Bump workflow actions to their current majors

The older majors ran on Node 20, which GitHub now forces onto Node 24
with a deprecation warning on every run.

.github/workflows/arm-script-test.yml

@@ -39,7 +39,7 @@ jobs:
     timeout-minutes: 120
     steps:
       # The harness from the default branch, as in pr-script-test.yml.
-      - uses: actions/checkout@v5
+      - uses: actions/checkout@v7
         with:
           sparse-checkout: .github/workflows/scripts
           persist-credentials: false
@@ -106,7 +106,7 @@ jobs:
           } >>"$GITHUB_STEP_SUMMARY"
 
       - if: always()
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v7
         with:
           name: result-arm
           path: ${{ runner.temp }}/result

.github/workflows/pocketbase-bot.yml

@@ -2,7 +2,7 @@ name: PocketBase Bot
 
 on:
   issue_comment:
-    types: [created]
+    types: [created, edited]
 
 permissions:
   issues: write
@@ -25,6 +25,9 @@ jobs:
           POCKETBASE_ADMIN_PASSWORD: ${{ secrets.POCKETBASE_ADMIN_PASSWORD }}
           COMMENT_BODY: ${{ github.event.comment.body }}
           COMMENT_ID: ${{ github.event.comment.id }}
+          EVENT_ACTION: ${{ github.event.action }}
+          SENDER: ${{ github.event.sender.login }}
+          PREVIOUS_BODY: ${{ github.event.changes.body.from }}
           ISSUE_NUMBER: ${{ github.event.issue.number }}
           IS_PR_COMMENT: ${{ github.event.issue.pull_request && 'true' || 'false' }}
           REPO_OWNER: ${{ github.repository_owner }}
@@ -87,7 +90,7 @@ jobs:
             const issueNumber = parseInt(process.env.ISSUE_NUMBER, 10);
             const isPrComment = process.env.IS_PR_COMMENT === 'true';
             const commentId = parseInt(process.env.COMMENT_ID, 10);
-            const actor = process.env.ACTOR;
+            let actor = process.env.ACTOR;
 
             function ghRequest(path, method, body) {
               const headers = {
@@ -278,7 +281,65 @@ jobs:
             // Accept /pocketbase at the start of ANY line (leading whitespace ok),
             // so the command works even when preceded by other text. Mid-sentence
             // mentions and blockquoted ("> ...") examples are ignored.
-            const commentBody = process.env.COMMENT_BODY || '';
+            let commentBody = process.env.COMMENT_BODY || '';
+
+            // ── Metadata requests from the website ─────────────────────────────
+            // One comment per requested change, each with Accept/Decline boxes.
+            // Ticking one edits the comment; that edit is the only event handled.
+            const REQUEST_MARKER = '<!-- pb-request -->';
+            const DONE_MARKER = '<!-- pb-request:done -->';
+            const isRequest = commentBody.includes(REQUEST_MARKER);
+            let decidedBy = null;
+            if (process.env.EVENT_ACTION === 'edited' && !isRequest) process.exit(0);
+            if (isRequest) {
+              if (process.env.EVENT_ACTION !== 'edited' || commentBody.includes(DONE_MARKER)) process.exit(0);
+              const ticked = (body, word) =>
+                new RegExp('^[ \\t]*[-*] \\[[xX]\\] ' + word + '[ \\t]*\\r?$', 'm').test(body);
+              const previous = process.env.PREVIOUS_BODY || '';
+              const accept = ticked(commentBody, 'Accept') && !ticked(previous, 'Accept');
+              const decline = ticked(commentBody, 'Decline') && !ticked(previous, 'Decline');
+              if (accept === decline) process.exit(0);
+
+              const sender = process.env.SENDER;
+              const permRes = await ghRequest(
+                '/repos/' + owner + '/' + repo + '/collaborators/' + encodeURIComponent(sender) + '/permission'
+              );
+              const level = permRes.ok ? JSON.parse(permRes.body).permission : 'none';
+              if (level !== 'admin' && level !== 'write') {
+                await postComment(
+                  '❌ **PocketBase Bot**: @' + sender + ' is not authorized to decide metadata requests.\n' +
+                  'Only maintainers with write access can accept or decline them.'
+                );
+                process.exit(0);
+              }
+
+              const heading = (commentBody.match(/^### (.+)$/m) || [])[1] || 'this change';
+              const command = commentBody.match(/<!-- pb-command -->\r?\n([\s\S]*?)\r?\n<!-- \/pb-command -->/);
+              const decided = commentBody
+                .replace(/^[ \t]*[-*] \[[ xX]\] (Accept|Decline)[ \t]*\r?\n?/gm, '')
+                .replace(
+                  REQUEST_MARKER,
+                  '**' + (accept ? '✅ Accepted' : '❌ Declined') + '** by @' + sender + '\n\n' +
+                  REQUEST_MARKER + '\n' + DONE_MARKER
+                );
+              const markRes = await ghRequest(
+                '/repos/' + owner + '/' + repo + '/issues/comments/' + commentId, 'PATCH', { body: decided }
+              );
+              if (!markRes.ok) console.warn('Could not mark the request as decided:', markRes.body);
+
+              if (decline) {
+                await postComment('❌ **PocketBase Bot**: **' + heading + '** declined by @' + sender + '.');
+                process.exit(0);
+              }
+              if (!command) {
+                await postComment('❌ **PocketBase Bot**: no command found in the request for **' + heading + '**.');
+                process.exit(0);
+              }
+              commentBody = command[1];
+              decidedBy = sender;
+              actor = sender;
+            }
+
             const cmdLine = commentBody
               .split('\n')
               .map(l => l.trim())
@@ -292,7 +353,7 @@ jobs:
 
