Compare commits

...
Author SHA1 Message Date
community-scripts-pr-app[bot]andgithub-actions[bot] c1e45c1c51 Update CHANGELOG.md (#18599)
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
2026-11-10 18:36:57 +00:00
Rögl-Brunner Michel e1dd96614d pr-script-test: approve migrations only when they change nothing else
Anyone with write access can push to a push-app-to-main branch, and the
bot's approval counts toward the main ruleset's required review. The
approval now only covers what a migration contains: if the PR changes
anything besides ct/*.sh and install/*.sh (workflows, misc/, ...), it is
tested as usual but not approved, and the result comment says why.
2026-11-10 16:34:29 +01:00
Rögl-Brunner Michel cfada6ed6a pr-script-test: run on DevScripts migrations by itself, approve passes
PRs that the push-app-to-main app opens on a branch of this repository
(scripts moving over from DevScripts, reviewed and tested there,
writable only by writers) are now built on pull_request_target for
every push, pinned to the pushed commit; an event for an older push is
ignored. When every script passes, the bot approves the tested commit,
which counts toward the main ruleset's required review; a new push
drops it as stale, and a later run that does not fully pass withdraws
it. Untestable (CPU) and setup errors are never approved. All other
PRs still need /test and are never approved.
2026-11-10 16:32:40 +01:00
Rögl-Brunner Michel cc27caf470 pr-script-test: say why nothing was built
A /test on a PR that only changes vm/ (#18578, TrueNAS) found nothing
to build and stopped without a word. The gate now always answers:
"Nothing to build" with the reason for each file (vm/ not covered,
tools/ not built, install/ script without a ct/ script of the same
name, more than 10 scripts), and when some scripts are built the rest
are listed as "Not built" in the progress and the result comment.
File names from the PR are put in code spans with backticks and line
breaks removed.
2026-11-10 16:24:59 +01:00
community-scripts-pr-app[bot]andgithub-actions[bot] dc9d75a5ad Update CHANGELOG.md (#18597)
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
2026-11-10 15:22:34 +00:00
Michel Roegl-Brunner 79481f3a74 pr-script-test: pre-answer prompt variables built in a loop (#18596)
OpenCloud reads var_opencloud_${server}_fqdn for each server in
servers=("opencloud" "collabora" "wopi"), so the scan for literal
var_ names found nothing and the build stopped at the first prompt
(#18586). Templates of the form var_...${x}... are now expanded with
the words of `for x in ...`, or of the array that loop walks; nothing
is executed. Hostname-like answers are derived from the variable name
(ci-opencloud-collabora.example.test), so several cannot collide.
2026-11-10 16:22:06 +01:00
community-scripts-pr-app[bot]andgithub-actions[bot] bd7f6f60bd Update CHANGELOG.md (#18595)
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
2026-11-10 15:05:16 +00:00
Michel Roegl-Brunner db30bcd960 Build PR scripts on a Proxmox test host via /test (#18594)
A maintainer comments `/test` on a pull request and a self-hosted
runner on a disposable Proxmox VE host builds every ct/ script the PR
touches (and the ct/ entry point of every changed install/ script),
checks the result and reports back on the PR.

- Gate: only users with write access; builds the latest commit only if
  nothing was pushed to the fork after the comment, or a pinned
  `/test <full sha>`.
- Build: the PR tree at that commit is served from the test host with
  every raw.githubusercontent.com/community-scripts/ProxmoxVE/main URL
  rewritten to it, so engine changes in the PR are tested too.
  Unattended (mode=default); prompts declared the AGENTS.md way are
  pre-answered with CI values.
- Checks: build and install succeed, container running, no failed
  units, the URL the script prints answers 200. Screenshot of that URL.
- Report: check run, one comment (results, screenshot, install log,
  mentions of the PR author and whoever asked), changes requested on a
  real failure, dismissed again on a pass. CPU-feature failures (no
  AVX2 on the test host) are marked untestable, not failed.
- The test job has a read-only token and no secrets; its runner carries
  only the proxmox-test label. Container IDs 9050-9099.

Same file as in DevScripts; github.repository selects the URLs and the
container ID range.
2026-11-10 16:04:32 +01:00
4 changed files with 1232 additions and 1 deletions

No files matched your search

+571
View File
@@ -0,0 +1,571 @@
name: PR script test
# Builds the ct/ scripts a pull request touches on a real Proxmox host and
# reports the outcome on the PR, as a check run and as one comment kept up to
# date.
#
# A self-hosted runner that executes pull request code as root is only
# acceptable if a human decides what runs on it. So it starts on a push only
# for trusted PRs: those the push-app-to-main app opens on a branch of this
# repository when a script moves over from DevScripts (already reviewed and
# tested there; only writers can push to such a branch). Those are built
# automatically and approved when every script passes. Everything else waits
# for someone with write access to comment `/test` (optionally followed by a
# full SHA to pin, and the ct/ paths to build). Only a commit that existed when
# the comment was written gets built: a push after it is never covered, not
# even one that lands in the seconds before this job runs.
#
# gate ubuntu-latest who asked, which scripts, which commit
# test proxmox-test build + checks, one matrix job per script/OS
# report ubuntu-latest check run + PR comment
#
# The test job gets a read-only token and never sees a secret: the host it runs
# on executes untrusted code and has to be treated as compromised.
#
# This file is identical in DevScripts and ProxmoxVE; github.repository decides
# which URLs get rewritten and which container ID range is used.
on:
issue_comment:
types: [created]
# Trusted PRs only, see above. pull_request_target runs this file from main,
# never the PR's copy of it.
pull_request_target:
types: [opened, reopened, synchronize, ready_for_review]
paths:
- "ct/**"
- "install/**"
workflow_dispatch:
inputs:
pr:
description: "Pull request number"
required: true
type: number
sha:
description: "Pin a commit (full 40-character SHA); empty = the latest, if nothing was pushed since"
required: false
type: string
scripts:
description: "Space-separated ct/ paths (empty = everything the PR touches)"
required: false
type: string
permissions: {}
jobs:
gate:
if: >-
github.event_name == 'workflow_dispatch' ||
(github.event_name == 'issue_comment' && github.event.issue.pull_request && startsWith(github.event.comment.body, '/test')) ||
(github.event_name == 'pull_request_target' && !github.event.pull_request.draft &&
github.event.pull_request.user.login == 'push-app-to-main[bot]' &&
github.event.pull_request.head.repo.full_name == github.repository)
runs-on: ubuntu-latest
permissions:
actions: read # when the run was dispatched, for a /test without SHA
checks: write
contents: read
issues: write
pull-requests: write
outputs:
matrix: ${{ steps.plan.outputs.matrix }}
pr: ${{ steps.plan.outputs.pr }}
repo: ${{ steps.plan.outputs.repo }}
sha: ${{ steps.plan.outputs.sha }}
check_id: ${{ steps.plan.outputs.check_id }}
actor: ${{ steps.plan.outputs.actor }}
not_built: ${{ steps.plan.outputs.not_built }}
trusted: ${{ steps.plan.outputs.trusted }}
steps:
- id: plan
uses: actions/github-script@v8
with:
script: |
const { owner, repo } = context.repo;
const MAX = 10;
core.setOutput('matrix', '[]');
// Trusted: a PR the push-app-to-main app opened on a branch of this
// repository, i.e. a script moving over from DevScripts.
const TRUSTED_AUTHOR = 'push-app-to-main[bot]';
const isTrusted = p => !!p && p.user && p.user.login === TRUSTED_AUTHOR && p.user.type === 'Bot' &&
!!p.head.repo && p.head.repo.full_name === `${owner}/${repo}`;
let prNumber, requested = '';
if (context.eventName === 'pull_request_target') {
const p = context.payload.pull_request;
if (!isTrusted(p)) return;
prNumber = p.number;
// Exactly the pushed commit; a later push starts its own run.
requested = p.head.sha;
} else if (context.eventName === 'workflow_dispatch') {
prNumber = Number(context.payload.inputs.pr);
requested = `${context.payload.inputs.sha || ''} ${context.payload.inputs.scripts || ''}`;
} else {
prNumber = context.payload.issue.number;
const body = context.payload.comment.body.trim();
if (!/^\/test(\s|$)/.test(body)) return;
// author_association hides private org members, so ask for the permission itself.
