@camunda/e2e-test-suite 0.0.1164 → 0.0.1165

package/dist/resources/test-api-v2-complete.sh

@@ -394,24 +394,31 @@ _curl_common() {
   # runs surface as empty responses (`<none>` status) and false-positive
   # FAILEDs in print_result. --max-time still bounds total time per attempt.
   #
-  # Outer gateway-retry loop: an nginx `502 Bad Gateway` / `504 Gateway
-  # Timeout` (or an empty/no-status response) is emitted by the ingress when
-  # it cannot reach the upstream orchestration REST API — e.g. while an
-  # orchestration pod is rolling during an upgrade-minor run. That is never a
-  # valid response to any assertion in this suite (the API returns JSON status
-  # bodies, never a bare nginx HTML page), so re-issue the whole request with
-  # capped backoff to ride out a brief upstream restart window. curl's own
-  # --retry budget (~4s) is too small for a pod restart; this widens it to
-  # ~90s. A sustained outage still returns the real 502/504 once the budget is
-  # exhausted, so it surfaces as a genuine FAILED (not masked).
+  # Outer gateway-retry loop: an nginx `502 Bad Gateway` / `503 Service
+  # Temporarily Unavailable` / `504 Gateway Timeout` (or an empty/no-status
+  # response) is emitted by the ingress when it cannot reach the upstream
+  # orchestration REST API — 502/504 when an upstream is reachable but not
+  # answering, 503 when the ingress has no ready endpoint at all (every
+  # orchestration pod unready, e.g. while a pod is rolling during an
+  # upgrade-minor run or while probes fail under the load this suite
+  # generates). Such a response is never a valid answer to any assertion in
+  # this suite (the API returns JSON status bodies, never a bare nginx HTML
+  # page), so re-issue the whole request with capped backoff to ride out a
+  # brief upstream restart window. curl's own --retry budget (~4s) is too
+  # small for a pod restart; this widens it to ~90s. A sustained outage still
+  # returns the real 502/503/504 once the budget is exhausted, so it surfaces
+  # as a genuine FAILED (not masked).
   local gw_attempt=0 gw_max=12 gw_response gw_status
   while :; do
     gw_response="$(curl --http1.1 --max-time 30 --connect-timeout 10 \
       --retry 3 --retry-delay 1 --retry-connrefused --retry-all-errors \
       -i -s -H 'Expect:' "$@")"
     gw_status="$(echo "$gw_response" | grep -m1 -E '^HTTP/[0-9.]+' | awk '{print $2}' || true)"
-    if { [[ "$gw_status" != "502" && "$gw_status" != "504" && -n "$gw_status" ]]; } \
-        || (( gw_attempt >= gw_max )); then
+    if { [[ -n "$gw_status" && "$gw_status" != "502" \
+            && "$gw_status" != "503" && "$gw_status" != "504" ]] \
+         || { [[ "$gw_status" == "503" \
+                 && "$gw_response" != *'<center>nginx</center>'* ]]; } \
+         || (( gw_attempt >= gw_max )); then
       printf '%s\n' "$gw_response"
       return 0
     fi

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@camunda/e2e-test-suite",
-  "version": "0.0.1164",
+  "version": "0.0.1165",
   "description": "End-to-end test helpers for Camunda 8",
   "repository": {
     "type": "git",

package/resources/test-api-v2-complete.sh

@@ -394,24 +394,31 @@ _curl_common() {
   # runs surface as empty responses (`<none>` status) and false-positive
   # FAILEDs in print_result. --max-time still bounds total time per attempt.
   #
-  # Outer gateway-retry loop: an nginx `502 Bad Gateway` / `504 Gateway
-  # Timeout` (or an empty/no-status response) is emitted by the ingress when
-  # it cannot reach the upstream orchestration REST API — e.g. while an
-  # orchestration pod is rolling during an upgrade-minor run. That is never a
-  # valid response to any assertion in this suite (the API returns JSON status
-  # bodies, never a bare nginx HTML page), so re-issue the whole request with
-  # capped backoff to ride out a brief upstream restart window. curl's own
-  # --retry budget (~4s) is too small for a pod restart; this widens it to
-  # ~90s. A sustained outage still returns the real 502/504 once the budget is
-  # exhausted, so it surfaces as a genuine FAILED (not masked).
+  # Outer gateway-retry loop: an nginx `502 Bad Gateway` / `503 Service
+  # Temporarily Unavailable` / `504 Gateway Timeout` (or an empty/no-status
+  # response) is emitted by the ingress when it cannot reach the upstream
+  # orchestration REST API — 502/504 when an upstream is reachable but not
+  # answering, 503 when the ingress has no ready endpoint at all (every
+  # orchestration pod unready, e.g. while a pod is rolling during an
+  # upgrade-minor run or while probes fail under the load this suite
+  # generates). Such a response is never a valid answer to any assertion in
+  # this suite (the API returns JSON status bodies, never a bare nginx HTML
+  # page), so re-issue the whole request with capped backoff to ride out a
+  # brief upstream restart window. curl's own --retry budget (~4s) is too
+  # small for a pod restart; this widens it to ~90s. A sustained outage still
+  # returns the real 502/503/504 once the budget is exhausted, so it surfaces
+  # as a genuine FAILED (not masked).
   local gw_attempt=0 gw_max=12 gw_response gw_status
   while :; do
     gw_response="$(curl --http1.1 --max-time 30 --connect-timeout 10 \
       --retry 3 --retry-delay 1 --retry-connrefused --retry-all-errors \
       -i -s -H 'Expect:' "$@")"
     gw_status="$(echo "$gw_response" | grep -m1 -E '^HTTP/[0-9.]+' | awk '{print $2}' || true)"
-    if { [[ "$gw_status" != "502" && "$gw_status" != "504" && -n "$gw_status" ]]; } \
-        || (( gw_attempt >= gw_max )); then
+    if { [[ -n "$gw_status" && "$gw_status" != "502" \
+            && "$gw_status" != "503" && "$gw_status" != "504" ]] \
+         || { [[ "$gw_status" == "503" \
+                 && "$gw_response" != *'<center>nginx</center>'* ]]; } \
+         || (( gw_attempt >= gw_max )); then
       printf '%s\n' "$gw_response"
       return 0
     fi