             // ── Permission check ───────────────────────────────────────────────
             const association = process.env.ACTOR_ASSOCIATION;
-            if (association !== 'OWNER' && association !== 'MEMBER') {
+            if (!decidedBy && association !== 'OWNER' && association !== 'MEMBER') {
               await addReaction('-1');
               await postComment(
                 '❌ **PocketBase Bot**: @' + actor + ' is not authorized to use this command.\n' +
@@ -957,6 +1018,35 @@ jobs:
                   );
                   process.exit(0);
                 }
+                if (fieldName === 'app_vars') {
+                  const TYPES = ['text', 'password', 'number', 'boolean', 'select'];
+                  const problems = [];
+                  if (!Array.isArray(parsedJson)) problems.push('must be a JSON array');
+                  else parsedJson.forEach(function (v, i) {
+                    const at = 'entry ' + (i + 1) + ': ';
+                    if (!v || typeof v !== 'object' || Array.isArray(v)) { problems.push(at + 'must be an object'); return; }
+                    if (typeof v.name !== 'string' || !/^var_[a-z0-9_]+$/.test(v.name)) problems.push(at + '`name` must look like `var_example`');
+                    if (typeof v.label !== 'string' || !v.label.trim()) problems.push(at + '`label` is required');
+                    if (TYPES.indexOf(v.type) === -1) problems.push(at + '`type` must be one of ' + TYPES.join(', '));
+                    if (v.default !== undefined && typeof v.default !== 'string') problems.push(at + '`default` must be a string');
+                    if (v.type === 'select') {
+                      const opts = v.options;
+                      if (!Array.isArray(opts) || opts.length === 0 || opts.some(function (o) { return typeof o !== 'string' || !o; })) {
+                        problems.push(at + 'a select needs `options`, a non-empty array of strings');
+                      } else if (v.default !== undefined && opts.indexOf(v.default) === -1) {
+                        problems.push(at + '`default` must be one of its options');
+                      }
+                    }
+                    ['required', 'secret'].forEach(function (k) {
+                      if (v[k] !== undefined && typeof v[k] !== 'boolean') problems.push(at + '`' + k + '` must be true or false');
+                    });
+                  });
+                  if (problems.length > 0) {
+                    await addReaction('-1');
+                    await postComment('❌ **PocketBase Bot**: `app_vars` was not saved.\n\n- ' + problems.join('\n- '));
+                    process.exit(0);
+                  }
+                }
                 setPayload[fieldName] = parsedJson;
               } else {
                 setPayload[fieldName] = codeBlockValue;
@@ -1236,3 +1326,58 @@ jobs:
           });
           ENDSCRIPT
         shell: bash
+
+      - name: Close a metadata request once every change is decided
+        if: always() && github.event.action == 'edited' && contains(github.event.comment.body, '<!-- pb-request -->')
+        env:
+          ISSUE_NUMBER: ${{ github.event.issue.number }}
+          REPO: ${{ github.repository }}
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        run: |
+          node << 'ENDSCRIPT'
+          (async function () {
+            const api = 'https://api.github.com/repos/' + process.env.REPO + '/issues/' + process.env.ISSUE_NUMBER;
+            const headers = {
+              'Authorization': 'Bearer ' + process.env.GITHUB_TOKEN,
+              'Accept': 'application/vnd.github+json',
+              'Content-Type': 'application/json',
+              'User-Agent': 'pocketbase-bot'
+            };
+            const https = require('https');
+            function call(method, url, body) {
+              return new Promise(function (resolve, reject) {
+                const req = https.request(url, { method: method, headers: headers }, function (res) {
+                  let data = '';
+                  res.on('data', function (d) { data += d; });
+                  res.on('end', function () { resolve({ ok: res.statusCode < 300, status: res.statusCode, body: data }); });
+                });
+                req.on('error', reject);
+                if (body) req.write(JSON.stringify(body));
+                req.end();
+              });
+            }
+            const requests = [];
+            for (let page = 1; page <= 10; page++) {
+              const res = await call('GET', api + '/comments?per_page=100&page=' + page);
+              if (!res.ok) throw new Error('Could not list comments: ' + res.status);
+              const comments = JSON.parse(res.body);
+              for (const c of comments) {
+                if ((c.body || '').includes('<!-- pb-request -->')) requests.push(c.body);
+              }
+              if (comments.length < 100) break;
+            }
+            const open = requests.filter(function (b) { return !b.includes('<!-- pb-request:done -->'); });
+            if (requests.length === 0 || open.length > 0) {
+              console.log(open.length + ' of ' + requests.length + ' changes still undecided.');
+              return;
+            }
+            const accepted = requests.filter(function (b) { return b.includes('✅ Accepted'); }).length;
+            const res = await call('PATCH', api, { state: 'closed', state_reason: accepted > 0 ? 'completed' : 'not_planned' });
+            if (!res.ok) throw new Error('Could not close the issue: ' + res.status);
+            console.log('Closed: ' + accepted + ' accepted, ' + (requests.length - accepted) + ' declined.');
+          })().catch(function (e) {
+            console.error(e.message || e);
+            process.exit(1);
+          });
+          ENDSCRIPT
+        shell: bash