const user = context.payload.comment.user.login;
const { data: perm } = await github.rest.repos.getCollaboratorPermissionLevel({ owner, repo, username: user });
if (!['admin', 'maintain', 'write'].includes(perm.permission)) {
core.notice(`${user} has ${perm.permission} access; /test needs write`);
return;
}
await github.rest.reactions.createForIssueComment({
owner, repo, comment_id: context.payload.comment.id, content: 'rocket',
});
requested = body.split('\n')[0].slice('/test'.length);
}
const { data: pr } = await github.rest.pulls.get({ owner, repo, pull_number: prNumber });
if (pr.state !== 'open' || !pr.head.repo) { core.notice('PR is closed or its fork is gone'); return; }
const head = pr.head.repo;
if (context.eventName === 'pull_request_target' && pr.head.sha !== requested) {
core.notice('Pushed to again since; the newer push has its own run');
return;
}
core.setOutput('trusted', isTrusted(pr) ? 'true' : 'false');
const MARK = '<!-- pr-script-test -->';
const sticky = async body => {
const comments = await github.paginate(github.rest.issues.listComments, { owner, repo, issue_number: prNumber });
// Only our own comment: anyone can paste the marker into theirs.
const mine = comments.find(c => c.body.includes(MARK) && c.user.type === 'Bot');
if (mine) await github.rest.issues.updateComment({ owner, repo, comment_id: mine.id, body });
else await github.rest.issues.createComment({ owner, repo, issue_number: prNumber, body });
};
// Which commit the approval covers. Reading pr.head.sha alone would
// build whatever was pushed between the comment and this job
// starting, i.e. code nobody looked at. Two ways to be sure:
//
// /test <sha> that exact commit, and only while it is the head. A
// full SHA: the author writes both commits, so a short
// prefix is a quick birthday collision away.
// /test the latest commit, if nothing reached the fork after
// the comment. pushed_at is set by GitHub on every push
// and read after the PR, so a push in between shows up.
// Branches of this repository belong to people with
// write access, who could build anything anyway.
const tokens = requested.split(/\s+/).filter(Boolean);
const isSha = t => /^[0-9a-f]{7,40}$/.test(t);
const approved = tokens.filter(isSha);
let entries = tokens.filter(t => !isSha(t));
const refuse = async (why, retry) => {
await sticky([MARK, '### Proxmox script test', '', `⚠️ ${why} Nothing was built.`, '', retry].join('\n'));
core.notice(why);
};
if (approved.length > 1) {
await refuse('`/test` takes at most one commit.', 'Comment `/test` to build the latest commit.');
return;
}
if (approved.length === 1 && approved[0] !== pr.head.sha) {
await refuse(
approved[0].length !== 40 && pr.head.sha.startsWith(approved[0])
? 'A short SHA is not enough, it has to be all 40 characters.'
: `\`${approved[0]}\` is not the head of this PR (any more).`,
'Comment `/test` to build the latest commit.');
return;
}
if (approved.length === 0 && head.full_name !== `${owner}/${repo}`) {
const askedAt = context.eventName === 'workflow_dispatch'
? (await github.rest.actions.getWorkflowRun({ owner, repo, run_id: context.runId })).data.created_at
: context.payload.comment.created_at;
const { data: fork } = await github.rest.repos.get({ owner: head.owner.login, repo: head.name });
if (new Date(fork.pushed_at) >= new Date(askedAt)) {
await refuse(
`Something was pushed to ${head.full_name} after the \`/test\`, so it is not certain which commit was meant.`,
'Comment `/test` again once you have looked at the latest commit.');
return;
}
}
if (entries.some(e => !/^ct\/[A-Za-z0-9._-]+\.sh$/.test(e))) {
core.setFailed(`Only ct/<name>.sh paths can be requested, got: ${requested}`);
return;
}
// Everything the PR changes that this test cannot build is named in
// the comment, so a /test never ends in silence.
const notBuilt = [];
const code = s => '`' + String(s).replace(/[`\r\n]/g, '') + '`';
if (entries.length === 0) {
const files = await github.paginate(github.rest.pulls.listFiles, { owner, repo, pull_number: prNumber });
const set = new Set();
for (const f of files) {
if (f.status === 'removed') continue;
let m;
if ((m = f.filename.match(/^install\/([A-Za-z0-9._-]+)-install\.sh$/))) set.add(`ct/${m[1]}.sh`);
else if (/^ct\/[A-Za-z0-9._-]+\.sh$/.test(f.filename)) set.add(f.filename);
else if (/^vm\/.*\.sh$/.test(f.filename)) notBuilt.push(`${code(f.filename)}: \`vm/\` scripts are not covered`);
else if (/^(tools|turnkey)\/.*\.sh$/.test(f.filename)) notBuilt.push(`${code(f.filename)}: only \`ct/\` scripts are built`);
}
entries = [...set].sort();
}
const include = [];
for (const script of entries.slice(0, MAX)) {
let text;
try {
const res = await github.rest.repos.getContent({ owner: head.owner.login, repo: head.name, path: script, ref: pr.head.sha });
text = Buffer.from(res.data.content || '', 'base64').toString('utf8');
} catch (e) {
notBuilt.push(`\`${script}\` does not exist in this PR (an \`install/\` script is built through the \`ct/\` script of the same name)`);
continue;
}
// Scripts that ask for the OS get built once per OS.
const oses = /msg_menu\s+"Choose the container OS"/.test(text) ? ['debian', 'alpine'] : [''];
const name = script.slice(3, -3);
for (const os of oses) include.push({ script, os, name: os ? `${name}-${os}` : name });
}
for (const script of entries.slice(MAX)) notBuilt.push(`\`${script}\`: at most ${MAX} scripts per run, name it with \`/test ${script}\``);
if (include.length === 0) {
await sticky([
MARK, '### Proxmox script test', '',
'ℹ️ Nothing to build: this test builds `ct/` scripts (and changed `install/` scripts through their `ct/` script).', '',
...(notBuilt.length ? notBuilt.map(n => `- ${n}`) : ['- this PR changes no `ct/` or `install/` script']),
].join('\n'));
core.notice('No ct/ script to build');
return;
}
const runUrl = `${context.serverUrl}/${owner}/${repo}/actions/runs/${context.runId}`;
const { data: check } = await github.rest.checks.create({
owner, repo, name: 'Proxmox script test', head_sha: pr.head.sha, status: 'in_progress',
details_url: runUrl,
output: { title: `Building ${include.length} on Proxmox`, summary: include.map(i => `- \`${i.script}\` ${i.os}`).join('\n') },
});
core.setOutput('matrix', JSON.stringify(include));
core.setOutput('pr', String(prNumber));
core.setOutput('repo', head.full_name);
core.setOutput('sha', pr.head.sha);
core.setOutput('check_id', String(check.id));
core.setOutput('not_built', JSON.stringify(notBuilt));
// Who asked: the /test commenter, or whoever dispatched the run.
core.setOutput('actor', context.actor);
await sticky([
MARK, '### Proxmox script test', '',
`⏳ Building ${include.length} at ${pr.head.sha.slice(0, 7)} — [watch it live](${runUrl}) (open the \`test\` job)`, '',
...include.map(i => `- \`${i.script}\`${i.os ? ` (${i.os})` : ''}`),
...(notBuilt.length ? ['', '**Not built:**', ...notBuilt.map(n => `- ${n}`)] : []),
].join('\n'));
test:
needs: gate
if: needs.gate.outputs.matrix != '[]' && needs.gate.outputs.matrix != ''
# Only `proxmox-test`, never `self-hosted`: the runner is registered without
# default labels so that no other workflow can be scheduled onto it.
runs-on: proxmox-test
permissions:
contents: read
timeout-minutes: 60
strategy:
fail-fast: false
matrix:
include: ${{ fromJSON(needs.gate.outputs.matrix) }}
name: test (${{ matrix.name }})
steps:
