+     * Optional: JSON string to configure the cache TTLs (in minutes) and size limits for specific backend API routes or global fallbacks.
-     * Example: '{"defaultMinutes": 15, "longMinutes": 60, "getBrandingSettings": 120}'
+     * Example: '{"defaultMinutes": 15, "longMinutes": 60, "maxSize": 200, "getBrandingSettings": 120}'
-     *
+     * 
      * Properties:
      * - \`defaultMinutes\`: The globally utilized default TTL in minutes (falls back to 15 if not set).
      * - \`longMinutes\`: The TTL utilized string for long-cached routes, like branding/translation (falls back to 60 if not set).
+     * - \`maxSize\`: Maximum number of entries the in-memory cache may hold. Oldest entries are evicted when capacity is reached (defaults to 100).
      * - \`[route_name]\`: Explicit overrides per specific API method (e.g., \`getHostedLoginTranslation\`).
      */
     API_CACHE_CONFIG?: string;

apps/login/package.json

@@ -41,6 +41,7 @@
     "copy-to-clipboard": "^3.3.3",
     "deepmerge": "^4.3.1",
     "escape-html": "^1.0.3",
+    "lru-cache": "^11.3.2",
     "lucide-react": "^0.577.0",
     "moment": "^2.30.1",
     "next": "16.2.2",