.github/workflows/pr-script-test.yml

@@ -95,9 +95,10 @@ jobs:
       trusted: ${{ steps.plan.outputs.trusted }}
       kind: ${{ steps.plan.outputs.kind }}
       platforms: ${{ steps.plan.outputs.platforms }}
+      base_sha: ${{ steps.plan.outputs.base_sha }}
     steps:
       - id: plan
-        uses: actions/github-script@v8
+        uses: actions/github-script@v9
         with:
           script: |
             const { owner, repo } = context.repo;
@@ -258,6 +259,31 @@ jobs:
             }
 
             const include = [];
+            // Upgrade test: install the ct/ script from the PR's base branch,
+            // then update with the PR, as a user who installed before the merge
+            // would. Only in ProxmoxVE, where users have installed it, and only
+            // when the PR changes update_script(): with the same update code an
+            // existing install updates exactly as it does today. New scripts
+            // have nothing to upgrade from. Proxmox only, once per script.
+            const UPGRADE_REPO = 'community-scripts/ProxmoxVE';
+            let baseSha = '';
+            if (!isIssue && `${owner}/${repo}` === UPGRADE_REPO) {
+              const { data: b } = await github.rest.repos.getBranch({ owner, repo, branch: pr.base.ref });
+              baseSha = b.commit.sha;
+            }
+            // From `function update_script() {` to the first `}` at the start of
+            // a line; null without one.
+            const updateCode = t => ((t || '').match(/^(function[ \t]+)?update_script[ \t]*\(\)[ \t]*\{[\s\S]*?^\}/m) || [null])[0];
+            const updateChanged = async (p, headText) => {
+              if (!baseSha) return false;
+              let baseText;
+              try {
+                const res = await github.rest.repos.getContent({ owner, repo, path: p, ref: baseSha });
+                baseText = Buffer.from(res.data.content || '', 'base64').toString('utf8');
+              } catch (e) { return false; }
+              const before = updateCode(baseText), after = updateCode(headText);
+              return before !== null && after !== null && before !== after;
+            };
             const fetchScript = async p => {
               try {
                 const res = await github.rest.repos.getContent({ owner: head.owner.login, repo: head.name, path: p, ref: headSha });
@@ -302,7 +328,11 @@ jobs:
                 platforms = platforms.filter(p => only.has(p));
               }
               for (const platform of platforms) {
-                for (const os of oses) include.push({ script, os, platform, arm64: platform === 'arm' ? arm : '', name: `${os ? `${name}-${os}` : name}-${platform}` });
+                for (const os of oses) include.push({ script, os, platform, arm64: platform === 'arm' ? arm : '', mode: '', name: `${os ? `${name}-${os}` : name}-${platform}` });
+              }
+              if (!isVm && platforms.includes('pve') && await updateChanged(script, text)) {
+                const os = oses[0];
+                include.push({ script, os, platform: 'pve', arm64: '', mode: 'upgrade', name: `${os ? `${name}-${os}` : name}-pve-upgrade` });
               }
             }
             for (const script of entries.slice(MAX)) notBuilt.push(`\`${script}\`: at most ${MAX} scripts per run, name it with \`/test ${script}\``);
@@ -327,7 +357,7 @@ jobs:
                 // full test, e.g. as a required check.
