fix(acceptance): 联动压力采样与负载停止

零副本 Worker 站点没有 metrics,应从聚合采样中跳过;有副本的站点采样失败仍保持硬失败。补充缺失字段诊断,避免只输出无上下文的采样失败。\n\n容量轮次和混合长稳压测现在持续监视压力采样进程;采样器提前退出时立即终止当前负载并走失败收口,避免在资源监控失效后继续造任务。\n\n验证:bash -n、ShellCheck、零副本站点测试、采样失败联动停止测试、manual-release-test。
This commit is contained in:
2026-08-01 14:37:45 +08:00
parent 8749f38eb3
commit 53fc79691a
2 changed files with 108 additions and 10 deletions
@@ -1,4 +1,5 @@
#!/usr/bin/env bash
# shellcheck disable=SC2034 # Variables are consumed by dynamically sourced functions.
set -euo pipefail
root=$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)
@@ -15,6 +16,15 @@ awk '
# shellcheck source=/dev/null
source "$tmp/is-release-ancestor.sh"
awk '
/^workload_metrics_for_site\(\)/ { capture = 1 }
capture && /^current_max_pod_memory_mib\(\)/ { exit }
capture { print }
' "$script" >"$tmp/pressure-monitor.sh"
# shellcheck source=/dev/null
source "$tmp/pressure-monitor.sh"
cluster_root=$tmp/repository
git init -q "$cluster_root"
git -C "$cluster_root" config user.name 'Acceptance Script Test'
@@ -35,4 +45,39 @@ if is_release_ancestor invalid "$third"; then
exit 1
fi
remote_kubectl() {
printf 0
}
cluster_ssh() {
echo 'zero-replica Worker attempted to collect metrics' >&2
return 1
}
namespace=easyai
if workload_metrics_for_site worker ningbo; then
echo 'zero-replica Worker site returned metrics success' >&2
exit 1
else
metrics_status=$?
fi
[[ $metrics_status == 2 ]]
long_running_load() {
while :; do
sleep 1
done
}
(sleep 1; exit 1) &
pressure_pid=$!
AI_GATEWAY_ACCEPTANCE_PROCESS_WATCH_SECONDS=1
active_load_pid=
if run_with_pressure_monitor "$pressure_pid" long_running_load; then
echo 'load continued after the pressure sampler failed' >&2
exit 1
fi
wait "$pressure_pid" >/dev/null 2>&1 || true
[[ -z $active_load_pid ]]
echo 'production_acceptance_script_tests=PASS'