From 051e870cfaf23c99439ac313854ea125c6e20913 Mon Sep 17 00:00:00 2001
From: bgard68 <30295154+bgard68@users.noreply.github.com>
Date: Mon, 24 Aug 2026 10:42:06 -0500
Subject: [PATCH] fix(deploy): require three consecutive ready responses before
 the smoke test
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

The readiness fix closed one race and revealed a second. webapps-deploy returns
while the old process is still serving, so a single /ready 200 can come from the
outgoing instance seconds before App Service restarts it. The wait passed on
that stale answer and the smoke test landed mid-restart — the same 500-on-every-
endpoint signature, on a build that served 200 everywhere once it settled.

One ready answer proves the old instance was alive. Three in a row spanning
twenty seconds cannot all be the doomed instance.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
---
 .github/workflows/deploy.yml | 28 +++++++++++++++++++---------
 1 file changed, 19 insertions(+), 9 deletions(-)

diff --git a/.github/workflows/deploy.yml b/.github/workflows/deploy.yml
index 85f5b1d..de6303c 100644
--- a/.github/workflows/deploy.yml
+++ b/.github/workflows/deploy.yml
@@ -188,19 +188,29 @@ jobs:
           BASE_URL: ${{ vars.API_BASE_URL }}
         run: |
           set -uo pipefail
+          # Two rules, each earned by a deploy that failed without them:
+          #
+          # /ready, not /live — liveness answers as soon as the process listens, before the
+          # app can serve, and the smoke test then measured a half-started app (500 on every
+          # endpoint, including readiness itself).
+          #
+          # Consecutive successes, not one — webapps-deploy returns while the OLD process is
+          # still serving, so a single 200 can come from the outgoing instance seconds before
+          # the restart. The smoke test then landed mid-restart and saw the same 500s. Three
+          # in a row spanning twenty seconds cannot all be the doomed instance.
+          streak=0
           for attempt in $(seq 1 30); do
-            # /live, not /ready, was the original poll target, and it answers as soon as the
-            # process is listening — before the container has finished resolving. The smoke
-            # test then ran against a half-started app and reported 500 on every endpoint,
-            # including "Health readiness expected 200 got 500", which is the readiness probe
-            # saying it was not ready while the liveness probe said the app was up. That is
-            # the distinction the two endpoints exist for; the wait was asking the wrong one.
             code=$(curl -s -o /dev/null -w "%{http_code}" --max-time 90 "$BASE_URL/api/health/ready" || true)
             if [[ "$code" == "200" ]]; then
-              echo "up after ${attempt} attempt(s)"
-              exit 0
+              streak=$((streak + 1))
+              if [[ "$streak" -ge 3 ]]; then
+                echo "ready: 3 consecutive 200s by attempt ${attempt}"
+                exit 0
+              fi
+            else
+              streak=0
             fi
-            echo "attempt ${attempt}: HTTP ${code:-none}"
+            echo "attempt ${attempt}: HTTP ${code:-none} (streak ${streak})"
             sleep 10
           done
           echo "::error::App did not return 200 from /api/health/ready within the wait window."