Hacktoberfest 2026：メンテナが10月に向けて印を付けた、オープンで初心者向けの issue。 Hacktoberfest の issue を見る

[Bug][dora] calculateChangeLeadTime loads the whole commits_diffs deployment join into memory

クローズ
  • 何か問題がありますか？
#9,280 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 2 日以内に返信

関連するプルリクエストがすでにマージされています。

  • #9281 @mdelacruz-avantio による — マージ済み

評価

難易度
3/5
見積もり時間
1〜2日
初心者へのやさしさ
25/100
issue の種類
バグ
明瞭さ
明確に書かれている
活発さ
停滞
技術スタック
go
領域
backend, data-engineering

調査の方向性

Start with batchFetchDeployments in backend/plugins/dora/tasks/change_lead_time_calculator.go and inspect how the second phase joins cicd_deployment_commits with commits_diffs. The issue proposes streaming rows into a three-column struct while preserving the joins, filters, ordering, and first-wins tie-break; an open pull request already tracks this work. Done means memory scales with distinct merge commits and the tie-break has an end-to-end test.

索引モデルが issue の本文から書いたものです。

説明

Search before asking
  • I had searched in the issues and found no similar issues.
What happened

batchFetchDeployments in backend/plugins/dora/tasks/change_lead_time_calculator.go resolves merge commits to deployments in two phases. The second phase joins cicd_deployment_commits with commits_diffs, loads every joined row into a slice of full CicdDeploymentCommit structs with db.All, and only then keeps the first row per merge commit in Go.

Memory therefore grows with the size of the join, not with the number of merge commits. The join has one row per commit between a production deployment and its previous successful one, for every merge commit in that range. On a project where deployments are far apart in commit history, or where a deployment job records a commit from a second repository with a long history, the join becomes very large.

On our instance the query returns about 936,000 rows for 14,088 distinct merge commits (1,693 production deployment commits). The lake container peaks at about 1.7 GiB in this subtask, is OOMKilled at a 2 GiB limit on every nightly run, and then crash loops because each restart resumes the same pipeline at the same subtask. The row count grows every week.

What do you expect to happen

The memory used by this subtask stays proportional to the number of distinct merge commits, as the first phase already does, whatever the size of the commits_diffs join.

How to reproduce
  1. A project with a few hundred production deployment commits where consecutive deployments are thousands of commits apart, so commits_diffs holds many rows per deployment pair.
  2. Run the DORA pipeline and watch the lake container memory during calculateChangeLeadTime, or count the rows the second phase loads:
SELECT COUNT(*), COUNT(DISTINCT cd.commit_sha)
FROM cicd_deployment_commits dc
LEFT JOIN cicd_deployment_commits p ON dc.prev_success_deployment_commit_id = p.id
INNER JOIN commits_diffs cd ON cd.new_commit_sha = dc.commit_sha AND cd.old_commit_sha = COALESCE(p.commit_sha, '')
LEFT JOIN project_mapping pm ON pm.table = 'cicd_scopes' AND pm.row_id = dc.cicd_scope_id
WHERE dc.prev_success_deployment_commit_id <> '' AND dc.environment = 'PRODUCTION' AND dc.result = 'SUCCESS' AND pm.project_name = '<project>';

The first number is what the subtask materialises, the second is what it keeps.

Anything else

Streaming the second phase with db.Cursor and db.Fetch into a three-column struct and keeping the first row per merge commit, with the same joins, filters and ordering, makes the memory independent of the join size. I will open a pull request with that change and an e2e case for the first-wins tie-break.

Raising the container memory limit and setting GOMEMLIMIT is a stopgap only, the peak keeps growing with the join.

Version

v1.0.3-beta18. The function is unchanged on main at 625e512a4.

主要言語
Go
スター
3.2k
フォーク
832
平均マージ
2日 6時間
マージ済み PR（30日）
62

環境構築

Codespaces で開く

このプロジェクトの開発コンテナを、あなたの GitHub アカウントでブラウザ上に起動します。

  • Dockerfile・Docker Compose ファイルなし
  • プルリクエストのテンプレートあり
  • コントリビューションガイドなし

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。

apache/devlake のほかの issue

apache/devlake 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。