apps/login/src/lib/cache.test.ts

@@ -0,0 +1,251 @@
+import { afterEach, describe, expect, test, vi } from "vitest";
+import { PromiseCache } from "./cache";
+
+// Suppress logger output during tests
+vi.mock("./logger", () => ({
+  createLogger: () => ({
+    warn: vi.fn(),
+    info: vi.fn(),
+    error: vi.fn(),
+    debug: vi.fn(),
+  }),
+}));
+
+/**
+ * Creates a mock clock for lru-cache's `perf` option.
+ * lru-cache uses `performance.now()` internally for TTL tracking, and caches
+ * the reference at import time. The `perf` option lets us inject a custom clock.
+ */
+function mockClock(initial = 1000) {
+  let now = initial;
+  return {
+    perf: { now: () => now },
+    advance: (ms: number) => {
+      now += ms;
+    },
+  };
+}
+
+describe("PromiseCache", () => {
+  let cache: PromiseCache;
+
+  afterEach(() => {
+    cache?.clear();
+  });
+
+  describe("getOrFetch", () => {
+    test("should return the fetcher result on cache miss", async () => {
+      cache = new PromiseCache(10);
+      const result = await cache.getOrFetch("key1", () => Promise.resolve("value1"), 60_000);
+      expect(result).toBe("value1");
+      expect(cache.size).toBe(1);
+    });
+
+    test("should return cached value on cache hit", async () => {
+      cache = new PromiseCache(10);
+      let callCount = 0;
+      const fetcher = () => {
+        callCount++;
+        return Promise.resolve(`value-${callCount}`);
+      };
+
+      const first = await cache.getOrFetch("key1", fetcher, 60_000);
+      const second = await cache.getOrFetch("key1", fetcher, 60_000);
+
+      expect(first).toBe("value-1");
+      expect(second).toBe("value-1");
+      expect(callCount).toBe(1);
+    });
+
+    test("should return stale value immediately after TTL expires (SWR)", async () => {
+      const clock = mockClock();
+      cache = new PromiseCache(10, clock.perf);
+
+      let callCount = 0;
+      let resolveRevalidation: ((v: string) => void) | undefined;
+
+      const fetcher = () => {
+        callCount++;
+        if (callCount === 1) {
+          return Promise.resolve("value-1");
+        }
+        // Second call: return a promise we control
+        return new Promise<string>((resolve) => {
+          resolveRevalidation = resolve;
+        });
+      };
+
+      // First call — blocks on fetch
+      const first = await cache.getOrFetch("key1", fetcher, 100);
+      expect(first).toBe("value-1");
+      expect(callCount).toBe(1);
+
+      // Expire the entry
+      clock.advance(101);
+
+      // Second call after expiry — should get stale value immediately
+      const second = await cache.getOrFetch("key1", fetcher, 100);
+      expect(second).toBe("value-1"); // stale value, not blocking
+      expect(callCount).toBe(2); // revalidation was triggered
+
+      // Third call while revalidation is in-flight — also gets stale value
+      const third = await cache.getOrFetch("key1", fetcher, 100);
+      expect(third).toBe("value-1"); // stale value
+      expect(callCount).toBe(2); // no additional fetch (already revalidating)
+
+      // Resolve the background revalidation and flush the .then() chain
+      resolveRevalidation!("value-2");
+      await new Promise((r) => setTimeout(r, 0));
+      await new Promise((r) => setTimeout(r, 0));
+
+      // Now we should get the fresh value
+      const fourth = await cache.getOrFetch("key1", fetcher, 100);
+      expect(fourth).toBe("value-2");
+    });
+
+    test("should keep stale value when revalidation fails", async () => {
+      const clock = mockClock();
+      cache = new PromiseCache(10, clock.perf);
+
+      let callCount = 0;
+
+      const fetcher = () => {
+        callCount++;
+        if (callCount === 1) {
+          return Promise.resolve("value-1");
+        }
+        return Promise.reject(new Error("network error"));
+      };
+
+      const first = await cache.getOrFetch("key1", fetcher, 100);
+      expect(first).toBe("value-1");
+
+      clock.advance(101);
+
+      // After TTL: returns stale, triggers failing revalidation
+      const second = await cache.getOrFetch("key1", fetcher, 100);
+      expect(second).toBe("value-1");
+      expect(callCount).toBe(2);
+
+      // Flush the rejected promise's handlers
+      await new Promise((r) => setTimeout(r, 0));
+      await new Promise((r) => setTimeout(r, 0));
+
+      // Next call can retry
+      clock.advance(1);
+      const third = await cache.getOrFetch("key1", fetcher, 100);
+      expect(third).toBe("value-1"); // still stale
+      expect(callCount).toBe(3); // retried
+    });
+
+    test("should re-fetch after entry is evicted", async () => {
+      cache = new PromiseCache(10);
+      let callCount = 0;
+      const fetcher = () => {
+        callCount++;
+        return Promise.resolve(`value-${callCount}`);
+      };
+
+      const first = await cache.getOrFetch("key1", fetcher, 60_000);
+      expect(first).toBe("value-1");
+
+      // Evict the entry manually
+      cache.clear();
+
+      const second = await cache.getOrFetch("key1", fetcher, 60_000);
+      expect(second).toBe("value-2");
+      expect(callCount).toBe(2);
+    });
+
+    test("should reject on first-fetch failure", async () => {
+      cache = new PromiseCache(10);
+      const failingFetcher = () => Promise.reject(new Error("fail"));
+
+      await expect(cache.getOrFetch("key1", failingFetcher, 60_000)).rejects.toThrow();
+
+      // Wait a tick for internal cleanup
+      await new Promise((r) => setTimeout(r, 0));
+      expect(cache.size).toBe(0);
+    });
+
+    test("should deduplicate concurrent requests for the same key", async () => {
+      cache = new PromiseCache(10);
+      let callCount = 0;
+      const fetcher = () => {
+        callCount++;
+        return new Promise<string>((resolve) => setTimeout(() => resolve(`value-${callCount}`), 10));
+      };
+
+      const [a, b] = await Promise.all([
+        cache.getOrFetch("key1", fetcher, 60_000),
+        cache.getOrFetch("key1", fetcher, 60_000),
+      ]);
+
+      expect(a).toBe("value-1");
+      expect(b).toBe("value-1");
+      expect(callCount).toBe(1);
+    });
+  });
+
+  describe("maxSize eviction", () => {
+    test("should evict entries when maxSize is exceeded", async () => {
+      cache = new PromiseCache(3);
+
+      await cache.getOrFetch("a", () => Promise.resolve(1), 60_000);
+      await cache.getOrFetch("b", () => Promise.resolve(2), 60_000);
+      await cache.getOrFetch("c", () => Promise.resolve(3), 60_000);
+      expect(cache.size).toBe(3);
+
+      // Adding a 4th entry should trigger eviction of the LRU entry ("a")
+      await cache.getOrFetch("d", () => Promise.resolve(4), 60_000);
+      expect(cache.size).toBe(3);
+
+      // "a" should have been evicted — re-fetching should call a new fetcher
+      let refetched = false;
+      await cache.getOrFetch(
+        "a",
+        () => {
+          refetched = true;
+          return Promise.resolve(10);
+        },
+        60_000,
+      );
+      expect(refetched).toBe(true);
+    });
+
+    test("should respect maxSize of 1", async () => {
+      cache = new PromiseCache(1);
+
+      await cache.getOrFetch("a", () => Promise.resolve(1), 60_000);
+      expect(cache.size).toBe(1);
+
+      await cache.getOrFetch("b", () => Promise.resolve(2), 60_000);
+      expect(cache.size).toBe(1);
+
+      // Only "b" should remain
+      let aRefetched = false;
+      await cache.getOrFetch(
+        "a",
+        () => {
+          aRefetched = true;
+          return Promise.resolve(10);
+        },
+        60_000,
+      );
+      expect(aRefetched).toBe(true);
+    });
+  });
+
+  describe("clear", () => {
+    test("should remove all entries", async () => {
+      cache = new PromiseCache(10);
+
+      await cache.getOrFetch("a", () => Promise.resolve(1), 60_000);
+      await cache.getOrFetch("b", () => Promise.resolve(2), 60_000);
+      expect(cache.size).toBe(2);
+
+      cache.clear();
+      expect(cache.size).toBe(0);
+    });
+  });
+});