# The default branch, never the PR: the harness itself must be trusted.
- uses: actions/checkout@v5
with:
sparse-checkout: .github/workflows/scripts
persist-credentials: false
- name: Build and check ${{ matrix.script }}
env:
PR_REPO: ${{ needs.gate.outputs.repo }}
PR_SHA: ${{ needs.gate.outputs.sha }}
SCRIPT: ${{ matrix.script }}
OS: ${{ matrix.os }}
UPSTREAM: ${{ github.repository }}
CTID_BASE: ${{ github.repository == 'community-scripts/ProxmoxVE' && '9050' || '9000' }}
OUT_DIR: ${{ runner.temp }}/result
run: bash .github/workflows/scripts/pve-run-script.sh
- if: always()
uses: actions/upload-artifact@v4
with:
name: result-${{ matrix.name }}
path: ${{ runner.temp }}/result
if-no-files-found: ignore
retention-days: 14
report:
needs: [gate, test]
if: always() && needs.gate.result == 'success' && needs.gate.outputs.matrix != '[]' && needs.gate.outputs.matrix != ''
runs-on: ubuntu-latest
permissions:
actions: read
checks: write
contents: write # the ci-screenshots branch only; main is protected
issues: write
pull-requests: write
steps:
- uses: actions/download-artifact@v5
with:
pattern: result-*
path: results
- uses: actions/github-script@v8
env:
MATRIX: ${{ needs.gate.outputs.matrix }}
PR: ${{ needs.gate.outputs.pr }}
SHA: ${{ needs.gate.outputs.sha }}
CHECK_ID: ${{ needs.gate.outputs.check_id }}
TEST_RESULT: ${{ needs.test.result }}
ACTOR: ${{ needs.gate.outputs.actor }}
NOT_BUILT: ${{ needs.gate.outputs.not_built }}
TRUSTED: ${{ needs.gate.outputs.trusted }}
with:
script: |
const fs = require('fs');
const { owner, repo } = context.repo;
const runUrl = `${context.serverUrl}/${owner}/${repo}/actions/runs/${context.runId}`;
const matrix = JSON.parse(process.env.MATRIX);
// download-artifact unpacks a single artifact straight into results/
// and several into results/<name>/, so look for every result.json
// and match it by what it says it is.
const path = require('path');
const found = new Map();
const walk = dir => {
if (!fs.existsSync(dir)) return;
for (const e of fs.readdirSync(dir, { withFileTypes: true })) {
const p = path.join(dir, e.name);
if (e.isDirectory()) walk(p);
else if (e.name === 'result.json') {
try {
const r = JSON.parse(fs.readFileSync(p, 'utf8'));
// Written on the host that ran the PR's code: anything but a
// known status is an error, never something new.
if (!['pass', 'fail', 'unsupported', 'error'].includes(r.status)) r.status = 'error';
const log = path.join(dir, 'install.log');
r.install_log = fs.existsSync(log) ? fs.readFileSync(log, 'utf8') : '';
const shot = path.join(dir, 'screenshot.png');
r.shot_file = fs.existsSync(shot) ? shot : '';
found.set(`${r.script}|${r.os}`, r);
} catch (err) { core.warning(`unreadable ${p}: ${err.message}`); }
}
}
};
walk('results');
// A job that died before writing result.json still gets a row.
const rows = matrix.map(m => ({ ...(found.get(`${m.script}|${m.os || 'default'}`) || {
script: m.script, os: m.os || 'default',
status: process.env.TEST_RESULT === 'cancelled' ? 'cancelled' : 'error',
step: 'no result (runner or job failure)', duration: 0, log_tail: '', install_log: '', shot_file: '',
}), name: m.name }));
// Screenshots go to an orphan branch, because a comment can only show
// an image that has a URL. One file per PR and script, overwritten on
// every run; the link names the commit, so no cache shows an old one.
// The PNG comes from the host that ran the PR's code: check it again.
const BRANCH = 'ci-screenshots';
const prNum = Number(process.env.PR);
const isPng = buf => buf.length > 8 && buf.length < 3000000 &&
buf.subarray(0, 8).equals(Buffer.from([0x89, 0x50, 0x4e, 0x47, 0x0d, 0x0a, 0x1a, 0x0a]));
const ensureBranch = async () => {
try { await github.rest.git.getRef({ owner, repo, ref: `heads/${BRANCH}` }); return; }
catch (e) { if (e.status !== 404) throw e; }
const { data: blob } = await github.rest.git.createBlob({ owner, repo, encoding: 'utf-8',
content: 'Screenshots taken by pr-script-test.yml for the PR comments. Not code; never merge this branch.\n' });
const { data: tree } = await github.rest.git.createTree({ owner, repo, tree: [{ path: 'README.md', mode: '100644', type: 'blob', sha: blob.sha }] });
const { data: commit } = await github.rest.git.createCommit({ owner, repo, message: 'Start ci-screenshots', tree: tree.sha, parents: [] });
await github.rest.git.createRef({ owner, repo, ref: `refs/heads/${BRANCH}`, sha: commit.sha });
};
for (const r of rows.filter(r => r.shot_file)) {
const buf = fs.readFileSync(r.shot_file);
if (!isPng(buf)) { core.warning(`${r.script}: screenshot is not a plausible PNG, skipped`); continue; }
const file = `pr-${prNum}/${r.name}.png`;
for (let attempt = 1; attempt <= 3 && !r.shot_url; attempt++) {
try {
await ensureBranch();
let sha;
try { sha = (await github.rest.repos.getContent({ owner, repo, path: file, ref: BRANCH })).data.sha; }
catch (e) { if (e.status !== 404) throw e; }
const { data } = await github.rest.repos.createOrUpdateFileContents({
owner, repo, branch: BRANCH, path: file, sha, content: buf.toString('base64'),
message: `Screenshot: ${r.script} (${r.os}) for #${prNum} at ${process.env.SHA.slice(0, 7)}`,
});
r.shot_url = `https://raw.githubusercontent.com/${owner}/${repo}/${data.commit.sha}/${file}`;
} catch (e) {
// Two reports at once race on the branch head; try again.
core.warning(`${r.script}: screenshot upload attempt ${attempt} failed: ${e.message}`);
await new Promise(res => setTimeout(res, 2000 * attempt));
}
}
}
// unsupported: the script needs a CPU feature the test host lacks
// (AVX2 and the like). Not the PR's fault, so not a failure.
const icon = { pass: '✅', fail: '❌', error: '⚠️', cancelled: '⏹️', unsupported: '🖥️' };
const dur = s => `${Math.floor(s / 60)}m${String(s % 60).padStart(2, '0')}s`;
// Text from the build lands outside code blocks here: no table breaks,
// no code spans, and no @mention of a stranger (zero-width space).
const clean = s => String(s || '').replace(/[|\r\n]/g, ' ').replace(/`/g, "'").replace(/@/g, '@​').slice(0, 200);
const fence = s => String(s || '').replace(/~~~~/g, '~~~');
// fail: the app broke, the PR's business. error/cancelled: our test
// setup did, not the PR's. unsupported: the host's CPU is too old.
const count = s => rows.filter(r => r.status === s).length;
const passed = count('pass');
const untestable = count('unsupported');
const broken = count('fail');
const setupTrouble = count('error') + count('cancelled');
const allPass = passed === rows.length;
const cancelled = rows.every(r => r.status === 'cancelled');
const of = `**${passed}/${rows.length} passed**`;
const headline = broken
? `❌ ${of}`
: setupTrouble
? `⚠️ ${of}, ${setupTrouble} could not be tested because of a problem with the test setup (not the PR)`
: untestable
? `🖥️ ${of}, ${untestable} need${untestable === 1 ? 's' : ''} a newer CPU than the test host has (nothing failed)`
: `✅ ${of}`;
const table = [
'| Script | OS | Result | Duration | Detail |',
'|---|---|---|---|---|',
...rows.map(r => `| \`${r.script}\` | ${r.os} | ${icon[r.status] || '❔'} | ${dur(r.duration || 0)} | ${r.status === 'pass' ? (r.url ? `${clean(r.url)} → ${r.http_code}` : 'no URL to check') : clean(r.step)} |`),
].join('\n');
// A comment holds 65536 characters. Each script gets an equal share
// and a log that does not fit loses its beginning: the end is where
// a failure shows.
