From a37dd0210bdad34f67ab6071d68a3111de9b1b08 Mon Sep 17 00:00:00 2001
From: Vincent Koc <vincentkoc@ieee.org>
Date: Fri, 19 Jun 2026 07:49:01 +0200
Subject: [PATCH] fix(e2e): bound upgrade survivor probe retries

---
 .../lib/upgrade-survivor/probe-gateway.mjs    | 14 ++++++--
 .../upgrade-survivor-probe-gateway.test.ts    | 33 +++++++++++++++++++
 2 files changed, 44 insertions(+), 3 deletions(-)

diff --git a/scripts/e2e/lib/upgrade-survivor/probe-gateway.mjs b/scripts/e2e/lib/upgrade-survivor/probe-gateway.mjs
index 875d990c2915..b708fdf35d01 100644
--- a/scripts/e2e/lib/upgrade-survivor/probe-gateway.mjs
+++ b/scripts/e2e/lib/upgrade-survivor/probe-gateway.mjs
@@ -107,7 +107,7 @@ function matchesDegradedReadyExpectation(body) {
 
 async function fetchProbeText() {
   const elapsedMs = Date.now() - startedAt;
-  const remainingMs = Math.max(1, timeoutMs - elapsedMs);
+  const remainingMs = timeoutMs - elapsedMs;
   const controller = new AbortController();
   const attemptDeadlineMs = Math.min(attemptTimeoutMs, remainingMs);
   let timer;
@@ -138,7 +138,7 @@ const startedAt = Date.now();
 let lastError;
 let lastResult;
 
-while (Date.now() - startedAt <= timeoutMs) {
+while (Date.now() - startedAt < timeoutMs) {
   try {
     const { response, text } = await fetchProbeText();
     let body;
@@ -171,9 +171,17 @@ while (Date.now() - startedAt <= timeoutMs) {
   } catch (error) {
     lastError = error instanceof Error ? error.message : String(error);
   }
+  const remainingDelayMs = timeoutMs - (Date.now() - startedAt);
+  if (remainingDelayMs <= 0) {
+    break;
+  }
+  const delayMs = Math.min(500, remainingDelayMs);
   await new Promise((resolve) => {
-    setTimeout(resolve, 500);
+    setTimeout(resolve, delayMs);
   });
+  if (delayMs === remainingDelayMs) {
+    break;
+  }
 }
 
 const suffix = lastResult ? ` (last HTTP ${lastResult.status}: ${lastResult.text})` : "";
diff --git a/test/scripts/upgrade-survivor-probe-gateway.test.ts b/test/scripts/upgrade-survivor-probe-gateway.test.ts
index 67a82e5ae288..d10bc75f74b8 100644
--- a/test/scripts/upgrade-survivor-probe-gateway.test.ts
+++ b/test/scripts/upgrade-survivor-probe-gateway.test.ts
@@ -192,6 +192,39 @@ describe("scripts/e2e/lib/upgrade-survivor/probe-gateway.mjs", () => {
     }
   });
 
+  it("keeps failed probe retries inside the total timeout", async () => {
+    const server = createHttpServer((_request, response) => {
+      response.writeHead(503, { "content-type": "application/json" });
+      response.end(JSON.stringify({ ready: false, failing: ["gateway"] }));
+    });
+    const baseUrl = await listen(server);
+    const out = path.join(makeTempDir(), "ready-timeout.json");
+    const startedAt = Date.now();
+    try {
+      const result = await runProbe([
+        "--base-url",
+        baseUrl,
+        "--path",
+        "/readyz",
+        "--expect",
+        "ready",
+        "--out",
+        out,
+        "--timeout-ms",
+        "50",
+        "--attempt-timeout-ms",
+        "25",
+      ]);
+
+      expect(result.status).not.toBe(0);
+      expect(Date.now() - startedAt).toBeLessThan(300);
+      expect(result.stderr).toContain("probe did not satisfy ready within 50ms");
+      expect(fs.existsSync(out)).toBe(false);
+    } finally {
+      server.close();
+    }
+  });
+
   it("allows degraded ready responses only when degraded readiness is explicit", async () => {
     const server = createHttpServer((_request, response) => {
       response.writeHead(503, { "content-type": "application/json" });