apps/login/src/lib/cache.ts

@@ -0,0 +1,61 @@
+import { LRUCache } from "lru-cache";
+
+interface FetchContext {
+  fetcher: () => Promise<any>;
+}
+
+/**
+ * A bounded, stale-while-revalidate in-memory promise cache backed by lru-cache.
+ *
+ * Features:
+ * - True LRU eviction
+ * - Deduplicates concurrent requests (built-in to lru-cache's fetchMethod)
+ * - Serves stale data immediately while revalidating in the background
+ * - Keeps stale value on fetch rejection
+ * - Bounded to `maxSize` entries to prevent unbounded memory growth
+ */
+export class PromiseCache {
+  private readonly cache: LRUCache<string, any, FetchContext>;
+
+  constructor(maxSize = 100_000, perf?: { now: () => number }) {
+    this.cache = new LRUCache<string, any, FetchContext>({
+      max: Math.max(1, maxSize),
+      // A global TTL is required to initialize lru-cache's TTL tracking internals.
+      // Per-entry TTLs passed to fetch() will override this default.
+      ttl: 1,
+      allowStale: true,
+      noDeleteOnStaleGet: true,
+      noDeleteOnFetchRejection: true,
+      allowStaleOnFetchRejection: true,
+      fetchMethod: async (_key, _staleValue, { context }) => {
+        return context.fetcher();
+      },
+      ...(perf ? { perf, ttlResolution: 0 } : {}),
+    });
+  }
+
+  /**
+   * Get a cached value or execute the fetcher and cache its result.
+   *
+   * After the first successful fetch, expired entries return the stale
+   * value immediately and trigger a background revalidation (SWR).
+   * Only the very first call for a key (or after eviction) blocks
+   * on the fetch.
+   */
+  getOrFetch<T>(key: string, fetcher: () => Promise<T>, ttlMs: number): Promise<T> {
+    return this.cache.forceFetch(key, {
+      ttl: ttlMs,
+      context: { fetcher },
+    }) as Promise<T>;
+  }
+
+  /** Current number of entries (including stale). */
+  get size(): number {
+    return this.cache.size;
+  }
+
+  /** Clear all entries. */
+  clear(): void {
+    this.cache.clear();
+  }
+}

apps/login/src/lib/zitadel.ts

@@ -37,6 +37,7 @@ import { createLogger } from "./logger";
 import { errorClassificationInterceptor } from "@/lib/grpc/interceptors/error-classification";
 import { otelGrpcInterceptor } from "@/lib/grpc/interceptors/otel";
 import { Code, ConnectError, Interceptor } from "@connectrpc/connect";
+import { PromiseCache } from "./cache";
 import { createServiceForHost } from "./service";
 
 const logger = createLogger("zitadel");
@@ -67,29 +68,30 @@ function getTTLForKey(keyPrefix: string, fallbackTtl: number) {
   return fallbackTtl;
 }
 
