ExecuteTaskActivity retries forever, so a failing step never parks for an admin

オープン
#302 コメント 0 件 リアクション 0 件 担当者 0 名 GitHub で見る

メンテナーはふだん 1 日以内に返信

まだ誰も着手していません。

評価

難易度
4/5
見積もり時間
1〜2日
初心者へのやさしさ
55/100
issue の種類
バグ
明瞭さ
明確に書かれている
活発さ
活発
技術スタック
go

調査の方向性

Start in workflow/workflow.go at handleTaskNode and trace the ExecuteTaskActivity options and the error path that parks a node for an admin. Then inspect dynamic_split.go and other engine activities for the same retry behavior. Done means repeated retryable task start failures eventually surface as an error and park the node, while steps waiting for asynchronous completion are not cut off by a total time limit.

索引モデルが issue の本文から書いたものです。

説明

Type/Bug
Describe the Bug

ExecuteTaskActivity runs with no retry policy, so a step whose start keeps failing is retried forever and never reaches the admin park.

handleTaskNode (workflow/workflow.go) sets only ActivityID and a one-year StartToCloseTimeout. With no RetryPolicy, Temporal's default applies: unlimited attempts, with the delay between them growing to 100s. A retryable error from the activity, such as a plugin failing in StartTaskStep, is retried indefinitely. The engine parks a node for an admin only when ExecuteActivity(...).Get returns an error (withCategory(ParkCategoryTaskFailure, err)), and that never happens.

This affects both levels: the task's own workflow (task-wf-…, MICRO_WORKFLOW_QUEUE), where the failing step runs, and the parent workflow's TASK node waiting on it.

To Reproduce
  1. Run a flow with a PAYMENT step while the host's refid config has no issuer for the payment reference format (e.g. refid.issuers: []).
  2. Reach the payment step.
  3. The worker logs Activity error ... Attempt=N with refid: unknown issuer forever. The node stays RUNNING with no last error, the trader sees "Starting Step", and the admin debugger shows nothing to resolve.
Expected Behavior

After a bounded number of attempts the activity fails, the engine parks the node (AWAITING_ADMIN) with the error, and an admin can see it and retry, complete or abort it.

Suggested shape:

  • Set a RetryPolicy on the task activity with a MaximumAttempts cap (a handful), keeping the backoff.
  • Cap attempts, not total time. The same activity also waits for asynchronous completion (a user submitting a form, an agency responding), which can legitimately take days, so a ScheduleToCloseTimeout would cut off steps that are only waiting.
  • Check the other activities the engine runs (e.g. dynamic_split.go) for the same unbounded default.
Additional Context
  • Without a cap, a misconfiguration is only visible in the worker logs.
  • Related: errors that can never succeed should not be retried at all: #303.
  • Seen in nsw-srilanka local testing, 2026-10-12.
主要言語
Go
スター
6
フォーク
8
平均マージ
21時間 36分
マージ済み PR(30日)
42

環境構築

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

OpenNSW/core のほかの issue

OpenNSW/core の issue をすべて見る

似ている issue

Go の issue をもっと見る

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。