                 owner, repo, name: `Proxmox script test${onlyList.length ? ` (${onlyList.join(', ')} only)` : ''}`, head_sha: headSha, status: 'in_progress',
                 details_url: runUrl,
-                output: { title: `Building ${include.length}`, summary: include.map(i => `- \`${i.script}\` ${[i.platform, i.os].filter(Boolean).join(' · ')}`).join('\n') },
+                output: { title: `Building ${include.length}`, summary: include.map(i => `- \`${i.script}\` ${[i.platform, i.os, i.mode].filter(Boolean).join(' · ')}`).join('\n') },
               });
               checkId = String(check.id);
             }
@@ -339,13 +369,14 @@ jobs:
             core.setOutput('check_id', checkId);
             core.setOutput('not_built', JSON.stringify(notBuilt));
             core.setOutput('platforms', onlyList.join(' '));
+            core.setOutput('base_sha', baseSha);
             // 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 ${headSha.slice(0, 7)}${isIssue ? ' (default branch)' : ''}${onlyList.length ? `, only on ${onlyList.join(', ')}` : ''} — [watch it live](${runUrl}) (open the \`test\` job)`, '',
-              ...include.map(i => `- \`${i.script}\` (${[i.platform, i.os].filter(Boolean).join(' · ')})`),
+              ...include.map(i => `- \`${i.script}\` (${[i.platform, i.os, i.mode && 'upgrade from the base branch'].filter(Boolean).join(' · ')})`),
               ...(notBuilt.length ? ['', '**Not built:**', ...notBuilt.map(n => `- ${n}`)] : []),
             ].join('\n'));
 
@@ -363,7 +394,7 @@ jobs:
     continue-on-error: ${{ matrix.platform == 'arm' }}
     permissions:
       contents: read
-    timeout-minutes: 90 # install (45) + update (30) + checks
+    timeout-minutes: 120 # install (45) + update (30) + second update + checks
     strategy:
       fail-fast: false
       matrix:
@@ -371,7 +402,7 @@ jobs:
     name: test (${{ matrix.name }})
     steps:
       # The default branch, never the PR: the harness itself must be trusted.
-      - uses: actions/checkout@v5
+      - uses: actions/checkout@v7
         with:
           sparse-checkout: .github/workflows/scripts
           persist-credentials: false
@@ -392,6 +423,9 @@ jobs:
           # ARM is the Incus harness on another machine.
           PLATFORM: ${{ matrix.platform == 'arm' && 'incus' || matrix.platform }}
           VAR_ARM64: ${{ matrix.platform == 'arm' && 'yes' || '' }}
+          # upgrade: install from the base branch, update with the PR.
+          TEST_MODE: ${{ matrix.mode || '' }}
+          BASE_SHA: ${{ needs.gate.outputs.base_sha }}
           UPSTREAM: ${{ github.repository }}
           CTID_BASE: ${{ github.repository == 'community-scripts/ProxmoxVE' && '9050' || '9000' }}
           # Only so release lookups are not rate limited: unauthenticated, all
@@ -402,14 +436,14 @@ jobs:
         # The self-hosted runners run as root; the GitHub one needs sudo (pct
         # and incus do).
         run: >-
-          ${{ matrix.platform == 'arm' && 'sudo --preserve-env=PR_REPO,PR_SHA,SCRIPT,OS,PLATFORM,VAR_ARM64,UPSTREAM,CTID_BASE,GH_API_TOKEN,OUT_DIR,GITHUB_RUN_ID,GITHUB_RUN_ATTEMPT' || '' }}
+          ${{ matrix.platform == 'arm' && 'sudo --preserve-env=PR_REPO,PR_SHA,SCRIPT,OS,PLATFORM,VAR_ARM64,TEST_MODE,BASE_SHA,UPSTREAM,CTID_BASE,GH_API_TOKEN,OUT_DIR,GITHUB_RUN_ID,GITHUB_RUN_ATTEMPT' || '' }}
           bash .github/workflows/scripts/pve-run-script.sh
 