const budget = Math.floor(52000 / rows.length);
const tail = (text, max) => {
text = fence(text).trimEnd();
if (text.length <= max) return text;
const kept = text.slice(text.length - max);
const cut = text.slice(0, text.length - max).split('\n').length;
return `… ${cut} earlier lines cut, the full log is in the run artifacts\n${kept.slice(kept.indexOf('\n') + 1)}`;
};
const details = (summaryLine, text, max) => [
`<details><summary>${summaryLine}</summary>`, '',
'~~~~text', tail(text, max), '~~~~', '</details>',
].join('\n');
const logs = rows.map(r => {
const parts = [];
const failed = r.status !== 'pass';
const buildShare = failed && r.log_tail ? Math.min(r.log_tail.length, Math.floor(budget / 3)) : 0;
if (buildShare) parts.push(details(`${icon[r.status] || '❔'} ${r.script} (${r.os}) — build log`, r.log_tail, buildShare));
if (r.install_log) {
const lines = r.install_log.trimEnd().split('\n').length;
parts.push(details(`📄 ${r.script} (${r.os}) — install log (${lines} lines)`, r.install_log, budget - buildShare - 300));
} else if (!buildShare) {
parts.push(`<sub>${r.script} (${r.os}): no install log was captured</sub>`);
}
return parts.join('\n\n');
});
const shots = rows.filter(r => r.shot_url).map(r => [
`<details open><summary>📸 ${r.script} (${r.os}) — ${clean(r.url)}</summary>`, '',
`<img src="${r.shot_url}" width="640" alt="Screenshot of ${r.script}">`, '', '</details>',
].join('\n'));
const summary = [
`${headline} at \`${process.env.SHA.slice(0, 7)}\``, '',
table, '',
...shots, '',
...(() => { try { const n = JSON.parse(process.env.NOT_BUILT || '[]'); return n.length ? ['**Not built:**', ...n.map(x => `- ${x}`), ''] : []; } catch (e) { return []; } })(),
...logs, '',
`Full logs: [workflow run](${runUrl}) (artifacts \`result-*\`). Comment \`/test\` to build again.`,
].join('\n');
await github.rest.checks.update({
owner, repo, check_run_id: Number(process.env.CHECK_ID), status: 'completed',
// Untestable here is "test it by hand", not a pass: action_required,
// because a required check would count neutral as passed.
conclusion: cancelled ? 'cancelled' : allPass ? 'success' : broken ? 'failure' : 'action_required',
output: { title: `${passed}/${rows.length} passed`, summary: summary.slice(0, 65000) },
});
// The result pings whoever started the test. A mention added by
// editing a comment does not reliably notify, so the old comment
// goes and the result is posted as a new one, at the end of the
// thread where it is looked for.
// Whoever started the test and the PR's author, once each; no bots.
const pr = Number(process.env.PR);
const { data: prInfo } = await github.rest.pulls.get({ owner, repo, pull_number: pr });
const author = prInfo.user && prInfo.user.type === 'User' ? prInfo.user.login : '';
const login = /^[A-Za-z0-9](?:[A-Za-z0-9-]{0,38})$/;
const ping = [...new Set([process.env.ACTOR || '', author])].filter(l => login.test(l)).map(l => `@${l} `).join('');
const MARK = '<!-- pr-script-test -->';
const body = `${MARK}\n### Proxmox script test\n\n${ping}${summary}`.slice(0, 65000);
const comments = await github.paginate(github.rest.issues.listComments, { owner, repo, issue_number: pr });
for (const c of comments.filter(c => c.body.includes(MARK) && c.user.type === 'Bot')) {
await github.rest.issues.deleteComment({ owner, repo, comment_id: c.id });
}
const { data: posted } = await github.rest.issues.createComment({ owner, repo, issue_number: pr, body });
// A real failure requests changes on the tested commit. Untestable
// (CPU), runner errors and cancelled runs are not the PR's fault and
// leave the review state alone. Our previous "changes requested" is
// always dismissed first: a pass lifts it, a new failure replaces it.
const REVIEW_MARK = '<!-- pr-script-test-review -->';
const failedRows = cancelled ? [] : rows.filter(r => r.status === 'fail');
const sha7 = process.env.SHA.slice(0, 7);
// Trusted PRs (scripts moving over from DevScripts) are approved when
// every script passes -- not on untestable or setup trouble, which
// means "not tested", not "good". That approval counts toward the
// ruleset's required review; a push dismisses it as stale, and a
// later run that does not fully pass withdraws it.
// Anyone with write access can push to a migration branch, and the
// approval counts toward the required review. So it only covers
// what a migration contains: ct/ and install/ scripts. A PR that
// also changes anything else (workflows, misc/, ...) is tested but
// never approved.
let approve = process.env.TRUSTED === 'true' && allPass;
let noApproveWhy = '';
if (approve) {
const changed = await github.paginate(github.rest.pulls.listFiles, { owner, repo, pull_number: pr });
const other = changed.map(f => f.filename).filter(n => !/^(ct|install)\/[A-Za-z0-9._-]+\.sh$/.test(n));
if (other.length) {
approve = false;
noApproveWhy = `Not approved automatically: besides \`ct/\` and \`install/\` scripts this PR changes ${other.slice(0, 5).map(n => '`' + n.replace(/[`\r\n]/g, '') + '`').join(', ')}${other.length > 5 ? ', …' : ''}. That needs a human review.`;
}
}
if (noApproveWhy) {
await github.rest.issues.updateComment({ owner, repo, comment_id: posted.id, body: `${body}\n\n> ${noApproveWhy}`.slice(0, 65000) });
}
try {
const reviews = await github.paginate(github.rest.pulls.listReviews, { owner, repo, pull_number: pr });
const ours = state => reviews.filter(v => v.user && v.user.type === 'Bot' && v.state === state && (v.body || '').includes(REVIEW_MARK));
for (const v of ours('CHANGES_REQUESTED')) {
await github.rest.pulls.dismissReview({
owner, repo, pull_number: pr, review_id: v.id,
message: failedRows.length ? 'Replaced by a newer Proxmox script test.' : `Proxmox script test passed at ${sha7}.`,
});
}
if (!approve && !cancelled) {
for (const v of ours('APPROVED')) {
await github.rest.pulls.dismissReview({
owner, repo, pull_number: pr, review_id: v.id,
message: `Withdrawn: the Proxmox script test at ${sha7} did not fully pass.`,
});
}
}
if (approve && !ours('APPROVED').some(v => v.commit_id === process.env.SHA)) {
await github.rest.pulls.createReview({
owner, repo, pull_number: pr, commit_id: process.env.SHA, event: 'APPROVE',
body: [
REVIEW_MARK, `**Proxmox script test passed** at \`${sha7}\`: ${rows.map(r => `\`${r.script}\``).join(', ')}.`, '',
`Script moved over from DevScripts, built and checked on a real Proxmox host. [Test result](${posted.html_url}).`,
].join('\n'),
});
}
if (failedRows.length) {
await github.rest.pulls.createReview({
owner, repo, pull_number: pr, commit_id: process.env.SHA, event: 'REQUEST_CHANGES',
body: [
REVIEW_MARK, `**Proxmox script test failed** at \`${sha7}\`:`, '',
...failedRows.map(r => `- \`${r.script}\` (${r.os}): ${clean(r.step)}`), '',
`Logs and details: [test result](${posted.html_url}). Push a fix, then ask for \`/test\` again.`,
].join('\n'),
});
}
} catch (e) {
core.warning(`Could not update the review state: ${e.message}`);
}
+203
View File
@@ -0,0 +1,203 @@
#!/usr/bin/env bash
# One-time setup of the Proxmox test host behind pr-script-test.yml.
# Safe to re-run. Run as root on the test host (a disposable PVE, never a
# production node: it builds pull request code as root).
#
# pve-ci-host-setup.sh host only
# pve-ci-host-setup.sh <owner/repo> <token> host + a runner for that repo
#
# The token is a runner *registration* token (valid one hour):
# gh api -X POST repos/<owner/repo>/actions/runners/registration-token --jq .token
set -euo pipefail
RUNNER_VERSION="${RUNNER_VERSION:-2.338.2}"
SERVE_ROOT=/srv/ci
SERVE_PORT=18080
UPLINK="${UPLINK:-vmbr0}"
export DEBIAN_FRONTEND=noninteractive
apt-get update -qq
apt-get install -y -qq jq curl python3 dnsmasq iptables chromium fonts-dejavu fonts-noto-color-emoji >/dev/null
# Screenshots of the app under test are taken as this user, never as root.
id ci-shot &>/dev/null || useradd --system --create-home --shell /usr/sbin/nologin ci-shot
