fix(ci): fit the install retry inside every caller and read a failed probe's log from its kept pod

Author: waleedlatif1

.github/actions/setup/action.yml

@@ -69,16 +69,17 @@ runs:
     # Bun retries a failed request itself, but a registry connection that stalls
     # without failing leaves `bun install` waiting until the job's own timeout
     # cancels it, the whole budget spent on one hung step. Healthy installs finish
-    # in under three minutes cold, so an attempt past five is stalled: kill it and
-    # retry. Any other failure is real and fails at once.
+    # in under two and a half minutes cold, so an attempt past three is stalled:
+    # kill it and retry once, which leaves room inside the shortest caller's
+    # 10-minute job timeout. Any other failure is real and fails at once.
     - name: Install dependencies
       shell: bash
       run: |
-        for attempt in 1 2 3; do
-          timeout --kill-after=30s 5m bun install --frozen-lockfile --ignore-scripts && exit 0
+        for attempt in 1 2; do
+          timeout --kill-after=15s 3m bun install --frozen-lockfile --ignore-scripts && exit 0
           status=$?
           if [ "$status" -ne 124 ] && [ "$status" -ne 137 ]; then exit "$status"; fi
-          echo "::warning title=bun install stalled::attempt ${attempt} of 3 ran past 5 minutes"
+          echo "::warning title=bun install stalled::attempt ${attempt} of 2 ran past 3 minutes"
         done
-        echo "::error title=bun install stalled::all 3 attempts ran past 5 minutes"
+        echo "::error title=bun install stalled::both attempts ran past 3 minutes"
         exit 1

.github/workflows/helm.yml

@@ -188,11 +188,12 @@ jobs:
             --values helm/sim/ci/kind-overlay.yaml \
             --wait --timeout 15m
 
-      # `--logs` prints the probe's output, so a failure says which URL refused.
       - name: Run helm test
-        run: helm test sim --namespace sim --timeout 5m --logs
+        run: helm test sim --namespace sim --timeout 5m
 
-      # After the test, so it also explains a failed probe, not only a failed install.
+      # After the test, so it also explains a failed probe, not only a failed
+      # install. A failed test pod is kept (the hook deletes it only on success),
+      # so its log says which URL refused.
       - name: Diagnostics on failure
         if: failure()
         run: |
@@ -202,6 +203,7 @@ jobs:
           kubectl -n sim logs deploy/sim-app -c migrations --tail=50 || true
           kubectl -n sim logs deploy/sim-app --tail=80 || true
           kubectl -n sim logs deploy/sim-realtime --tail=80 || true
+          kubectl -n sim logs sim-test-connection || true
 
   # Resolved once and shared, so the two publish paths cannot package the same
   # immutable chart version with different appVersions -- they run in parallel,

helm/sim/templates/tests/test-connection.yaml

@@ -42,16 +42,15 @@ spec:
           set -eu
           # A pod turns Ready before every node routes its Service to it, so a
           # probe sent the moment `helm install --wait` returns can be refused.
-          # Each URL gets a minute to answer before the test fails.
+          # Each URL is retried until a minute has passed, then the test fails.
           probe() {
-            attempt=1
+            deadline=$(($(date +%s) + 60))
             while ! wget -q -T {{ .Values.tests.timeoutSeconds }} --spider "$1"; do
-              if [ "$attempt" -ge 12 ]; then
-                echo "FAILED: $1 did not answer after $attempt attempts"
+              if [ "$(date +%s)" -ge "$deadline" ]; then
+                echo "FAILED: $1 did not answer within 60s"
                 return 1
               fi
-              echo "Attempt $attempt: $1 did not answer; retrying in 5s"
-              attempt=$((attempt + 1))
+              echo "$1 did not answer; retrying in 5s"
               sleep 5
             done
             echo "OK: $1"