fix(acceptance): 清理中止验收的未提交任务
验收失败后，排队及尚未提交上游的任务会继续占用队列与 API 数据库连接池，导致下一轮容量测试被历史负载污染。 中止验收时仅取消确认未提交上游的任务，删除其未提交 attempt 与准入记录，释放租约并生成幂等退款事件；提交中或已收到上游响应的任务继续自然收敛。新建验收前也会清理历史失败 Run 的安全遗留任务，并等待其余任务终态。 验证：Go 全量测试通过；临时 PostgreSQL 集成测试覆盖排队、运行未提交、提交中与退款事件；gofmt、bash -n、ShellCheck、git diff --check 通过。
This commit is contained in:
@@ -153,6 +153,12 @@ func (s *Store) CreateAcceptanceRun(ctx context.Context, input CreateAcceptanceR
|
||||
}
|
||||
config, _ := json.Marshal(sanitizeAcceptanceMetadata(sanitizeJSONForStorage(input.Config)))
|
||||
tokenHash := acceptanceTokenSHA256(input.Token)
|
||||
if err := s.beginTransaction(ctx, func(tx pgx.Tx) error {
|
||||
_, err := cancelSafeAcceptanceTasksTx(ctx, tx, "")
|
||||
return err
|
||||
}); err != nil {
|
||||
return AcceptanceRun{}, err
|
||||
}
|
||||
return scanAcceptanceRun(s.pool.QueryRow(ctx, `
|
||||
INSERT INTO gateway_acceptance_runs (
|
||||
release_sha, api_image_digest, worker_image_digest, api_key_id, user_id,
|
||||
@@ -370,6 +376,9 @@ FOR UPDATE`, SystemSettingGatewayTrafficMode).Scan(&value); err != nil {
|
||||
current.WorkerImageDigest != strings.TrimSpace(input.WorkerImageDigest) {
|
||||
return GatewayTrafficMode{}, ErrAcceptanceStateConflict
|
||||
}
|
||||
if _, err := cancelSafeAcceptanceTasksTx(ctx, tx, current.RunID); err != nil {
|
||||
return GatewayTrafficMode{}, err
|
||||
}
|
||||
next := GatewayTrafficMode{Mode: "live", Revision: current.Revision + 1}
|
||||
nextValue, _ := json.Marshal(next)
|
||||
if _, err := tx.Exec(ctx, `
|
||||
@@ -391,6 +400,169 @@ WHERE id = $1::uuid AND status IN ('running', 'failed', 'passed')`, current.RunI
|
||||
return next, nil
|
||||
}
|
||||
|
||||
func cancelSafeAcceptanceTasksTx(ctx context.Context, tx pgx.Tx, runID string) (int64, error) {
|
||||
runID = strings.TrimSpace(runID)
|
||||
tag, err := tx.Exec(ctx, `
|
||||
UPDATE gateway_tasks task
|
||||
SET status = 'cancelled',
|
||||
error = NULL,
|
||||
error_code = 'acceptance_run_aborted',
|
||||
error_message = 'acceptance run ended before upstream submission',
|
||||
billing_status = CASE
|
||||
WHEN gateway_user_id IS NULL THEN 'not_required'
|
||||
WHEN reservation_amount > 0 THEN 'pending'
|
||||
ELSE 'released'
|
||||
END,
|
||||
billing_updated_at = now(),
|
||||
remote_task_payload = '{}'::jsonb,
|
||||
locked_by = NULL,
|
||||
locked_at = NULL,
|
||||
heartbeat_at = NULL,
|
||||
execution_token = NULL,
|
||||
execution_lease_expires_at = NULL,
|
||||
finished_at = now(),
|
||||
updated_at = now()
|
||||
WHERE task.run_mode = 'acceptance'
|
||||
AND task.status IN ('queued', 'running')
|
||||
AND COALESCE(task.remote_task_id, '') = ''
|
||||
AND NOT EXISTS (
|
||||
SELECT 1
|
||||
FROM gateway_task_attempts attempt
|
||||
WHERE attempt.task_id = task.id
|
||||
AND attempt.upstream_submission_status <> 'not_submitted'
|
||||
)
|
||||
AND (
|
||||
task.acceptance_run_id = NULLIF($1, '')::uuid
|
||||
OR (
|
||||
NULLIF($1, '') IS NULL
|
||||
AND EXISTS (
|
||||
SELECT 1
|
||||
FROM gateway_acceptance_runs run
|
||||
WHERE run.id = task.acceptance_run_id
|
||||
AND run.status IN ('failed', 'aborted')
|
||||
)
|
||||
)
|
||||
)`, runID)
|
||||
if err != nil {
|
||||
return 0, err
|
||||
}
|
||||
if _, err := tx.Exec(ctx, `
|
||||
DELETE FROM gateway_task_param_preprocessing_logs log
|
||||
USING gateway_task_attempts attempt, gateway_tasks task
|
||||
WHERE log.attempt_id = attempt.id
|
||||
AND attempt.task_id = task.id
|
||||
AND attempt.upstream_submission_status = 'not_submitted'
|
||||
AND task.error_code = 'acceptance_run_aborted'
|
||||
AND (
|
||||
