Hold replay batches until Retry-After has passed

tests/browser/replay-feed.test.js

@@ -0,0 +1,79 @@
+// Run with: node --test tests/browser/replay-feed.test.js
+//
+// The replay feed's answer to a 429, run rather than read: a fixed one-second
+// retry passes every text check and still asks the server again sixty times
+// inside the window it named.
+
+import { test } from "node:test";
+import assert from "node:assert/strict";
+
+const replayFeed = await import("../../web/replay-feed.js");
+
+function fakeClock() {
+  const real = { now: Date.now, setTimeout: globalThis.setTimeout, clearTimeout: globalThis.clearTimeout };
+  let now = 0;
+  let nextId = 1;
+  const pending = new Map();
+  Date.now = () => now;
+  globalThis.setTimeout = (fn, ms) => {
+    pending.set(nextId, { at: now + ms, fn });
+    return nextId++;
+  };
+  globalThis.clearTimeout = (id) => pending.delete(id);
+  return {
+    tick(ms) {
+      now += ms;
+      for (const [id, timer] of [...pending]) {
+        if (timer.at > now) continue;
+        pending.delete(id);
+        timer.fn();
+      }
+    },
+    restore() {
+      Date.now = real.now;
+      globalThis.setTimeout = real.setTimeout;
+      globalThis.clearTimeout = real.clearTimeout;
+    },
+  };
+}
+
+test("replay-feed waits out Retry-After", async () => {
+  const clock = fakeClock();
+  const posts = [];
+  const statuses = [429, 204];
+  globalThis.fetch = async (_url, init) => {
+    posts.push(JSON.parse(init.body).events.length);
+    const status = statuses.shift();
+    return { status, headers: { get: (name) => (name === "Retry-After" ? "30" : null) } };
+  };
+  try {
+    replayFeed.initReplay({
+      state: { interviewId: "i1" },
+      nodes: {},
+      recordingEnabled: true,
+      consentVersion: "v1",
+      replayVersion: 1,
+    });
+    replayFeed.recordReplay("lifecycle", { state: "started" });
+    await replayFeed.flushReplay();
+    assert.deepEqual(posts, [1], "the first batch is refused");
+
+    replayFeed.recordReplay("lifecycle", { state: "still_here" });
+    await replayFeed.flushReplay();
+    assert.deepEqual(posts, [1], "a flush inside the window sends nothing");
+
+    for (let second = 1; second < 30; second += 1) {
+      clock.tick(1000);
+      await new Promise(setImmediate);
+    }
+    assert.deepEqual(posts, [1], "nor does any timer before the window has passed");
+
+    clock.tick(1000);
+    await new Promise(setImmediate);
+    assert.deepEqual(posts, [1, 2], "then the kept batch goes, with what queued behind it");
+  } finally {
+    replayFeed.closeReplay();
+    clock.restore();
+    delete globalThis.fetch;
+  }
+});

web/interview.js

@@ -1473,9 +1473,11 @@ function endInterview(reason) {
   // who ends while it is still up would read the report status through it.
   globalThis.clearTimeout(frameworkHintTimer);
   nodes.frameworkHint.hidden = true;
-  // Last event, and sent rather than queued: the page is about to stop being
-  // the kind of page that flushes timers, and an "ended" nobody sent leaves a
-  // replay that just stops.
+  // Last event, and flushed now rather than left to the batching timer: the
+  // page is about to stop being the kind of page that flushes timers, and an
+  // "ended" nobody sent leaves a replay that just stops. Inside a Retry-After
+  // window the flush sends nothing and the event still waits on that timer,
+  // so a tab closed before the window passes loses it.
   recordReplay("lifecycle", { state: "ended", reason });
   void flushReplay();
   // The end_interview payload carries the final buffer, so drop any debounced

web/replay-feed.js

@@ -56,7 +56,13 @@ export function recordReplay(kind, payload) {
     void flushReplay();
     return;
   }
-  replayTimer ||= setTimeout(() => void flushReplay(), REPLAY_FLUSH_MS);
+  scheduleFlush();
+}
+
+function scheduleFlush() {
+  if (replayTimer) return;
+  const delay = Math.max(REPLAY_FLUSH_MS, retryAfter - Date.now());
+  replayTimer = setTimeout(() => void flushReplay(), delay);
 }
 
 /// How long to wait after the server says it has heard enough for now.
@@ -91,8 +97,10 @@ export function closeReplay() {
 /// that batch is kept and sent when the minute is up.
 /// One flush at a time, in the order they were asked for.
 ///
-/// `recordReplay` starts one whenever the queue fills, and the interview's last
-/// act awaits one. Left unserialized, a batch posted while another was still
+/// `recordReplay` starts one whenever the queue fills, and the interview's end
+/// asks for one without waiting on it. Inside a Retry-After window a flush
+/// resolves without posting and leaves the queue to the timer that window
+/// armed. Left unserialized, a batch posted while another was still
 /// awaiting `fetch` could commit first, and the replay would be ordered by
 /// whichever request the server happened to finish rather than by what the
 /// candidate did. Chained rather than skipped, because a caller that is told
@@ -108,6 +116,10 @@ export function flushReplay() {
 
 async function sendQueuedBatch() {
   if (!replayQueue.length || replayClosed) return;
+  if (Date.now() < retryAfter) {
+    scheduleFlush();
+    return;
+  }
   const batch = replayQueue.splice(0, REPLAY_MAX_BATCH);
   try {
     const response = await fetch(`/api/interviews/${encodeURIComponent(state.interviewId)}/events`, {
@@ -154,7 +166,7 @@ async function sendQueuedBatch() {
   } catch {
     // Offline. The interview is what matters and it is still running.
   }
-  if (replayQueue.length) replayTimer ||= setTimeout(() => void flushReplay(), REPLAY_FLUSH_MS);
+  if (replayQueue.length) scheduleFlush();
 }
 
 /// The problem heading, as the recording shows it.