reduce e2e test flakiness

e2e/fixtures.ts

@@ -12,13 +12,36 @@ import * as path from "path";
 export const test = base.extend<{ server: { baseURL: string }; page: Page }>({
   server: [
     async ({}, use) => {
+      const server = await startServer();
+      await use({ baseURL: server.baseURL });
+      await killTree(server.child);
+      // Clean up the DB files (including -wal / -shm).
+      for (const suffix of ["", "-wal", "-shm"]) {
+        fs.rmSync(server.dbPath + suffix, { force: true });
+      }
+    },
+    { scope: "test", auto: true },
+  ],
+
+  // Provide a page whose baseURL points at this test's server.
+  page: async ({ browser, server }, use) => {
+    const context = await browser.newContext({ baseURL: server.baseURL });
+    const page = await context.newPage();
+    await use(page);
+    await context.close();
+  },
+});
+
+/** Starts a server, retrying on a fresh port if the first attempt fails to bind. */
+async function startServer() {
+  for (let attempt = 0; attempt < 5; attempt++) {
     const dbPath = path.join(
       os.tmpdir(),
       `sustenance-e2e-${process.pid}-${Date.now()}-${Math.random()
         .toString(36)
         .slice(2)}.db`,
     );
-      const port = 3200 + Math.floor(Math.random() * 2000);
+    const port = 20000 + Math.floor(Math.random() * 30000);
     const baseURL = `http://127.0.0.1:${port}`;
 
     const child = spawn(
@@ -33,33 +56,35 @@ export const test = base.extend<{ server: { baseURL: string }; page: Page }>({
           // Point SEED_CONFIG at a nonexistent file so no default user is created.
           SEED_CONFIG: path.join(os.tmpdir(), "sustenance-e2e-no-seed.json"),
         },
-          stdio: "ignore",
+        stdio: ["ignore", "ignore", "pipe"],
         // Run in its own process group so we can kill the whole tree.
         detached: true,
       },
     );
 
+    let stderr = "";
+    child.stderr?.on("data", (chunk) => {
+      stderr += chunk.toString();
+    });
+
+    try {
       await waitForServer(baseURL, child);
-
+      return { baseURL, child, dbPath };
-      await use({ baseURL });
+    } catch (error) {
-
+      // The server may have failed to bind (port collision). Clean up and retry.
       await killTree(child);
-      // Clean up the DB files (including -wal / -shm).
       for (const suffix of ["", "-wal", "-shm"]) {
         fs.rmSync(dbPath + suffix, { force: true });
       }
-    },
+      if (attempt === 4) {
-    { scope: "test", auto: true },
+        throw new Error(
-  ],
+          `server failed to start after retries; last stderr:\n${stderr}\n${error}`,
-
+        );
-  // Provide a page whose baseURL points at this test's server.
+      }
-  page: async ({ browser, server }, use) => {
+    }
-    const context = await browser.newContext({ baseURL: server.baseURL });
+  }
-    const page = await context.newPage();
+  throw new Error("unreachable");
-    await use(page);
+}
-    await context.close();
-  },
-});
 
 async function waitForServer(baseURL: string, child: ChildProcess) {
   const deadline = Date.now() + 60_000;

e2e/playwright.config.ts

@@ -3,7 +3,7 @@ import { defineConfig } from "@playwright/test";
 export default defineConfig({
   testDir: "./tests",
   timeout: 30_000,
-  retries: 0,
+  retries: 2,
   use: {
     trace: "on-first-retry",
   },