fix(acceptance): 支持失败验收无缝重试

在 validation 保持正式流量关闭时,允许为同一 release 与镜像 digest 创建新 Run,并仅在确认旧 Run 已失败后通过 CAS 中止旧 Run、立即激活新 Run。避免验收控制连接异常后必须先长时间切回 live 才能重试。\n\n验证:bash -n;shellcheck -x -P .;git diff --check。
This commit is contained in:
2026-07-31 00:25:52 +08:00
parent 3053ba4925
commit cec3d429af
+51 -1
View File
@@ -243,6 +243,9 @@ deploy_protocol_emulator() {
create_and_activate_run() {
local create_response=$temporary_root/create-run.json
local activate_response=$temporary_root/activate-run.json
local traffic_response=$temporary_root/pre-activate-traffic.json
local previous_run_response=$temporary_root/previous-run.json
local abort_response=$temporary_root/abort-previous-run.json
local body
body=$(jq -cn \
--arg releaseSha "$release_sha" \
@@ -274,8 +277,55 @@ create_and_activate_run() {
report_root=${AI_GATEWAY_ACCEPTANCE_REPORT_DIR:-"$cluster_root/dist/acceptance/$run_id"}
mkdir -p "$report_root"
chmod 0700 "$report_root"
admin_request GET /api/admin/system/acceptance/traffic-mode '' "$traffic_response"
local traffic_mode previous_run_id previous_status
traffic_mode=$(jq -r '.mode // empty' "$traffic_response")
if [[ $traffic_mode == validation ]]; then
previous_run_id=$(jq -r '.runId // empty' "$traffic_response")
[[ $previous_run_id =~ ^[0-9a-f-]{36}$ && $previous_run_id != "$run_id" ]] || {
echo 'validation mode does not reference a different previous Run' >&2
return 1
}
[[ $(jq -r '.releaseSha // empty' "$traffic_response") == "$release_sha" &&
$(jq -r '.apiImageDigest // empty' "$traffic_response") == "$api_digest" &&
$(jq -r '.workerImageDigest // empty' "$traffic_response") == "$worker_digest" ]] || {
echo 'refusing to replace validation mode for a different production release' >&2
return 1
}
admin_request GET "/api/admin/system/acceptance/runs/$previous_run_id" '' \
"$previous_run_response"
previous_status=$(jq -r '.status // empty' "$previous_run_response")
[[ $previous_status == failed ]] || {
echo "refusing to replace previous acceptance Run in status $previous_status" >&2
return 1
}
body=$(jq -cn \
--argjson revision "$(jq -r '.revision' "$traffic_response")" \
--arg releaseSha "$release_sha" \
--arg apiDigest "$api_digest" \
--arg workerDigest "$worker_digest" \
'{
revision:$revision,
releaseSha:$releaseSha,
apiImageDigest:$apiDigest,
workerImageDigest:$workerDigest
}')
admin_request POST \
"/api/admin/system/acceptance/runs/$previous_run_id/abort" \
"$body" "$abort_response"
[[ $(jq -r '.mode // empty' "$abort_response") == live ]] || {
echo 'previous failed acceptance Run did not return traffic to live' >&2
return 1
}
elif [[ $traffic_mode != live ]]; then
echo "unsupported production traffic mode before acceptance: $traffic_mode" >&2
return 1
fi
admin_request POST "/api/admin/system/acceptance/runs/$run_id/activate" '' "$activate_response"
[[ $(jq -r '.mode' "$activate_response") == validation ]]
[[ $(jq -r '.mode' "$activate_response") == validation &&
$(jq -r '.runId' "$activate_response") == "$run_id" ]]
}
mark_run_failed() {