# The page Chromium renders is served by code under test, and a redirect or an
# <img> could point it at this host (the PVE UI on :8006) or the lab network;
# the screenshot would then publish that on the PR. So ci-shot may reach the
# test containers, DNS and the public internet (apps load fonts and scripts
# from CDNs), and nothing private.
cat >/usr/local/sbin/ci-shot-firewall <<'EOF'
#!/usr/bin/env bash
set -euo pipefail
uid=$(id -u ci-shot)
for fw in iptables ip6tables; do
$fw -w -N CI_SHOT 2>/dev/null || $fw -w -F CI_SHOT
$fw -w -C OUTPUT -m owner --uid-owner "$uid" -j CI_SHOT 2>/dev/null ||
$fw -w -I OUTPUT -m owner --uid-owner "$uid" -j CI_SHOT
done
iptables -w -A CI_SHOT -d 10.99.0.1 -j REJECT
iptables -w -A CI_SHOT -d 10.99.0.0/24 -j ACCEPT
for ns in $(awk '/^nameserver/ && $2 !~ /:/ {print $2}' /etc/resolv.conf); do
iptables -w -A CI_SHOT -d "$ns" -p udp --dport 53 -j ACCEPT
iptables -w -A CI_SHOT -d "$ns" -p tcp --dport 53 -j ACCEPT
done
for net in 0.0.0.0/8 10.0.0.0/8 100.64.0.0/10 127.0.0.0/8 169.254.0.0/16 172.16.0.0/12 192.168.0.0/16 224.0.0.0/3; do
iptables -w -A CI_SHOT -d "$net" -j REJECT
done
for net in ::1/128 fc00::/7 fe80::/10 ff00::/8; do
ip6tables -w -A CI_SHOT -d "$net" -j REJECT
done
EOF
chmod +x /usr/local/sbin/ci-shot-firewall
cat >/etc/systemd/system/ci-shot-firewall.service <<'EOF'
[Unit]
Description=Keep the screenshot browser away from this host and the lab network
After=network-online.target
[Service]
Type=oneshot
RemainAfterExit=yes
ExecStart=/usr/local/sbin/ci-shot-firewall
[Install]
WantedBy=multi-user.target
EOF
# Containers get their own NATed bridge instead of the lab network: no DHCP
# lease surprises (a lab lease once handed a container the host's own address),
# and an app listening on 8080 can never be confused with this host.
cat >/etc/network/interfaces.d/ci-nat <<EOF
auto vmbr1
iface vmbr1 inet static
address 10.99.0.1/24
bridge-ports none
bridge-stp off
bridge-fd 0
post-up sysctl -qw net.ipv4.ip_forward=1
post-up iptables -t nat -C POSTROUTING -s 10.99.0.0/24 -o ${UPLINK} -j MASQUERADE 2>/dev/null || iptables -t nat -A POSTROUTING -s 10.99.0.0/24 -o ${UPLINK} -j MASQUERADE
post-down iptables -t nat -D POSTROUTING -s 10.99.0.0/24 -o ${UPLINK} -j MASQUERADE || true
EOF
ifreload -a
cat >/etc/dnsmasq.d/ci.conf <<'EOF'
interface=vmbr1
bind-interfaces
except-interface=lo
dhcp-range=10.99.0.100,10.99.0.250,255.255.255.0,2h
dhcp-option=option:router,10.99.0.1
dhcp-option=option:dns-server,10.99.0.1
no-resolv
EOF
# Upstream = whatever this host resolves with. Debian's dnsmasq would otherwise
# ask resolvconf, which has nothing to offer on a PVE host.
awk '/^nameserver/ {print "server=" $2}' /etc/resolv.conf >>/etc/dnsmasq.d/ci.conf
systemctl enable dnsmasq >/dev/null
systemctl restart dnsmasq
# A PVE ISO install writes /etc/timezone; a cloud-image install does not, and
# the engine's pct_timezone aborts without it.
[[ -s /etc/timezone ]] || {
timedatectl set-timezone Europe/Vienna
echo Europe/Vienna >/etc/timezone
}
# Containers live on the root filesystem; there is no LVM in a cloud-image PVE.
pvesm set local --content iso,vztmpl,backup,rootdir,images
# Templates are fetched once here rather than on every first build.
pveam update >/dev/null
for t in $(pveam available --section system | awk '/_amd64\./ && /debian-13-standard|alpine-3\.2[0-9]-default|ubuntu-24\.04-standard/ {print $2}'); do
pveam list local | grep -q "$t" || pveam download local "$t" >/dev/null
done
# The rewritten PR trees, served to the engine and to the containers.
mkdir -p "$SERVE_ROOT"
cat >/etc/systemd/system/ci-serve.service <<EOF
[Unit]
Description=Serve pull request trees to script builds
After=network-online.target
[Service]
ExecStart=/usr/bin/python3 -m http.server ${SERVE_PORT} --bind 10.99.0.1 --directory ${SERVE_ROOT}
DynamicUser=yes
Restart=always
[Install]
WantedBy=multi-user.target
EOF
# Leftovers of a job that was killed before its cleanup trap ran.
cat >/usr/local/sbin/ci-sweep <<'EOF'
#!/usr/bin/env bash
# Destroy CI containers (9000-9099) that no running job holds, and stale trees.
for id in $(pct list | awk 'NR>1 && $1>=9000 && $1<=9099 {print $1}'); do
[[ -e "/run/ci-jobs/$id" ]] && continue
pct stop "$id" --skiplock 1 &>/dev/null || true
pct destroy "$id" --purge 1 --force 1 &>/dev/null && echo "swept CT $id"
done
find /srv/ci -mindepth 1 -maxdepth 1 -type d -mmin +180 -exec rm -rf {} +
# Engine logs that a killed job never got to remove.
find /tmp -maxdepth 1 -type f \( -name 'create-lxc-*.log' -o -name 'pct_create_*.log' \
-o -regex '/tmp/.*-9[0-9][0-9][0-9]-[0-9a-f]+\.log' \) -mmin +180 -delete
find /var/log/community-scripts -type f -mtime +2 -delete 2>/dev/null
find /opt/actions-runner/*/_diag -type f -mtime +14 -delete 2>/dev/null
true
EOF
chmod +x /usr/local/sbin/ci-sweep
cat >/etc/systemd/system/ci-sweep.service <<'EOF'
[Unit]
Description=Remove leftover CI containers
[Service]
Type=oneshot
ExecStart=/usr/local/sbin/ci-sweep
EOF
cat >/etc/systemd/system/ci-sweep.timer <<'EOF'
[Unit]
Description=Remove leftover CI containers every 3 hours
[Timer]
OnCalendar=0/3:30
Persistent=true
[Install]
WantedBy=timers.target
EOF
systemctl daemon-reload
systemctl enable --now ci-serve.service ci-sweep.timer ci-shot-firewall.service >/dev/null
systemctl restart ci-serve.service ci-shot-firewall.service
# --- Runner (one per repository) ----------------------------------------------
if [[ $# -eq 2 ]]; then
repo="$1"
token="$2"
[[ "$repo" =~ ^[A-Za-z0-9._-]+/[A-Za-z0-9._-]+$ ]] || { echo "bad repo" >&2; exit 2; }
dir="/opt/actions-runner/${repo#*/}"
mkdir -p "$dir"
cd "$dir"
if [[ ! -x ./config.sh ]]; then
curl -fsSL "https://github.com/actions/runner/releases/download/v${RUNNER_VERSION}/actions-runner-linux-x64-${RUNNER_VERSION}.tar.gz" | tar -xz
./bin/installdependencies.sh >/dev/null
fi
# A registered runner is reconfigured from scratch: stop and remove the old
# registration (the token works for removal too).
if [[ -f .runner ]]; then
./svc.sh stop >/dev/null 2>&1 || true
./svc.sh uninstall >/dev/null 2>&1 || true
RUNNER_ALLOW_RUNASROOT=1 ./config.sh remove --token "$token" >/dev/null 2>&1 || true
fi
# Root, because pct needs it; the host is disposable by design.
# --no-default-labels: without it the runner also answers to `self-hosted`,
# and any other workflow with `runs-on: self-hosted` (and its secrets) could
# land on this host.
RUNNER_ALLOW_RUNASROOT=1 ./config.sh --unattended --replace --no-default-labels \
--url "https://github.com/${repo}" --token "$token" \
--name "pve-ci-${repo#*/}" --labels proxmox-test --work _work
./svc.sh install root >/dev/null
./svc.sh start >/dev/null
echo "runner for ${repo} registered"
fi
echo "test host ready: ${SERVE_ROOT} on :${SERVE_PORT}, sweeper every 3h"
