fix(vitest-cloud): retry transient post-execute status/log queries

scripts/test/cloud-vitest-wrapper.test.sh

@@ -139,6 +139,39 @@ else
   FAILURES=$((FAILURES + 1))
 fi
 
+# Check 10: transient execution-status query failures are retried.
+# Regression: a single flaky `gcloud run jobs executions describe` right after
+# `execute --wait` completed must not fail an otherwise-green run (observed
+# live 2026-08-18: execution succeeded on all shards, wrapper exited 1 because
+# the status query errored once).
+cat > "$FAKE_GCLOUD_DIR/gcloud" << 'FAKE2'
+#!/usr/bin/env bash
+echo "FAKE_GCLOUD: $@" >> "${FAKE_GCLOUD_LOG:-/dev/null}"
+if [[ "$*" == *"artifacts docker images describe"* ]] || [[ "$*" == *"artifacts repositories describe"* ]] || [[ "$*" == *"builds submit"* ]]; then exit 0; fi
+if [[ "$*" == *"auth print-access-token"* ]]; then echo "fake-token"; exit 0; fi
+if [[ "$*" == *"info"* ]]; then echo "/usr/lib/google-cloud-sdk"; exit 0; fi
+if [[ "$*" == *"logs read"* ]]; then echo "Test Files  1 passed (1)"; exit 0; fi
+if [[ "$*" == *"executions describe"* ]]; then
+  if [[ "$*" == *"succeededCount"* ]]; then
+    CC=$(cat "${FAKE_GCLOUD_LOG}.desccount" 2>/dev/null || echo 0); CC=$((CC+1)); echo "$CC" > "${FAKE_GCLOUD_LOG}.desccount"
+    if [ "$CC" -le 2 ]; then exit 1; fi
+    echo "1"   # one succeeded shard
+  else
+    echo "0"   # zero failed shards
+  fi
+  exit 0
+fi
+if [[ "$*" == *"executions list"* ]]; then echo "test-execution-1"; exit 0; fi
+if [[ "$*" == *"run jobs execute"* ]]; then echo "Execution test-execution-1"; exit 0; fi
+if [[ "$*" == *"run jobs"* ]]; then exit 0; fi
+exit 0
+FAKE2
+chmod +x "$FAKE_GCLOUD_DIR/gcloud"
+rm -f "${FAKE_GCLOUD_LOG}.desccount"
+rm -f "$FAKE_GCLOUD_LOG"; touch "$FAKE_GCLOUD_LOG"
+check "transient describe failures retried (2 failures, then success)" bash -c "bash '$SCRIPT' run --cloud --config=default >/dev/null 2>&1"
+check "describe was retried (>=3 describe calls logged)" bash -c "[ \$(grep -c 'executions describe' '$FAKE_GCLOUD_LOG') -ge 3 ]"
+
 # Cleanup
 rm -rf "$FAKE_GCLOUD_DIR"
 

scripts/vitest-cloud.sh

@@ -423,10 +423,15 @@ cmd_run() {
     exit 1
   fi
 
-  # Fetch logs
+  # Fetch logs (one short retry: right after --wait completes, log reads can
+  # transiently return empty; observed live 2026-08-18).
   echo "[vitest-cloud] Fetching logs..."
   local log_output
   log_output=$(gcloud beta run jobs executions logs read $(gcloud_flags) "$execution_id" 2>/dev/null || true)
+  if [ -z "$log_output" ]; then
+    sleep 3
+    log_output=$(gcloud beta run jobs executions logs read $(gcloud_flags) "$execution_id" 2>/dev/null || true)
+  fi
 
   # Print full log output from ALL shards.
   echo "$log_output"
@@ -437,23 +442,33 @@ cmd_run() {
   echo "$log_output" | grep -E '(\[vitest-entrypoint\]|Test Files|Tests )' || true
 
   # Check execution status — propagate query errors instead of normalizing to 0.
+  # Transient describe failures right after `execute --wait` returns are a real
+  # flake class (observed live 2026-08-18: execution succeeded on all 4 shards
+  # while a single describe errored, failing the wrapper); retry briefly before
+  # declaring the run failed.
+  query_count() {
+    local field="$1" val attempt
+    for attempt in 1 2 3 4 5; do
+      if val=$(gcloud run jobs executions describe $(gcloud_flags) "$execution_id" \
+        --format="value($field)" 2>/dev/null); then
+        echo "${val:-0}"
+        return 0
+      fi
+      sleep 3
+    done
+    return 1
+  }
   local succeeded
   local failed
-  if ! succeeded=$(gcloud run jobs executions describe $(gcloud_flags) "$execution_id" \
-    --format="value(status.succeededCount)" 2>/dev/null); then
+  if ! succeeded=$(query_count status.succeededCount); then
     echo "[vitest-cloud] ERROR: failed to query execution status"
     exit 1
   fi
-  if ! failed=$(gcloud run jobs executions describe $(gcloud_flags) "$execution_id" \
-    --format="value(status.failedCount)" 2>/dev/null); then
+  if ! failed=$(query_count status.failedCount); then
     echo "[vitest-cloud] ERROR: failed to query execution status"
     exit 1
   fi
 
-  # Normalize empty/null to 0
-  succeeded="${succeeded:-0}"
-  failed="${failed:-0}"
-
   echo ""
   echo "[vitest-cloud] Succeeded tasks: $succeeded"
   echo "[vitest-cloud] Failed tasks: $failed"