       - if: always() && matrix.platform == 'arm'
         run: sudo chmod -R a+rX "$RUNNER_TEMP/result" 2>/dev/null || true
 
       - if: always()
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@v7
         with:
           name: result-${{ matrix.name }}
           path: ${{ runner.temp }}/result
@@ -427,12 +461,12 @@ jobs:
       issues: write
       pull-requests: write
     steps:
-      - uses: actions/download-artifact@v5
+      - uses: actions/download-artifact@v8
         with:
           pattern: result-*
           path: results
 
-      - uses: actions/github-script@v8
+      - uses: actions/github-script@v9
         env:
           MATRIX: ${{ needs.gate.outputs.matrix }}
           PR: ${{ needs.gate.outputs.pr }}
@@ -477,6 +511,12 @@ jobs:
                     if (r.status === 'pass' && r.update_status === 'error') { r.status = 'error'; r.step = `update ${r.update_step}`; }
                     const ulog = path.join(dir, 'update.log');
                     r.update_log = fs.existsSync(ulog) ? fs.readFileSync(ulog, 'utf8') : '';
+                    const u2log = path.join(dir, 'update2.log');
+                    r.update2_log = fs.existsSync(u2log) ? fs.readFileSync(u2log, 'utf8') : '';
+                    if (!['upgrade', ''].includes(r.mode || '')) r.mode = '';
+                    r.mode = r.mode || '';
+                    r.notes = Array.isArray(r.notes) ? r.notes.filter(n => typeof n === 'string').slice(0, 8) : [];
+                    r.resources = typeof r.resources === 'string' ? r.resources : '';
                     const ushot = path.join(dir, 'screenshot-update.png');
                     r.shot_update_file = fs.existsSync(ushot) ? ushot : '';
                     // ARM runs the Incus harness on an arm64 machine; result.json
@@ -484,7 +524,7 @@ jobs:
                     // unknown one would never match a matrix row anyway.
                     if (r.arch === 'aarch64') r.platform = 'arm';
                     else if (!['pve', 'incus'].includes(r.platform)) r.platform = 'pve';
-                    found.set(`${r.script}|${r.platform}|${r.os}`, r);
+                    found.set(`${r.script}|${r.platform}|${r.os}|${r.mode}`, r);
                   } catch (err) { core.warning(`unreadable ${p}: ${err.message}`); }
                 }
               }
@@ -492,19 +532,20 @@ jobs:
             walk('results');
 