+451
View File
@@ -0,0 +1,451 @@
#!/usr/bin/env bash
# Build one ct/ script from a pull request on this Proxmox host, check that the
# container really works, destroy it, and leave a result.json behind.
#
# Runs on the self-hosted runner, which *is* the disposable test host. The
# harness comes from the default branch; only the script under test comes from
# the pull request, pinned to the commit a maintainer approved.
#
# The engine hardcodes raw.githubusercontent.com/<upstream>/main for every fetch
# (ct -> build.func -> install script -> install.func -> tools.func). Pointing it
# at a fork branch therefore means rewriting those URLs: the PR tree is unpacked
# under SERVE_ROOT, every upstream main URL in it is replaced by this host's own
# HTTP server, and the containers fetch from there. Engine changes in the PR are
# tested along with the script, which is what a reviewer wants anyway.
set -uo pipefail
: "${PR_REPO:?}" "${PR_SHA:?}" "${SCRIPT:?}" "${UPSTREAM:?}" "${CTID_BASE:?}" "${OUT_DIR:?}"
OS="${OS:-}"
CTID_COUNT="${CTID_COUNT:-50}"
SERVE_ROOT="${SERVE_ROOT:-/srv/ci}"
SERVE_URL="${SERVE_URL:-http://10.99.0.1:18080}"
BRIDGE="${BRIDGE:-vmbr1}"
BUILD_TIMEOUT="${BUILD_TIMEOUT:-45m}"
URL_WAIT="${URL_WAIT:-120}"
# Every value below reaches a path, a URL or a command line.
[[ "$PR_REPO" =~ ^[A-Za-z0-9._-]+/[A-Za-z0-9._-]+$ ]] || { echo "bad PR_REPO" >&2; exit 2; }
[[ "$UPSTREAM" =~ ^[A-Za-z0-9._-]+/[A-Za-z0-9._-]+$ ]] || { echo "bad UPSTREAM" >&2; exit 2; }
[[ "$PR_SHA" =~ ^[0-9a-f]{40}$ ]] || { echo "bad PR_SHA" >&2; exit 2; }
[[ "$SCRIPT" =~ ^ct/[A-Za-z0-9._-]+\.sh$ ]] || { echo "bad SCRIPT" >&2; exit 2; }
[[ -z "$OS" || "$OS" =~ ^(debian|ubuntu|alpine)$ ]] || { echo "bad OS" >&2; exit 2; }
[[ "$CTID_BASE" =~ ^[0-9]+$ ]] || { echo "bad CTID_BASE" >&2; exit 2; }
mkdir -p "$OUT_DIR"
RAW_LOG="$OUT_DIR/build.raw.log"
LOG="$OUT_DIR/build.log"
RESULT="$OUT_DIR/result.json"
INSTALL_LOG_OUT="$OUT_DIR/install.log"
app=$(basename "$SCRIPT" .sh)
job="${GITHUB_RUN_ID:-local}-${GITHUB_RUN_ATTEMPT:-1}-${app}-${OS:-default}"
tree="$SERVE_ROOT/$job"
base="$SERVE_URL/$job"
started=$(date +%s)
status="fail"
step=""
exit_code=""
ctid=""
ct_ip=""
url=""
http_code=""
failed_units=""
strip_ansi() {
sed -E 's/\x1B\[[0-9;?]*[A-Za-z]//g; s/\x1B[()][A-Z0-9]//g; s/\r/\n/g' | grep -vE '^\s*$'
}
# The install script's own output: everything $STD hides from the build log.
# Both engines write it inside the container as /root/.install-<session>.log;
# dev_mode=keep makes sure a failed container is still there to read it from.
# If it is gone anyway, the copy the engine leaves on the host is the fallback.
collect_install_log() {
[[ -n "$ctid" ]] || return 0
# Exactly one build ran in this container, so the newest install log is ours.
if pct status "$ctid" &>/dev/null &&
pct exec "$ctid" -- sh -c 'cat "$(ls -t /root/.install-*.log 2>/dev/null | head -n1)"' 2>/dev/null |
strip_ansi >"$INSTALL_LOG_OUT" && [[ -s "$INSTALL_LOG_OUT" ]]; then
return 0
fi
# The engine names its host-side files after the session: pct_create_<ctid>_<date>_<session>.log
local sid f
sid=$(find /tmp -maxdepth 1 -name "pct_create_${ctid}_*.log" -newer "$RAW_LOG.start" -printf '%f\n' 2>/dev/null |
sed -E 's/.*_([0-9a-f]+)\.log$/\1/' | head -n1)
[[ -n "$sid" ]] || { rm -f "$INSTALL_LOG_OUT"; return 0; }
for f in /tmp/*-"${ctid}"-"${sid}".log /var/log/community-scripts/install-"${sid}"-combined-*.log; do
[[ -s "$f" ]] && { strip_ansi <"$f" >"$INSTALL_LOG_OUT"; return 0; }
done
rm -f "$INSTALL_LOG_OUT"
}
# "Requires AVX2" and friends: the script may be fine, the test host's CPU is
# not (Xeon E5-2650, Sandy Bridge: AVX, no AVX2). Shown as untestable rather
# than as a failure of the PR -- but the log is written by the code under test,
# so its word is not enough. Only the engine's error lines count, only when the
# build or install itself failed, and the feature they name must really be
# missing from this host's /proc/cpuinfo. Claiming a feature the host has
# changes nothing.
CPU_RE='(requir|need)[a-z]*[^.]{0,60}(AVX(2|-?512)?|SSE4[._]?2|x86-64-v[234])\b|\b(AVX(2|-?512)?|x86-64-v[234])\b[^.]{0,60}(not (supported|available)|does not have|is missing|unsupported)'
classify_cpu_failure() {
local line feature flags
line=$(strip_ansi <"$RAW_LOG" | grep -E '✖|\[ERROR\]' | grep -m1 -iE "$CPU_RE" |
sed -E 's/^\s*(✖️?|\[[^]]*\]|\s)*//' | cut -c1-200)
[[ -n "$line" ]] || return 0
# The CPU flag that would have to be missing for the message to be true.
case "${line,,}" in
*avx-512* | *avx512* | *x86-64-v4*) feature=avx512f ;;
*avx2* | *x86-64-v3*) feature=avx2 ;;
*sse4.2* | *sse4_2* | *x86-64-v2*) feature=sse4_2 ;;
*) return 0 ;;
esac
flags=" $(grep -m1 '^flags' /proc/cpuinfo | cut -d: -f2) "
[[ "$flags" == *" ${feature} "* ]] && return 0
status="unsupported"
step="needs a newer CPU than the test host has (no ${feature}): ${line}"
}
finish() {
[[ -f "$RAW_LOG" ]] && strip_ansi <"$RAW_LOG" >"$LOG"
[[ -f "$RAW_LOG" ]] && collect_install_log
jq -n \
--arg script "$SCRIPT" --arg os "${OS:-default}" --arg status "$status" \
--arg step "$step" --arg exit_code "$exit_code" --arg sha "$PR_SHA" \
--arg ctid "$ctid" --arg url "$url" --arg http_code "$http_code" \
--arg failed_units "$failed_units" \
--argjson duration "$(($(date +%s) - started))" \
--arg tail "$([[ -f "$LOG" ]] && tail -n 60 "$LOG")" \
'{script: $script, os: $os, status: $status, step: $step, exit_code: $exit_code,
sha: $sha, ctid: $ctid, url: $url, http_code: $http_code,
failed_units: $failed_units, duration: $duration, log_tail: $tail}' >"$RESULT"
echo "::group::result"
jq 'del(.log_tail)' "$RESULT"
echo "::endgroup::"
}
# --- Live output ------------------------------------------------------------------
# The job log on GitHub streams while the job runs, so mirroring the build there
# is all "live" takes. Two streams: the engine's own output, and the install
# log inside the container, which holds everything $STD hides (apt, npm, ...).
#
# Both are written by the code under test. A line starting with `::` would be
# read by the runner as a workflow command, so command processing is switched
# off for as long as they run, with a token the build cannot know.
stream_token=""
log_pid=""
install_pid=""
# Colours stay, cursor movement and spinner redraws go.
live_filter() {
sed -u -E 's/\r/\n/g; s/\x1B\[[0-9;?]*[ABCDEFGHJKSTfhilnsu]//g' | sed -u -E '/^\s*$/d'
}
start_streams() {
stream_token=$(head -c 16 /dev/urandom | od -An -tx1 | tr -d ' \n')
echo "::stop-commands::${stream_token}"
tail -n +1 -F --pid="$build_pid" "$RAW_LOG" 2>/dev/null > >(live_filter) &
log_pid=$!
# setsid: its own process group, so stop_streams can take down the
# pct exec / tail pair inside it in one go.
setsid bash -c '
for _ in $(seq 1 900); do
f=$(pct exec "$1" -- sh -c "ls -t /root/.install-*.log 2>/dev/null | head -n1" 2>/dev/null)
if [ -n "$f" ]; then
echo "──────── install log of CT $1 (live) ────────"
exec pct exec "$1" -- tail -n +1 -F "$f"
fi
sleep 2
done' _ "$ctid" 2>/dev/null > >(live_filter | sed -u 's/^/ │ /') &
install_pid=$!
