another attempt at improving e2e flakiness

docker/docker-compose.end2end.yml

@@ -47,6 +47,13 @@ services:
     ports:
       - '6379:6379'
 
+  usage-ingestor:
+    environment:
+      # Wait for the async insert to flush before acking, so a reported operation is queryable in
+      # Insights immediately instead of racing ClickHouse's non-deterministic async-insert timer
+      # (the source of usage e2e flakiness). Prod keeps the default fire-and-forget behavior.
+      CLICKHOUSE_WAIT_FOR_ASYNC_INSERT: '1'
+
   server:
     environment:
       SUPERTOKENS_RATE_LIMIT: '0'

e2e/helpers/usage.ts

@@ -90,9 +90,9 @@
 
             return page.getByRole('link').filter({ hasText: operationName }).count();
           },
-          { timeout: 60_000, intervals: [2_000] },
+          { timeout: 30_000, intervals: [2_000] },
         )
         .toBeGreaterThan(0);
     },
     async expectInsightVersion(version) {
       await expect
@@ -102,9 +102,9 @@
 
             return page.getByText(version, { exact: true }).count();
           },
-          { timeout: 60_000, intervals: [2_000] },
+          { timeout: 30_000, intervals: [2_000] },
         )
         .toBeGreaterThan(0);
     },
   };
 }

e2e/specs/usage.spec.ts

@@ -7,7 +7,7 @@ import type { UsageHelper } from '../helpers/usage';
 // Each test waits on ClickHouse ingestion through several stacked polls (see helpers/usage.ts).
 // Keep the timeout above the sum of those poll budgets so a slow-but-succeeding poll isn't cut
 // off mid-wait by the per-test deadline.
-test.describe.configure({ mode: 'serial', timeout: 180_000 });
+test.describe.configure({ mode: 'serial', timeout: 120_000 });
 
 type UsageReport = {
   size: number;

integration-tests/testkit/oidc-integration.ts

@@ -299,7 +299,7 @@ export async function createOIDCIntegration(args: {
 
           const rawBody = await result.json();
 
-          const body = z
+          const parsed = z
             .object({
               user: z.object({
                 id: z.string(),
@@ -311,7 +311,16 @@ export async function createOIDCIntegration(args: {
                 ),
               }),
             })
-            .parse(rawBody);
+            .safeParse(rawBody);
+
+          if (!parsed.success) {
+            // SuperTokens answers 200 with a non-OK status (e.g. SIGN_IN_UP_NOT_ALLOWED) and no
+            // `user`. Surface the actual body so a flaky sign-in is diagnosable instead of a bare
+            // ZodError that hides why it failed.
+            throw new Error('Unexpected signinup response: ' + JSON.stringify(rawBody));
+          }
+
+          const body = parsed.data;
           const cookies = setCookie.parse(result.headers.getSetCookie());
           return {
             accessToken: cookies.find(c => c.name === 'sAccessToken')?.value ?? '',

packages/services/usage-ingestor/src/environment.ts

@@ -81,6 +81,11 @@ const ClickHouseModel = zod.object({
   CLICKHOUSE_PASSWORD: zod.string(),
   CLICKHOUSE_ASYNC_INSERT_BUSY_TIMEOUT_MS: emptyString(NumberFromString.optional()),
   CLICKHOUSE_ASYNC_INSERT_MAX_DATA_SIZE: emptyString(NumberFromString.optional()),
+  // Whether ClickHouse waits for the async insert to flush before acking. Defaults to 0
+  // (fire-and-forget) for throughput; e2e sets it to 1 so reports are queryable immediately.
+  CLICKHOUSE_WAIT_FOR_ASYNC_INSERT: emptyString(
+    zod.union([zod.literal('0'), zod.literal('1')]).optional(),
+  ),
 });
 
 const PrometheusModel = zod.object({
@@ -238,6 +243,7 @@ export const env = {
     password: clickhouse.CLICKHOUSE_PASSWORD,
     async_insert_busy_timeout_ms: clickhouse.CLICKHOUSE_ASYNC_INSERT_BUSY_TIMEOUT_MS ?? 30_000,
     async_insert_max_data_size: clickhouse.CLICKHOUSE_ASYNC_INSERT_MAX_DATA_SIZE ?? 200_000_000,
+    wait_for_async_insert: clickhouse.CLICKHOUSE_WAIT_FOR_ASYNC_INSERT === '1' ? 1 : 0,
   },
   heartbeat: base.HEARTBEAT_ENDPOINT ? { endpoint: base.HEARTBEAT_ENDPOINT } : null,
   sentry: sentry.SENTRY === '1' ? { dsn: sentry.SENTRY_DSN } : null,

packages/services/usage-ingestor/src/writer.ts

@@ -20,6 +20,7 @@ export interface ClickHouseConfig {
   password: string;
   async_insert_busy_timeout_ms: number;
   async_insert_max_data_size: number;
+  wait_for_async_insert: number;
 }
 
 const operationsFields = operationsOrder.join(', ');
@@ -153,7 +154,7 @@ async function writeCsv(
       searchParams: {
         query,
         async_insert: 1,
-        wait_for_async_insert: 0,
+        wait_for_async_insert: config.wait_for_async_insert,
         async_insert_busy_timeout_ms: config.async_insert_busy_timeout_ms,
         async_insert_max_data_size: config.async_insert_max_data_size,
       },