release

作者: clickhouse

Drive a package release of `ClickHouse/clickhouse-js` end to end: bump the version, sync the protected `release` branch from `main`, watch the npm publish…

npx skills add https://github.com/clickhouse/clickhouse-js --skill release

Releasing a clickhouse-js package

This skill is the source of truth for the release process. It supersedes the old RELEASING.md (which now just points here).

Before you start — read this

  1. Releases are per package. There is no "release everything" button. Ask the user which package(s) they want to release and confirm — the packages are versioned independently and ship on independent cadences:

    • @clickhouse/client — Node.js client (packages/client-node)
    • @clickhouse/client-web — Web client (packages/client-web)
    • @clickhouse/client-commondeprecated, effectively frozen; only cut a final standalone release if explicitly asked.
    • @clickhouse/datatype-parser — standalone type parser (packages/datatype-parser)
    • @clickhouse/rowbinary — standalone RowBinary codec skill/package (skills/clickhouse-js-node-rowbinary)

    The flow forks into two families — workspace client packages (client / client-web / client-common) and standalone packages (datatype-parser / rowbinary) — which differ in version bumping and publish workflow. Pick the right section below.

  2. release is a long-lived, protected branch. It is not a per-release branch. The npm-publish GitHub Actions environment only permits the release branch to deploy, and it additionally requires a manual approval before any publish job runs. You will hand the human an approval link and wait.

  3. You drive, the human judges. Run the workflows with gh, watch CI, and pause at: (a) reviewing the release PR, (b) the npm-publish approval gate, (c) the GitHub Release text. Never push to release directly — if the release PR needs fixes, route them through main (see the fix-release-pr skill).

One workflow per package. Each package has its own publish workflow:

  • @clickhouse/clientpublish-client.yml
  • @clickhouse/client-webpublish-client-web.yml
  • @clickhouse/client-commonpublish-client-common.yml
  • @clickhouse/datatype-parserpublish-datatype-parser.yml
  • @clickhouse/rowbinarypublish-skill-rowbinary.yml

Each client workflow has two triggers: an automatic head publish on push to release and a manual latest publish via workflow_dispatch. The standalone packages have the manual trigger only.

The client head triggers are path-scoped: a change touching only one client's own sources no longer republishes the others. Both @clickhouse/client and @clickhouse/client-web bundle the shared common sources via the src/common symlink (packages/*/src/commonpackages/client-common/src), so packages/client-common/** is also an input to both client workflows — a change to the common sources publishes a new head for every client that bundles them.


Part A — Workspace client packages (@clickhouse/client, -web, -common)

Step 1 — Verify the package CHANGELOG on main

Each package now keeps its own CHANGELOG.md (the repo-wide root CHANGELOG.md is frozen). The package you're releasing maps to:

  • @clickhouse/clientpackages/client-node/CHANGELOG.md
  • @clickhouse/client-webpackages/client-web/CHANGELOG.md
  • @clickhouse/client-commonpackages/client-common/CHANGELOG.md

These are normally updated inside feature PRs as they merge to main, under a top # <version> header. A common mistake: the in-progress entries sit under the wrong header — e.g. still under the previously released version, or under a version that doesn't match the one you're about to cut. (Note: shared/common code is bundled into both clients, so such changes should appear in both the client-node and client-web changelogs.)

  • Compute the version this release will produce: current version + the bump_type you'll pick. Current version:
    node -p "require('./packages/client-node/package.json').version"   # or client-web / client-common
    
  • Confirm that package's CHANGELOG.md has a top # <new-version> section that actually contains this release's notes.
  • If it's wrong, fix it in a small separate PR to main and merge it quickly before bumping — do not bundle the changelog fix into the bump PR.

Step 2 — Bump the version (opens a PR to main)

Dispatch the bump-version workflow. It bumps the selected package's package.json + src/version.ts and opens a release-<pkg>-<ver> PR against main:

gh workflow run bump-version.yml --ref main \
  -f package='@clickhouse/client' \
  -f bump_type=patch        # patch | minor | major

Find and watch the resulting PR, get it reviewed, and merge it into main:

gh run list --workflow=bump-version.yml --limit 1
gh pr list --search 'head:release-client-' --state open   # locate the bump PR

Step 3 — Sync release from main (the "release PR")

Open a real PR with base release, head main. This PR is the snapshot that gets reviewed once more as a whole (code-review tooling runs on the full diff):

gh pr create --base release --head main \
  --title "<version> release" \
  --body "Sync release from main to cut <package> <version>."
  • If review surfaces issues, do not push to release. Fix on main via a separate PR (use the fix-release-pr skill), then the release PR picks the fix up automatically. Keep release and main from drifting.
  • When it's clean, merge the release PR.