}
stop_streams() {
[[ -n "$stream_token" ]] || return 0
[[ -n "$install_pid" ]] && kill -TERM -- "-$install_pid" &>/dev/null
[[ -n "$log_pid" ]] && kill -TERM "$log_pid" &>/dev/null
wait "$log_pid" "$install_pid" &>/dev/null
sleep 1 # let the filters flush
install_pid=""
log_pid=""
echo "::${stream_token}::"
stream_token=""
}
build_pid=""
cleanup() {
stop_streams
if [[ -n "$build_pid" ]]; then
pkill -TERM -P "$build_pid" &>/dev/null || true
kill -TERM "$build_pid" &>/dev/null || true
sleep 2
fi
if [[ -n "$ctid" ]]; then
if pct status "$ctid" &>/dev/null; then
pct stop "$ctid" --skiplock 1 &>/dev/null || true
pct destroy "$ctid" --purge 1 --force 1 &>/dev/null || true
fi
rm -f "/run/ci-jobs/$ctid"
# What the engine leaves on the host for this container and its session.
local f sid
for f in /tmp/pct_create_"${ctid}"_*.log; do
[[ -f "$f" && "$f" -nt "$RAW_LOG.start" ]] || continue
sid=$(sed -E 's/.*_([0-9a-f]+)\.log$/\1/' <<<"$f")
rm -f -- "$f" /tmp/create-lxc-"${sid}".log /tmp/*-"${ctid}"-"${sid}".log
done
fi
rm -f "$RAW_LOG.start"
rm -rf -- "$tree"
}
trap cleanup EXIT
# A cancelled job sends TERM; without this the EXIT trap would not run.
trap 'exit 143' TERM
trap 'exit 130' INT
# fail <reason> [fail|unsupported]: only the build/install branches pass the
# second argument, after classify_cpu_failure has checked the host.
#
# The exit code says whether *this job* worked, not whether the app did. A
# script that fails to build is a result: it is written down, reported on the
# PR (check, comment, review) and the job exits 0 -- a red run per broken app
# would only be notification noise. Only trouble with the test setup itself
# (setup_error) fails the job, because that is ours to look at.
fail() {
step="$1"
status="${2:-fail}"
[[ "$status" == "unsupported" ]] || status="fail"
finish
echo "::warning title=${SCRIPT}::${step}"
exit 0
}
setup_error() {
step="$1"
status="error"
finish
echo "::error title=${SCRIPT}::test setup: ${step}"
exit 1
}
# --- Reserve a container ID ---------------------------------------------------
# A lock file per ID, not just "first free one", because the engine only creates
# the container minutes later and two jobs would otherwise pick the same ID.
mkdir -p /run/ci-jobs
exec 9>/run/ci-jobs/.lock
flock 9
for ((id = CTID_BASE; id < CTID_BASE + CTID_COUNT; id++)); do
[[ -e "/run/ci-jobs/$id" ]] && continue
pvesh get /cluster/nextid --vmid "$id" &>/dev/null || continue
echo "$job" >"/run/ci-jobs/$id"
ctid=$id
break
done
flock -u 9
exec 9>&-
[[ -n "$ctid" ]] || setup_error "no free container ID in ${CTID_BASE}-$((CTID_BASE + CTID_COUNT - 1))"
# --- Fetch the approved commit and point it at ourselves ------------------------
mkdir -p "$tree"
curl -fsSL --retry 3 "https://codeload.github.com/${PR_REPO}/tar.gz/${PR_SHA}" |
tar -xz -C "$tree" --strip-components=1 || setup_error "could not download ${PR_REPO}@${PR_SHA:0:7}"
[[ -f "$tree/$SCRIPT" ]] || fail "${SCRIPT} does not exist at ${PR_SHA:0:7}"
upstream_re="https://raw\.githubusercontent\.com/${UPSTREAM//./\\.}/(refs/heads/)?main"
grep -rlE "$upstream_re" "$tree" --include='*.sh' --include='*.func' |
xargs -r sed -i -E "s#${upstream_re}#${base}#g"
# --- Build ----------------------------------------------------------------------
echo "Building ${SCRIPT} (${OS:-default OS}) as CT ${ctid} from ${PR_REPO}@${PR_SHA:0:7}"
hostname_safe="ci-$(tr -c 'a-z0-9-\n' '-' <<<"${app,,}" | cut -c1-50)"
build_env=(
PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
HOME=/root TERM=dumb LANG=C.UTF-8
mode=default MODE=default PHS_SILENT=1 DIAGNOSTICS=no
var_ctid="$ctid" var_hostname="$hostname_safe"
var_container_storage=local var_template_storage=local var_brg="$BRIDGE"
"dev_mode=timing,keep" COMMUNITY_SCRIPTS_URL="$base"
)
touch "$RAW_LOG.start"
[[ -n "$OS" ]] && build_env+=(var_os="$OS")
# --- Answers for the script's questions ------------------------------------------
# Nobody types into CI, so anything the install asks has to be answered up front.
# AGENTS.md rule 24: a script reads var_<x> first and prompts only when it is
# unset (or uses prompt_input_required ... "var_x"). Every such variable gets a
# CI value here. The names come from json/<app>.json app_vars (DevScripts) and
# from the read/prompt_ lines of the install script (both repos). A bare `read`
# that ignores its variable still fails, and the report says why.
ENGINE_VARS=" var_ctid var_hostname var_os var_version var_cpu var_ram var_disk var_brg var_net var_container_storage var_template_storage var_unprivileged var_tags var_arm64 var_testurl var_verbose var_timezone var_http_proxy var_http_no_proxy var_appversion "
host_app=$(tr -c 'a-z0-9-\n' '-' <<<"${app,,}")
ci_value() {
case "${1,,}" in
*pass* | *secret* | *token* | *key*) head -c 32 /dev/urandom | base64 | tr -dc 'A-Za-z0-9' | cut -c1-20 ;;
*email* | *mail*) echo "ci@example.test" ;;
*url* | *endpoint*) echo "https://ci-${host_app}.example.test" ;;
# One name per variable: var_opencloud_collabora_fqdn ->
# ci-opencloud-collabora.example.test, so several of them cannot collide.
*host* | *domain* | *fqdn*)
local part="${1#var_}"
part=$(sed -E 's/_?(fqdn|hostname|host|domain)$//' <<<"${part,,}" | tr '_' '-')
echo "ci-${part:-$host_app}.example.test"
;;
*user*) echo "ciadmin" ;;
*) echo "ci-test" ;;
esac
}
declare -A answers=()
json="$tree/json/${app}.json"
if [[ -f "$json" ]]; then
while IFS=$'\t' read -r name default; do
answers[$name]="$default"
done < <(jq -r '(if type == "array" then .[0] else . end).app_vars // [] | .[] |
select(.name | test("^var_[a-z0-9_]{1,40}$")) | [.name, (.default // "" | tostring)] | @tsv' "$json" 2>/dev/null)
fi
# Names built at run time from a loop, e.g. OpenCloud:
# servers=("opencloud" "collabora" "wopi")
# for server in "${servers[@]}"; do fqdn_var="var_opencloud_${server}_fqdn"
# Each var_...${x}... template is expanded with the words of `for x in ...`
# (or of the array that loop walks). Nothing is executed, only read.
expand_templates() {
local tpl x pre suf words arr w
while read -r tpl; do
x=$(sed -E 's/^[^$]*\$\{?([A-Za-z_][A-Za-z0-9_]*)\}?.*$/\1/' <<<"$tpl")
pre="${tpl%%\$*}"
suf=$(sed -E 's/^[^$]*\$\{?[A-Za-z_][A-Za-z0-9_]*\}?//' <<<"$tpl")
words=$(grep -m1 -oE "for ${x} in [^;]*" "$1" | sed -E "s/^for ${x} in //")
if [[ "$words" =~ \$\{([A-Za-z_][A-Za-z0-9_]*)\[@\]\} ]]; then
arr="${BASH_REMATCH[1]}"
words=$(grep -m1 -oE "^\s*${arr}=\([^)]*\)" "$1" | sed -E "s/^\s*${arr}=\(//; s/\)$//")
fi
for w in $words; do
w="${w//[\"\']/}"
[[ "${pre}${w}${suf}" =~ ^var_[a-z0-9_]{1,40}$ ]] && echo "${pre}${w}${suf}"
done
done < <(grep -oE 'var_[a-z0-9_]*\$\{?[A-Za-z_][A-Za-z0-9_]*\}?[a-z0-9_]*' "$1" | sort -u)
