test(webapp): run schedule timing benchmarks on demand, not in CI

Wall-clock speedup ratios are single-sample and swing well past the asserted
bounds on a shared runner, so the benchmarks now live behind a *.perf.test.ts
suffix with their own config and a test:perf script, following the pattern the
e2e suites already use. Correctness stays in the ordinary suite.

Also splits the malformed-expression test, whose name claimed a lastRun
degradation that its assertions did not cover.

Author: ericallam

apps/webapp/package.json

@@ -24,6 +24,7 @@
     "db:seed:webhooks": "tsx seed-webhook-deliveries.ts",
     "upload:sourcemaps": "bash ./upload-sourcemaps.sh",
     "test": "vitest --no-file-parallelism",
+    "test:perf": "vitest --config ./vitest.perf.config.ts --run",
     "eval:dev": "evalite watch"
   },
   "dependencies": {

…ebapp/test/scheduleTimings.bench.test.ts …webapp/test/scheduleTimings.perf.test.tsapps/webapp/test/scheduleTimings.bench.test.ts renamed to apps/webapp/test/scheduleTimings.perf.test.ts apps/webapp/test/scheduleTimings.bench.test.ts renamed to apps/webapp/test/scheduleTimings.perf.test.ts

@@ -266,23 +266,26 @@ describe("resolveScheduleTimings", () => {
     expect(fresh.lastRun).toEqual(new Date("2024-06-15T00:00:00.000Z"));
   });
 
-  it("degrades to undefined lastRun for a malformed expression rather than throwing", () => {
-    const rows = [input({ cron: "0 0 * * *" }), input({ cron: "not a cron" })];
-
-    expect(() =>
-      resolveScheduleTimings([rows[1]], {
-        phaseSecret: PHASE_SECRET,
-        includeLastRun: false,
-        now,
-      })
-    ).toThrow();
+  it("throws for a malformed expression, matching the previous behaviour", () => {
+    for (const includeLastRun of [false, true]) {
+      expect(() =>
+        resolveScheduleTimings([input({ cron: "not a cron" })], {
+          phaseSecret: PHASE_SECRET,
+          includeLastRun,
+          now,
+        })
+      ).toThrow();
+    }
+  });
 
-    const [valid] = resolveScheduleTimings([rows[0]], {
+  it("resolves lastRun for a valid expression", () => {
+    const [valid] = resolveScheduleTimings([input({ cron: "0 0 * * *" })], {
       phaseSecret: PHASE_SECRET,
       includeLastRun: true,
       now,
     });
-    expect(valid.lastRun).toBeDefined();
+
+    expect(valid.lastRun).toEqual(new Date("2024-06-15T00:00:00.000Z"));
   });
 
   it("honours a caller-supplied schedulePhase over the derived one", () => {

apps/webapp/test/scheduleTimings.test.ts

@@ -27,7 +27,7 @@ export default defineConfig({
     // *.e2e.test.ts: smoke matrix, run via vitest.e2e.config.ts.
     // *.e2e.full.test.ts: full auth suite, runs via vitest.e2e.full.config.ts
     // (needs a globalSetup-spawned webapp + Postgres container).
-    exclude: ["test/**/*.e2e.test.ts", "test/**/*.e2e.full.test.ts"],
+    exclude: ["test/**/*.e2e.test.ts", "test/**/*.e2e.full.test.ts", "test/**/*.perf.test.ts"],
     globals: true,
     pool: "forks",
     setupFiles: ["./test/setup.ts"], // load apps/webapp/.env

apps/webapp/vitest.config.ts

@@ -0,0 +1,22 @@
+import { defineConfig } from "vitest/config";
+import tsconfigPaths from "vite-tsconfig-paths";
+
+export default defineConfig({
+  test: {
+    include: ["test/**/*.perf.test.ts"],
+    globals: true,
+    pool: "forks",
+    /**
+     * These compare wall-clock timings between two implementations. Single
+     * samples on a shared CI runner swing by more than the ratios being
+     * asserted, so they are kept out of the default suite and run on demand
+     * with `pnpm run test:perf`. Correctness is covered by the ordinary
+     * suites; these exist to show the shape of the win and to catch a
+     * large regression locally.
+     */
+    fileParallelism: false,
+    testTimeout: 120_000,
+  },
+  // @ts-ignore
+  plugins: [tsconfigPaths({ projects: ["./tsconfig.json"] })],
+});