Step 4 — Watch the automatic head publish + e2e

Merging into release pushes to release, which triggers the path-scoped head publish for each package whose sources changed (publish-client.yml, publish-client-web.yml, publish-client-common.yml). It publishes a head pre-release build (an internal beta — not tracked as a GitHub Release); the client workflows then run an automatic e2e job against the published version and a live ClickHouse — the Node.js client installs it across Node 20/22/24/26 and runs integration tests, and the Web client runs a real-browser smoke (chromium + firefox) via Playwright.

  • These publishes run under the npm-publish environment → they pause for manual approval. Find the run(s) — one per package that changed — give the human the approval link, and wait:
    gh run list --workflow=publish-client.yml --branch release --limit 3
    gh run view <run-id> --json url -q .url      # hand this URL to the human to approve
    gh run watch <run-id>                        # waits for completion
    
    (Use publish-client-web.yml / publish-client-common.yml for the other packages.)
  • The e2e job in publish-client.yml / publish-client-web.yml is sufficient verification — no manual @head testing is needed. Just watch it.
  • If e2e fails: the broken build is already on npm under the head tag. Suggest moving that tag out of the way so nobody installs it, e.g.:
    npm dist-tag add @clickhouse/client@<broken-version> debugging
    
    Then fix forward on main and re-sync release.

Step 5 — Publish to latest (manual dispatch, per package)

Once the head e2e is green, dispatch the package's publish workflow manually from the release branch (no inputs). This builds, publishes to the latest tag (npm OIDC + provenance), and pushes a git tag (client-<ver>, client-web-<ver>, or client-common-<ver>):

gh workflow run publish-client.yml --ref release          # @clickhouse/client
# gh workflow run publish-client-web.yml --ref release     # @clickhouse/client-web
# gh workflow run publish-client-common.yml --ref release  # @clickhouse/client-common (deprecated)

This also runs under npm-publishapproval gate again. Same drill: hand over the approval URL, then watch:

gh run list --workflow=publish-client.yml --branch release --event workflow_dispatch --limit 1
gh run view <run-id> --json url -q .url
gh run watch <run-id>

Repeat for each package being released, dispatching its own workflow.

For the deprecated @clickhouse/client-common, if you ever cut one, also confirm its npm deprecation notice is in place (it should already be).

Step 6 — Create the GitHub Release

The auto-published head betas are not tracked as Releases. The latest release is. Create a GitHub Release from the matching section of that package's CHANGELOG.md (e.g. packages/client-node/CHANGELOG.md for @clickhouse/client).

  • The tag was pushed by the publish workflow: client-<ver> / client-web-<ver> / client-common-<ver> (note: client tags have no v prefix).
  • Use the CHANGELOG section that documents this release as the body. Extract it (the lines under # <version> up to the next # header):
    awk -v v="1.23.0" '$0=="# "v{f=1;next} /^# /{f=0} f' \
      packages/client-node/CHANGELOG.md > /tmp/notes.md
    
  • Do not pass --title — GitHub renders the title from the tag itself. These are not pre-releases, so do not pass --prerelease:
    gh release create client-1.23.0 --notes-file /tmp/notes.md
    
  • Backfill check: sometimes earlier releases are missing their GitHub Release. Compare pushed tags against existing releases and create any that are missing, pulling each one's notes from the matching CHANGELOG section:
    gh release list --limit 50
    git ls-remote --tags origin | grep -oE 'client(-web|-common)?-[0-9.]+'
    

Part B — Standalone packages (@clickhouse/datatype-parser, @clickhouse/rowbinary)

These ship independently. There is no bump-version workflow and no head beta for them.

Step 1 — Bump the version manually on main

Edit the version in the package's package.json (these have no src/version.ts):

  • @clickhouse/datatype-parserpackages/datatype-parser/package.json
  • @clickhouse/rowbinaryskills/clickhouse-js-node-rowbinary/package.json

Do this in a normal PR to main, together with the relevant entry in that package's own CHANGELOG.md (verify the changelog as in Part A, Step 1):

  • @clickhouse/datatype-parserpackages/datatype-parser/CHANGELOG.md
  • @clickhouse/rowbinaryskills/clickhouse-js-node-rowbinary/CHANGELOG.md

Merge it.