             // A job that died before writing result.json still gets a row.
-            const rows = matrix.map(m => ({ ...(found.get(`${m.script}|${m.platform || 'pve'}|${m.os || 'default'}`) || {
-              script: m.script, os: m.os || 'default', platform: m.platform || 'pve',
+            const rows = matrix.map(m => ({ ...(found.get(`${m.script}|${m.platform || 'pve'}|${m.os || 'default'}|${m.mode || ''}`) || {
+              script: m.script, os: m.os || 'default', platform: m.platform || 'pve', mode: m.mode || '', notes: [], resources: '', update2_log: '',
               status: process.env.TEST_RESULT === 'cancelled' ? 'cancelled' : 'error',
               step: 'no result (runner or job failure)', duration: 0, log_tail: '', install_log: '', shot_file: '',
               update_status: '', update_log: '', shot_update_file: '',
             }), name: m.name }));
-            // Where a row ran: "pve", "incus · alpine", "arm64", or "VM" (Proxmox only).
+            // Where a row ran: "pve", "incus · alpine", "arm64", "pve · upgrade", or "VM" (Proxmox only).
             const isVm = r => r.script.startsWith('vm/');
             const isArm = r => r.platform === 'arm';
             const where = r => {
               if (isVm(r)) return 'VM';
               const p = isArm(r) ? 'arm64' : r.platform;
-              return r.os && r.os !== 'default' ? `${p} · ${r.os}` : p;
+              const w = r.os && r.os !== 'default' ? `${p} · ${r.os}` : p;
+              return r.mode === 'upgrade' ? `${w} · upgrade` : w;
             };
             // ARM is information only: shown, never counted. The result, the
             // check, the review and the approval come from Proxmox and Incus.
@@ -647,6 +688,9 @@ jobs:
                 const lines = r.update_log.trimEnd().split('\n').length;
                 parts.push(details(`🔄 ${r.script} (${where(r)}) — update log (${lines} lines)`, r.update_log, updateShare));
               }
+              if (r.update2_log) {
+                parts.push(details(`🔁 ${r.script} (${where(r)}) — second update log`, r.update2_log, Math.min(r.update2_log.length, Math.floor(budget / 8))));
+              }
               return parts.join('\n\n');
             });
 
@@ -661,6 +705,22 @@ jobs:
               `${headline} at \`${process.env.SHA.slice(0, 7)}\``, '',
               table, '',
               ...(armLine ? [`<sub>${armLine}</sub>`, ''] : []),
+              ...(rows.some(r => r.mode === 'upgrade')
+                ? ['<sub>`upgrade`: installed from the base branch as it is now, then updated with this PR, as a user who installed before the merge would.</sub>', '']
+                : []),
+              ...(() => {
+                const noted = rows.filter(r => r.notes.length);
+                return noted.length
+                  ? ['**Notes** (information only, not counted):', ...noted.flatMap(r => r.notes.map(n => `- \`${r.script}\` (${where(r)}): ${clean(n)}`)), '']
+                  : [];
+              })(),
+              ...(() => {
+                const measured = rows.filter(r => r.resources);
+                return measured.length
+                  ? ['<details><summary>📊 Resources</summary>', '', '| Script | Platform | Usage |', '|---|---|---|',
+                     ...measured.map(r => `| \`${r.script}\` | ${where(r)} | ${clean(r.resources)} |`), '', '</details>', '']
+                  : [];
+              })(),
               ...(partial ? [`<sub>Built only on ${process.env.PLATFORMS.split(' ').join(', ')}; comment \`/test\` for every platform.</sub>`, ''] : []),
               ...shots, '',
               ...(() => { try { const n = JSON.parse(process.env.NOT_BUILT || '[]'); return n.length ? ['**Not built:**', ...n.map(x => `- ${x}`), ''] : []; } catch (e) { return []; } })(),