-const promiseCache = new Map<string, { promise: Promise<any>; expiresAt: number }>();
+/**
+ * Build a cache key scoped to the current instance.
+ * In multi-tenant mode serviceConfig.instanceHost distinguishes tenants;
+ * without it we fall back to "default" (single-tenant / self-hosted).
+ */
+function instanceCacheKey(serviceConfig: ServiceConfig, key: string): string {
+  return `${serviceConfig.instanceHost || "default"}:${key}`;
+}
+
+const promiseCache = new PromiseCache(Number(cacheConfig.maxSize) || 100);
 
 /**
  * A stale-while-revalidate in-memory cache to keep data fresh and deduplicate concurrent requests.
  * We cache the Promise, so concurrent requests share the exact same execution.
+ *
+ * The cache is bounded and periodically swept for expired entries
+ * to prevent unbounded memory growth.
  */
 function freshCache<T>(key: string, fetcher: () => Promise<T>, ttlMs: number): Promise<T> {
   if (!useCache) {
     return fetcher();
   }
 
-  const now = Date.now();
+  return promiseCache.getOrFetch(key, fetcher, ttlMs);
-  const cached = promiseCache.get(key);
-  if (cached && now < cached.expiresAt) {
-    return cached.promise;
-  }
-
-  const promise = fetcher();
-  promiseCache.set(key, { promise, expiresAt: now + ttlMs });
-
-  promise.catch(() => promiseCache.delete(key));
-
-  return promise;
 }
 
 export async function getHostedLoginTranslation({
@@ -125,7 +127,7 @@ export async function getHostedLoginTranslation({
   };
 
   return freshCache(
-    `getHostedLoginTranslation-${organization || "instance"}-${locale || "default"}`,
+    instanceCacheKey(serviceConfig, `getHostedLoginTranslation-${organization || "instance"}-${locale || "default"}`),
     fetcher,
     getTTLForKey("getHostedLoginTranslation", longCacheTTL),
   );
@@ -146,7 +148,7 @@ export async function getBrandingSettings({
   };
 
   return freshCache(
-    `getBrandingSettings-${organization || "instance"}`,
+    instanceCacheKey(serviceConfig, `getBrandingSettings-${organization || "instance"}`),
     fetcher,
     getTTLForKey("getBrandingSettings", longCacheTTL),
   );
@@ -167,7 +169,7 @@ export async function getLoginSettings({
   };
 
   return freshCache(
-    `getLoginSettings-${organization || "instance"}`,
+    instanceCacheKey(serviceConfig, `getLoginSettings-${organization || "instance"}`),
     fetcher,
     getTTLForKey("getLoginSettings", defaultCacheTTL),
   );
@@ -180,7 +182,11 @@ export async function getSecuritySettings({ serviceConfig }: WithServiceConfig)
     return settingsService.getSecuritySettings({}).then((resp) => (resp.settings ? resp.settings : undefined));
   };
 
-  return freshCache(`getSecuritySettings-instance`, fetcher, getTTLForKey("getSecuritySettings", defaultCacheTTL));
+  return freshCache(
+    instanceCacheKey(serviceConfig, `getSecuritySettings-instance`),
+    fetcher,
+    getTTLForKey("getSecuritySettings", defaultCacheTTL),
+  );
 }
 
 export async function getLockoutSettings({ serviceConfig, orgId }: WithServiceConfig<{ orgId?: string }>) {
@@ -193,7 +199,7 @@ export async function getLockoutSettings({ serviceConfig, orgId }: WithServiceCo
   };
 