Step 2 — Sync release from main

Same as Part A, Step 3: open a base release ← head main PR, get it reviewed as a whole, merge it. (Pushing to release does not auto-publish these standalone packages — only the client packages have an auto head publish.)

Step 3 — Publish to latest (manual dispatch, no inputs)

Dispatch the package's own publish workflow from the release branch. These take no inputs; the release-branch ref is required by the workflow's if guard:

# type parser:
gh workflow run publish-datatype-parser.yml --ref release
# rowbinary codec skill/package:
gh workflow run publish-skill-rowbinary.yml --ref release

Each workflow builds, packs + smoke-tests the tarball, publishes it to latest (OIDC + provenance), pushes a git tag (datatype-parser-v<ver> / rowbinary-v<ver> — note the v prefix here), then runs the e2e job across Node 20/22/24/26.

Same npm-publish approval gate applies — hand over the approval URL and watch:

gh run list --workflow=publish-datatype-parser.yml --branch release --limit 1
gh run view <run-id> --json url -q .url
gh run watch <run-id>

Step 4 — Create the GitHub Release

Same as Part A, Step 6, using the standalone tag (datatype-parser-v<ver> / rowbinary-v<ver>) and the notes from that package's own CHANGELOG.md (packages/datatype-parser/CHANGELOG.md or skills/clickhouse-js-node-rowbinary/CHANGELOG.md). No --title, no --prerelease. Backfill any missing releases.


Quick reference

Client packages:

  1. Verify/fix the top section of the package's own CHANGELOG.md (e.g. packages/client-node/CHANGELOG.md) on main (quick PR if wrong).
  2. gh workflow run bump-version.yml --ref main -f package=… -f bump_type=… → review & merge the bump PR to main.
  3. gh pr create --base release --head main → review the whole release PR → merge.
  4. Auto head publish (per-package publish-client*.yml, path-scoped) + e2e on push to releaseapprove at npm-publish, watch e2e. (If e2e fails: retag the bad build to debugging, fix forward.)
  5. gh workflow run publish-client.yml --ref release (or -web / -common) per package → approve, watch.
  6. gh release create <client-…-tag> --notes-file <changelog-section> (no --title, not prerelease); backfill missing releases.

Standalone packages (datatype-parser / rowbinary):

  1. Bump version in the package's package.json + its own CHANGELOG.md via a PR to main → merge.
  2. gh pr create --base release --head main → review → merge.
  3. gh workflow run publish-datatype-parser.yml --ref release (or publish-skill-rowbinary.yml) → approve, watch.
  4. gh release create <…-v…-tag> --notes-file <changelog-section>; backfill.

Always: per package · each package has its own CHANGELOG.md (root is frozen) · release is protected (fix via main) · every publish needs npm-publish approval · head betas are untracked · GitHub Release notes come from the package's CHANGELOG.md with no explicit --title.

来自 clickhouse 的更多技能

clickhouse-best-practices
clickhouse
28条ClickHouse最佳实践规则,按模式设计、查询优化和数据摄入策略组织。涵盖三个关键领域:主键与数据类型选择(不可变设计决策)、JOIN与查询优化、批量插入与避免突变。包含28条按影响程度排序的规则,其中模式设计和查询优化规则因ClickHouse的列式存储和稀疏索引机制被标记为关键。提供结构化审查流程用于...
official
clickhouse-js-node-coding
clickhouse
Write idiomatic application code with the ClickHouse Node.js client (`@clickhouse/client`). Use this skill whenever a user is *building* against the Node.js…
official
clickhousectl-cloud-deploy
clickhouse
当用户希望将ClickHouse部署到云端、投入生产环境、使用ClickHouse Cloud、托管托管式ClickHouse服务,或从本地迁移时使用…
official
clickstack-otel-collector
clickhouse
Use when a user wants to wire an OpenTelemetry collector into a Managed ClickStack service on ClickHouse Cloud, either by deploying a new local collector…
official
infra-clickhouse
clickhouse
Sets up and manages ClickHouse using the clickhousectl CLI — installs and runs a local ClickHouse server for development, and creates managed ClickHouse Cloud…
official
infra-postgres
clickhouse
Sets up and manages Postgres using the clickhousectl CLI — runs a local Docker-backed Postgres for development, and creates and operates managed ClickHouse…
official
clickhouse-best-practices
clickhouse
在审查ClickHouse模式、查询或配置时必须使用。包含31条在提供建议前必须检查的规则。始终阅读…
official
setup
clickhouse
引导用户完成此插件附带的ClickHouse MCP服务器连接的设置。在用户首次安装插件或遇到问题时使用…
official