Capture dev-stack logs in e2e runs and raise CI heap headroom

Author: RhysSullivan

e2e/setup/cloud.boot.ts

@@ -120,7 +120,15 @@ export const bootCloud = async (options: CloudBootOptions): Promise<CloudBooted>
           options.host ?? "127.0.0.1",
         ],
         cwd: cloudDir,
-        env,
+        // Mitigation, not a fix (the real per-request OOM growth is what the
+        // spec-compile LRU addresses): CI runners have 7GB, well above the
+        // default ~2GB V8 old-space ceiling, so give the dev server room
+        // before it hits the limit and starts 500ing. Scoped to this child
+        // only; dev-db does not need it.
+        env: {
+          ...env,
+          NODE_OPTIONS: `${process.env.NODE_OPTIONS ?? ""} --max-old-space-size=3072`.trim(),
+        },
         logFile: options.logFile,
       },
     ],

e2e/setup/cloud.globalsetup.ts

@@ -3,18 +3,35 @@
 // (cloud.boot.ts — emulated WorkOS + Autumn + the app's real dev stack, the
 // same recipe the dev CLI uses). Set E2E_CLOUD_URL to attach to a running
 // stack instead.
+import { mkdirSync } from "node:fs";
+import { resolve } from "node:path";
+
 import { claimAndBoot } from "../src/ports";
 import { E2E_COOKIE_PASSWORD, E2E_WORKOS_CLIENT_ID } from "../targets/cloud";
 import { waitForHttp } from "./boot";
 import { bootCloud } from "./cloud.boot";
 import { bootMotel, motelExporterEnv } from "./motel";
+import { RUNS_DIR } from "../src/scenario";
+
+// dev-db + vite dev stdout/stderr, swept into the failure-only artifact
+// upload (CI uploads e2e/runs/**): see .github/workflows/ci.yml. A
+// mid-shard OOM abort or a 500 cascade otherwise leaves no trace: boot.ts
+// defaults to stdio "ignore" unless E2E_VERBOSE=1. Skip the file when
+// E2E_VERBOSE is set so local `E2E_VERBOSE=1` keeps its inherited,
+// live-in-terminal output (boot.ts prioritizes logFile over E2E_VERBOSE, so
+// passing both would silently swallow the verbose stream into the file).
+const bootLogFile = process.env.E2E_VERBOSE
+  ? undefined
+  : resolve(RUNS_DIR, "cloud", ".server", "boot.log");
 
 export default async function setup(): Promise<(() => Promise<void>) | void> {
   if (process.env.E2E_CLOUD_URL) {
     await waitForHttp(process.env.E2E_CLOUD_URL);
     return;
   }
 
+  if (bootLogFile) mkdirSync(resolve(bootLogFile, ".."), { recursive: true });
+
   // Suite-owned trace store — every run captures distributed traces. Booted
   // once outside the retry: it binds its own OS-assigned port (not a claimed
   // one), so an EADDRINUSE on the claimed block doesn't implicate it.
@@ -50,6 +67,7 @@ export default async function setup(): Promise<(() => Promise<void>) | void> {
           workosClientId: E2E_WORKOS_CLIENT_ID,
           cookiePassword: E2E_COOKIE_PASSWORD,
           publicUrl,
+          logFile: bootLogFile,
           // Server + browser spans → the suite's motel. The app's exporter is
           // endpoint-agnostic, so the same layer that ships prod traces to
           // Axiom ships e2e traces to the suite store — "why was that page

e2e/setup/selfhost.globalsetup.ts

@@ -2,17 +2,32 @@
 // (src/ports.ts), then run the shared boot recipe (selfhost.boot.ts — the
 // same one the dev CLI uses). Set E2E_SELFHOST_URL to attach to a running
 // instance (with E2E_SELFHOST_ADMIN_EMAIL/PASSWORD matching it).
+import { mkdirSync } from "node:fs";
+import { resolve } from "node:path";
+
 import { claimAndBoot } from "../src/ports";
 import { SELFHOST_ADMIN } from "../targets/selfhost";
 import { waitForHttp } from "./boot";
 import { bootSelfhost } from "./selfhost.boot";
+import { RUNS_DIR } from "../src/scenario";
+
+// vite dev stdout/stderr, swept into the failure-only artifact upload (CI
+// uploads e2e/runs/**): see .github/workflows/ci.yml and cloud.globalsetup.ts
+// for the same pattern. Skip when E2E_VERBOSE=1 so local verbose runs keep
+// their inherited, live-in-terminal output (boot.ts prioritizes logFile over
+// E2E_VERBOSE).
+const bootLogFile = process.env.E2E_VERBOSE
+  ? undefined
+  : resolve(RUNS_DIR, "selfhost", ".server", "boot.log");
 
 export default async function setup(): Promise<(() => Promise<void>) | void> {
   if (process.env.E2E_SELFHOST_URL) {
     await waitForHttp(process.env.E2E_SELFHOST_URL);
     return;
   }
 
+  if (bootLogFile) mkdirSync(resolve(bootLogFile, ".."), { recursive: true });
+
   // Claim a free port (preferred block first, walk forward past squatters),
   // boot, and retry on EADDRINUSE (a Linux-CI ephemeral socket can grab a
   // claimed port between probe and bind). The claimed port is published via env
@@ -28,6 +43,7 @@ export default async function setup(): Promise<(() => Promise<void>) | void> {
         port,
         webBaseUrl: `http://localhost:${port}`,
         admin: SELFHOST_ADMIN,
+        logFile: bootLogFile,
       });
       return { teardown: procs.teardown, value: procs };
     },