feat: restart a workload in place, health-gated and audited

オープン
#219 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
3〜5日
初心者へのやさしさ
30/100
issue の種類
機能追加
明瞭さ
おおむね明確
活発さ
活発
技術スタック
docker, go
領域
cli, devops

調査の方向性

Start from the existing ob deploy path, which keeps replicas when images are unchanged (runtime_unchanged), and from the deploy lock/fence, the rolling/converge path, and the journal and ob audit code the issue names. The payload names no specific files or tests, so locate these by searching the repository. Done means ob restart shows a plan, asks for confirmation, restarts replicas one at a time with health gating, records a journal and audit entry, and leaves the release and ob status unchanged.

索引モデルが issue の本文から書いたものです。

説明

enhancement
The problem

An operator sometimes needs to restart a running workload without releasing anything new. A typical case: a singleton worker misses an upstream event, and only its startup catch-up recovers it. The image and configuration have not changed; the process just needs to start again.

Onebox has no command for this. ob deploy with unchanged images retains the workload (runtime_unchanged), and neither the top-level commands nor ob service offer a restart.

Proposed change

ob restart <workload> [--reason TEXT], using the same guarantees as a deploy, without changing the release:

  • Shows what will restart and asks for confirmation (-y to skip). Takes the deploy lock/fence, so it cannot race a concurrent deploy or resume.
  • Restarts multi-replica workloads one replica at a time, health-gated, using the existing rolling/converge path. A replica that fails its health check stops the restart and reports it.
  • Restarts a single-replica workload, waits for it to become healthy, and reports the outcome.
  • Records a journal entry and an ob audit entry: who, when, which workload, the release it belongs to, the reason and the result.
  • Does not change images, configuration or the recorded release. ob status stays consistent.

The same could later cover supporting services (ob service restart <name>), where a restart may affect the data plane and should say so in the plan.

Current workaround

ssh <host> docker restart <container>. It works because the container, image and configuration stay the same, but it bypasses Onebox:

  • no confirmation and no deploy-lock fencing;
  • no health gate, so a failed start shows up only in ob status or in logs;
  • no rolling for multi-replica workloads;
  • nothing in the journal or ob audit, so the restart is invisible in the deployment history.
Scope and safety

It stays inside the one-application, one-host scope. It acts on a running production system only on an explicit operator command, after showing a plan and getting confirmation, and it changes no release state.

主要言語
Go
スター
4
フォーク
0
平均マージ
36分
マージ済み PR(30日)
13

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。

labstack/onebox のほかの issue

labstack/onebox 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。