}
# The engine names the install script after APP (lower case, no spaces), which
# need not match the ct/ file name: ct/bluesky-pds.sh -> blueskypds-install.sh.
nsapp=$(grep -m1 -oE '^APP="[^"]+"' "$tree/$SCRIPT" | cut -d'"' -f2 | tr -d ' ' | tr '[:upper:]' '[:lower:]')
for f in "$tree/install/${app}-install.sh" "$tree/install/${nsapp}-install.sh" "$tree/$SCRIPT"; do
[[ -f "$f" ]] || continue
while read -r name; do
[[ -v "answers[$name]" ]] || answers[$name]=""
done < <({
grep -E '\bread\b|prompt_(input|input_required|confirm|select|password)\b' "$f" | grep -oE '\bvar_[a-z0-9_]{1,40}\b'
expand_templates "$f"
} | sort -u)
done
for name in "${!answers[@]}"; do
[[ "$ENGINE_VARS" == *" $name "* ]] && continue
value="${answers[$name]}"
[[ -n "$value" ]] || value=$(ci_value "$name")
build_env+=("$name=$value")
case "${name,,}" in
*pass* | *secret* | *token* | *key*) echo "Pre-answered ${name}=(random)" ;;
*) echo "Pre-answered ${name}=${value}" ;;
esac
done
# In the background so that a TERM from a cancelled job is handled at once:
# bash defers traps until a foreground command returns.
env -i "${build_env[@]}" timeout --kill-after=1m "$BUILD_TIMEOUT" \
bash -c 'bash -c "$(curl -fsSL "$1")"' _ "$base/$SCRIPT" </dev/null >"$RAW_LOG" 2>&1 &
build_pid=$!
start_streams
wait "$build_pid"
exit_code=$?
build_pid=""
stop_streams
if ((exit_code == 124 || exit_code == 137)); then
fail "build timed out after ${BUILD_TIMEOUT}"
fi
# The engine marks the step that broke with a cross. Its closing "Installation
# failed in container" line is generic, so the cross before it is the useful one.
failure_line() {
local line
line=$(strip_ansi <"$RAW_LOG" | grep -E '✖|\[ERROR\]|error:' | grep -v 'Installation failed in container' |
tail -n 1 | sed -E 's/^\s*(✖️?|\s)*//' | cut -c1-200)
# A prompt nobody can answer: say what to do instead.
# First, because the report keeps 200 characters.
if [[ "$line" == *"read exited with status"* || "$line" == *"executing command read "* ]]; then
local asked
asked=$(grep -oE '\bvar_[a-z0-9_]{1,40}\b' <<<"$line" | tail -n 1)
line="unanswerable prompt (AGENTS.md rule 24): ask only when ${asked:-var_<x>} is unset, or use prompt_input_required. ${line}"
fi
echo "$line"
}
if ((exit_code != 0)); then
last=$(failure_line)
step="build exited ${exit_code}${last:+: $last}"
status="fail"
classify_cpu_failure
fail "$step" "$status"
fi
# With dev_mode=keep (needed to read the install log) the engine carries on
# after a failed install and still prints "Completed Successfully", exit 0.
install_exit=$(strip_ansi <"$RAW_LOG" | grep -oE 'Installation failed in container [0-9]+ \(exit code: [0-9]+\)' |
grep -oE '[0-9]+\)$' | tr -d ')' | tail -n 1)
if [[ -n "$install_exit" ]]; then
last=$(failure_line)
step="install script exited ${install_exit}${last:+: $last}"
status="fail"
classify_cpu_failure
fail "$step" "$status"
fi
grep -qi 'completed successfully' "$RAW_LOG" || fail "build exited 0 but never reported success"
# --- Checks ---------------------------------------------------------------------
pct status "$ctid" 2>/dev/null | grep -q running || fail "container ${ctid} is not running"
ct_ip=$(pct exec "$ctid" -- sh -c "ip -4 -o addr show dev eth0 | awk '{print \$4}' | cut -d/ -f1" 2>/dev/null | head -n1)
if pct exec "$ctid" -- sh -c 'command -v systemctl' &>/dev/null; then
sleep 10
failed_units=$(pct exec "$ctid" -- systemctl --failed --no-legend --plain 2>/dev/null | awk '{print $1}' | paste -sd, -)
elif pct exec "$ctid" -- sh -c 'command -v rc-status' &>/dev/null; then
failed_units=$(pct exec "$ctid" -- rc-status --crashed 2>/dev/null | awk '{print $1}' | paste -sd, -)
fi
[[ -z "$failed_units" ]] || fail "failed services: ${failed_units}"
# What the app's page looks like, for the PR comment. Chromium runs as its own
# unprivileged user with its sandbox on: the page is served by code under test.
# ci-shot-firewall (pve-ci-host-setup.sh) keeps that user off this host and the
# lab network; without it there is no screenshot rather than an unguarded one.
take_screenshot() {
command -v chromium >/dev/null && id ci-shot &>/dev/null || return 0
iptables -w -S CI_SHOT 2>/dev/null | grep -q -- '-d 10.0.0.0/8 .*REJECT' || return 0
local dir
dir=$(mktemp -d)
chown ci-shot "$dir"
runuser -u ci-shot -- timeout 60 chromium --headless=new --disable-gpu --hide-scrollbars \
--ignore-certificate-errors --no-first-run --window-size=1280,800 --virtual-time-budget=10000 \
--user-data-dir="$dir/profile" --screenshot="$dir/shot.png" "$url" &>/dev/null
# Only a plausible PNG leaves this function.
if [[ -s "$dir/shot.png" && $(stat -c %s "$dir/shot.png") -lt 3000000 ]] &&
[[ "$(head -c 8 "$dir/shot.png" | od -An -tx1 | tr -d ' \n')" == "89504e470d0a1a0a" ]]; then
cp "$dir/shot.png" "$OUT_DIR/screenshot.png"
fi
rm -rf -- "$dir"
}
# The URL the script tells the user to open, e.g. http://${IP}:1337/admin, has
# to answer 200 (after redirects, so a jump to /login is fine). A script that
# prints no URL has no web UI to check.
pattern=$(grep -oE 'https?://\$\{IP\}(:[0-9]+)?(/[A-Za-z0-9._~/?=&%-]*)?' "$tree/$SCRIPT" | head -n1)
if [[ -n "$pattern" && -n "$ct_ip" ]]; then
url="${pattern/\$\{IP\}/$ct_ip}"
for ((waited = 0; waited < URL_WAIT; waited += 5)); do
http_code=$(curl -sk -L --max-redirs 5 -o /dev/null -w '%{http_code}' --max-time 10 "$url" || true)
[[ "$http_code" =~ ^2[0-9][0-9]$ ]] && break
sleep 5
done
# Also when the check fails: an error page explains more than a status code.
[[ "$http_code" =~ ^[1-5][0-9][0-9]$ ]] && take_screenshot
[[ "$http_code" =~ ^2[0-9][0-9]$ ]] || fail "${url} answered ${http_code/#000/nothing} instead of 200 within ${URL_WAIT}s"
fi
status="pass"
finish
+7 -1
View File
@@ -594,7 +594,13 @@ Exercise vigilance regarding copycat or coat-tailing sites that seek to exploit
### 💾 Core
- fix: flush the image before importing a VM disk [@michelroegl-brunner](https://github.com/michelroegl-brunner) ([core#223](https://github.com/community-scripts/core/pull/223))
- fix: keep a script's config file when a package installs it [@MickLesk](https://github.com/MickLesk) ([core#226](https://github.com/community-scripts/core/pull/226))
- fix: flush the image before importing a VM disk [@michelroegl-brunner](https://github.com/michelroegl-brunner) ([core#223](https://github.com/community-scripts/core/pull/223))
### 📂 Github
- pr-script-test: pre-answer prompt variables built in a loop [@michelroegl-brunner](https://github.com/michelroegl-brunner) ([#18596](https://github.com/community-scripts/ProxmoxVE/pull/18596))
- Build PR scripts on a Proxmox test host via /test [@michelroegl-brunner](https://github.com/michelroegl-brunner) ([#18594](https://github.com/community-scripts/ProxmoxVE/pull/18594))
## 2026-11-09