bug: install-stats by-release table credits smoke installs to the newest-published release, not the one installed

クローズ
  • 何か問題がありますか？
#1,622 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

関連するプルリクエストがすでにマージされています。

  • #1623 @Alan-Cha による — マージ済み

評価

難易度
3/5
見積もり時間
半日
初心者へのやさしさ
42/100
issue の種類
バグ
明瞭さ
おおむね明確
活発さ
活発
技術スタック
go, javascript, shell
領域
devtools, documentation

調査の方向性

Start in the install-stats page code that defines renderReleases, byPublication and ciByTag, and check how compareRelease is used there. Compare with scripts/release_smoke_test_linux.sh, which picks the highest stable release. The fix assigns each smoke run to the highest stable release published at or before the run's created_at. Done when the by-release table credits v0.8.1 for the Oct 4 to Oct 14 runs, with the tag-push residuals left documented.

索引モデルが issue の本文から書いたものです。

説明

bug

Problem

The by-release table on docs/install-stats/ subtracts each release's smoke-test CI installs from its Linux column. It assigns each smoke run to whichever release was published most recently when the run started (byPublication / ciByTag in renderReleases), pre-releases included.

The smoke test installs something else. A job's fresh install is the highest-versioned stable release: gh release list --exclude-pre-releases, then sort -V | tail -1 (scripts/release_smoke_test_linux.sh). The script has excluded pre-releases since it was added in d0be885c.

The two rules disagree whenever the most recently published release isn't the highest stable version:

  • A pre-release is newest. This happened from Oct 4 to Oct 14, while v0.9.0-rc.1 through rc.4 were published. Every smoke run in those windows installed v0.8.1. I confirmed this from one run log per window: each says Fresh install: v0.8.1. The page credited those runs to the rcs instead.
  • A backport is newest. For example, a v0.8.2 published after v0.9.4. The page would credit it with runs that installed v0.9.4. This hasn't happened yet.

Effect on today's page

Release Linux CI the page subtracts CI by the smoke test's rule
v0.8.1 93 8 (shows 85 people) 80 (13 people)
v0.9.0-rc.1 to rc.4 1 to 9 each 4, 25, 39, 4 (rows show ~0) none from main pushes
v0.9.0 and later unchanged unchanged

So about 72 CI installs are attributed to the wrong releases. The daily table isn't affected, because it subtracts the CI total whichever release was installed.

The page recomputes everything from data.json on each load, so fixing the rule also corrects the existing history. Nothing needs re-collecting.

Fix

Assign each run to the highest-versioned stable release published at or before the run started, using the page's existing compareRelease. That is the smoke script's own rule, applied to data data.json already has (prerelease, published_at and each run's created_at).

Known residuals

  • Tag-push runs. A run triggered by a stable tag push installs the stable version below that tag, but this rule picks the tag itself. That puts 1 install on v0.9.0 and 2 on v0.9.1 into the wrong row so far. Fixing it needs the collector to record which ref triggered each run; today the smoke run's own head_branch is main even for tag pushes.
  • Contradicting comments. The collector's header says a tag push makes the people figure run "1-2 low per job"; the page says "1-2 high". The page is right: a tag-push job also installs the tag itself, which the page doesn't subtract. The "1-2" estimate also predates the cpex steps added in #1431, which install the tag several more times per job.

Found during the review of #1457.

主要言語
Go
スター
23
フォーク
44
平均マージ
16時間 37分
マージ済み PR（30日）
250

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。

rossoctl/cortex のほかの issue

rossoctl/cortex 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。