task.acceptance_run_id = NULLIF($1, '')::uuid
|
||||
OR (
|
||||
NULLIF($1, '') IS NULL
|
||||
AND EXISTS (
|
||||
SELECT 1
|
||||
FROM gateway_acceptance_runs run
|
||||
WHERE run.id = task.acceptance_run_id
|
||||
AND run.status IN ('failed', 'aborted')
|
||||
)
|
||||
)
|
||||
)`, runID); err != nil {
|
||||
return 0, err
|
||||
}
|
||||
if _, err := tx.Exec(ctx, `
|
||||
DELETE FROM gateway_task_attempts attempt
|
||||
USING gateway_tasks task
|
||||
WHERE attempt.task_id = task.id
|
||||
AND attempt.upstream_submission_status = 'not_submitted'
|
||||
AND task.error_code = 'acceptance_run_aborted'
|
||||
AND (
|
||||
task.acceptance_run_id = NULLIF($1, '')::uuid
|
||||
OR (
|
||||
NULLIF($1, '') IS NULL
|
||||
AND EXISTS (
|
||||
SELECT 1
|
||||
FROM gateway_acceptance_runs run
|
||||
WHERE run.id = task.acceptance_run_id
|
||||
AND run.status IN ('failed', 'aborted')
|
||||
)
|
||||
)
|
||||
)`, runID); err != nil {
|
||||
return 0, err
|
||||
}
|
||||
if _, err := tx.Exec(ctx, `
|
||||
INSERT INTO settlement_outbox (
|
||||
task_id, event_type, action, amount, currency, pricing_snapshot, payload,
|
||||
status, next_attempt_at
|
||||
)
|
||||
SELECT task.id, 'task.billing.release', 'release', task.reservation_amount,
|
||||
task.billing_currency, task.pricing_snapshot,
|
||||
jsonb_build_object('taskId', task.id, 'reason', 'acceptance_run_aborted'),
|
||||
'pending', now()
|
||||
FROM gateway_tasks task
|
||||
WHERE task.error_code = 'acceptance_run_aborted'
|
||||
AND task.billing_status = 'pending'
|
||||
AND task.reservation_amount > 0
|
||||
AND (
|
||||
task.acceptance_run_id = NULLIF($1, '')::uuid
|
||||
OR (
|
||||
NULLIF($1, '') IS NULL
|
||||
AND EXISTS (
|
||||
SELECT 1
|
||||
FROM gateway_acceptance_runs run
|
||||
WHERE run.id = task.acceptance_run_id
|
||||
AND run.status IN ('failed', 'aborted')
|
||||
)
|
||||
)
|
||||
)
|
||||
ON CONFLICT (task_id, event_type) DO NOTHING`, runID); err != nil {
|
||||
return 0, err
|
||||
}
|
||||
if _, err := tx.Exec(ctx, `
|
||||
UPDATE gateway_concurrency_leases lease
|
||||
SET released_at = now()
|
||||
FROM gateway_tasks task
|
||||
WHERE lease.task_id = task.id
|
||||
AND lease.released_at IS NULL
|
||||
AND task.error_code = 'acceptance_run_aborted'
|
||||
AND (
|
||||
task.acceptance_run_id = NULLIF($1, '')::uuid
|
||||
OR (
|
||||
NULLIF($1, '') IS NULL
|
||||
AND EXISTS (
|
||||
SELECT 1
|
||||
FROM gateway_acceptance_runs run
|
||||
WHERE run.id = task.acceptance_run_id
|
||||
AND run.status IN ('failed', 'aborted')
|
||||
)
|
||||
)
|
||||
)`, runID); err != nil {
|
||||
return 0, err
|
||||
}
|
||||
if _, err := tx.Exec(ctx, `
|
||||
DELETE FROM gateway_task_admissions admission
|
||||
USING gateway_tasks task
|
||||
WHERE admission.task_id = task.id
|
||||
AND task.error_code = 'acceptance_run_aborted'
|
||||
AND (
|
||||
task.acceptance_run_id = NULLIF($1, '')::uuid
|
||||
OR (
|
||||
NULLIF($1, '') IS NULL
|
||||
AND EXISTS (
|
||||
SELECT 1
|
||||
FROM gateway_acceptance_runs run
|
||||
WHERE run.id = task.acceptance_run_id
|
||||
AND run.status IN ('failed', 'aborted')
|
||||
)
|
||||
)
|
||||
)`, runID); err != nil {
|
||||
return 0, err
|
||||
}
|
||||
if tag.RowsAffected() > 0 {
|
||||
if err := notifyTaskAdmissionTx(ctx, tx, "*"); err != nil {
|
||||
return 0, err
|
||||
}
|
||||
}
|
||||
return tag.RowsAffected(), nil
|
||||
}
|
||||
|
||||
func (s *Store) AuthorizeAcceptanceTask(ctx context.Context, runID string, token string, user *auth.User) (string, error) {
|
||||
mode, err := s.GetGatewayTrafficMode(ctx)
|
||||
if err != nil {
|
||||
|
||||
Reference in New Issue
Block a user