   return freshCache(
-    `getLockoutSettings-${orgId || "instance"}`,
+    instanceCacheKey(serviceConfig, `getLockoutSettings-${orgId || "instance"}`),
     fetcher,
     getTTLForKey("getLockoutSettings", defaultCacheTTL),
   );
@@ -209,7 +215,7 @@ export async function getPasswordExpirySettings({ serviceConfig, orgId }: WithSe
   };
 
   return freshCache(
-    `getPasswordExpirySettings-${orgId || "instance"}`,
+    instanceCacheKey(serviceConfig, `getPasswordExpirySettings-${orgId || "instance"}`),
     fetcher,
     getTTLForKey("getPasswordExpirySettings", defaultCacheTTL),
   );
@@ -251,7 +257,11 @@ export async function getAllowedLanguages({ serviceConfig }: WithServiceConfig)
     });
   };
 
-  return freshCache(`getGeneralSettings-instance`, fetcher, getTTLForKey("getGeneralSettings", longCacheTTL));
+  return freshCache(
+    instanceCacheKey(serviceConfig, `getGeneralSettings-instance`),
+    fetcher,
+    getTTLForKey("getGeneralSettings", longCacheTTL),
+  );
 }
 
 export async function getLegalAndSupportSettings({
@@ -269,7 +279,7 @@ export async function getLegalAndSupportSettings({
   };
 
   return freshCache(
-    `getLegalAndSupportSettings-${organization || "instance"}`,
+    instanceCacheKey(serviceConfig, `getLegalAndSupportSettings-${organization || "instance"}`),
     fetcher,
     getTTLForKey("getLegalAndSupportSettings", longCacheTTL),
   );
@@ -290,7 +300,7 @@ export async function getPasswordComplexitySettings({
   };
 
   return freshCache(
-    `getPasswordComplexitySettings-${organization || "instance"}`,
+    instanceCacheKey(serviceConfig, `getPasswordComplexitySettings-${organization || "instance"}`),
     fetcher,
     getTTLForKey("getPasswordComplexitySettings", defaultCacheTTL),
   );
@@ -850,7 +860,11 @@ export async function getDefaultOrg({ serviceConfig }: WithServiceConfig): Promi
   };
 
   return useCache
-    ? freshCache(`getDefaultOrg-${"instance"}`, fetcher, getTTLForKey("getDefaultOrg", defaultCacheTTL))
+    ? freshCache(
+        instanceCacheKey(serviceConfig, "getDefaultOrg-instance"),
+        fetcher,
+        getTTLForKey("getDefaultOrg", defaultCacheTTL),
+      )
     : fetcher();
 }
 

pnpm-lock.yaml

@@ -256,6 +256,9 @@ importers:
       escape-html:
         specifier: ^1.0.3
         version: 1.0.3
+      lru-cache:
+        specifier: ^11.3.2
+        version: 11.3.2
       lucide-react:
         specifier: ^0.577.0
         version: 0.577.0([email protected])
@@ -9286,6 +9289,10 @@ packages:
     resolution: {integrity: sha512-aY/R+aEsRelme17KGQa/1ZSIpLpNYYrhcrepKTZgE+W3WM16YMCaPwOHLHsmopZHELU0Ojin1lPVxKR0MihncA==}
     engines: {node: 20 || >=22}
 
+  [email protected]:
+    resolution: {integrity: sha512-wgWa6FWQ3QRRJbIjbsldRJZxdxYngT/dO0I5Ynmlnin8qy7tC6xYzbcJjtN4wHLXtkbVwHzk0C+OejVw1XM+DQ==}
+    engines: {node: 20 || >=22}
+
   [email protected]:
     resolution: {integrity: sha512-KpNARQA3Iwv+jTA0utUVVbrh+Jlrr1Fv0e56GGzAFOXN7dk/FviaDW8LHmK52DlcH4WP2n6gI8vN1aesBFgo9w==}
 
@@ -21867,6 +21874,8 @@ snapshots:
 
   [email protected]: {}
 
+  [email protected]: {}
+
   [email protected]:
     dependencies:
       yallist: 3.1.1