.github/workflows/scripts/ci-controller-setup.sh

@@ -0,0 +1,70 @@
+#!/usr/bin/env bash
+# Installs ci-vm-controller.sh on the hypervisor (prox1): the script, the
+# token reader, a systemd unit per CI VM, and each VM's config. Run as root
+# on prox1 from a directory that also holds ci-vm-controller.sh:
+#
+#   scp ci-controller-setup.sh ci-vm-controller.sh root@prox1:/root/ci-controller/
+#   ssh root@prox1 'cd /root/ci-controller && bash ci-controller-setup.sh'
+#
+# It does not start anything: a VM goes into service with
+#   systemctl enable --now ci-vm-controller@<vmid>
+# once it has a 'clean' snapshot without any runner registered (see
+# docs/guides/pr-script-testing.md, "Single-use hosts").
+#
+# The token (/etc/ci-controller/token) is put there by hand: a fine-grained
+# token for the two repositories with Administration: read and write only.
+
+set -euo pipefail
+cd "$(dirname "$0")"
+[[ -f ci-vm-controller.sh ]] || { echo "ci-vm-controller.sh missing next to this script" >&2; exit 1; }
+command -v jq >/dev/null || DEBIAN_FRONTEND=noninteractive apt-get install -y -qq jq >/dev/null
+
+install -m 755 ci-vm-controller.sh /usr/local/sbin/ci-vm-controller.sh
+install -d -m 700 /etc/ci-controller
+
+cat >/usr/local/sbin/ci-gh-token <<'EOF'
+#!/bin/sh
+# Prints the token ci-vm-controller.sh uses (Administration: write on the CI
+# repositories). Swap this for a GitHub App token later without touching
+# the controller.
+exec cat /etc/ci-controller/token
+EOF
+chmod 755 /usr/local/sbin/ci-gh-token
+
+cat >/etc/systemd/system/ci-vm-controller@.service <<'EOF'
+[Unit]
+Description=Single-use CI runner host, VM %i
+After=network-online.target pve-cluster.service pvedaemon.service
+Wants=network-online.target
+# A missing token is a setup error, not something to retry every 30 s.
+ConditionPathExists=/etc/ci-controller/token
+
+[Service]
+ExecStart=/usr/local/sbin/ci-vm-controller.sh %i
+Restart=always
+RestartSec=30
+# Stopping the controller leaves the VM as it is; the next start resets it,
+# which also ends a job that may still be running there.
+KillMode=mixed
+
+[Install]
+WantedBy=multi-user.target
+EOF
+
+# One repository per host: see ci-vm-controller.sh.
+conf() { # conf <vmid> <ip> <name> <label> <repo>
+  cat >"/etc/ci-controller/$1.conf" <<EOF
+VM_IP=$2
+HOST_NAME=$3
+LABEL=$4
+REPOS=$5
+EOF
+}
+conf 200 10.201.146.232 pve-ci     proxmox-test community-scripts/DevScripts
+conf 201 10.201.146.231 pve-ci-2   proxmox-test community-scripts/ProxmoxVE
+conf 204 10.201.146.235 pve-ci-3   proxmox-test community-scripts/ProxmoxVE
+conf 202 10.201.146.233 incus-ci   incus-test   community-scripts/DevScripts
+conf 203 10.201.146.234 incus-ci-2 incus-test   community-scripts/ProxmoxVE
+
+systemctl daemon-reload
+echo "Installed. Token: $([[ -s /etc/ci-controller/token ]] && echo present || echo 'MISSING (/etc/ci-controller/token)')"

.github/workflows/scripts/ci-vm-controller.sh

@@ -0,0 +1,201 @@
+#!/usr/bin/env bash
+# Keeps one CI VM (pve-ci, incus-ci, ...) as a single-use GitHub runner host.
+# Runs on the hypervisor (prox1), which PR code never touches.
+#
+# Every cycle:
+#   1. roll the VM back to its clean snapshot and boot it;
+#   2. register a just-in-time runner on it (generate-jitconfig) for the
+#      host's repository. Such a runner takes exactly one job and is deleted
+#      by GitHub afterwards, so its credentials are worthless once that job
+#      is over;
+#   3. (with several repositories in REPOS: as soon as one runner is busy,
+#      delete the others that are still idle, so no second job starts on a
+#      host that is about to be reset);
+#   4. once every runner of this cycle is gone, start over at 1.
+#
+# PR code runs as root on the VM, so nothing the VM says is trusted: when a
+# job is over is read from GitHub's API, never from the VM.
+#
+# Config: /etc/ci-controller/<vmid>.conf (or the environment), see below.
+# Usage:  ci-vm-controller.sh <vmid>          run forever (the systemd unit)
+#         ONE_CYCLE=1 ci-vm-controller.sh <vmid>   one job, then stop (testing)
+
+set -euo pipefail
+
+VMID=${1:?usage: ci-vm-controller.sh <vmid>}
+CONF=${CONF:-/etc/ci-controller/$VMID.conf}
+# shellcheck source=/dev/null
+[[ -f "$CONF" ]] && . "$CONF"
+
+: "${VM_IP:?VM_IP}"                       # address of the CI VM
+: "${HOST_NAME:?HOST_NAME}"               # runner name prefix, e.g. pve-ci
+: "${LABEL:?LABEL}"                       # proxmox-test or incus-test
+# One repository per host. Several would each have a runner waiting on the
+# VM, and a job of one could read the idle runner's JIT config of the other
+# and use it elsewhere before it is withdrawn: one repository's PR code
+# faking the other's results. (Several are fine within one trust domain.)
+: "${REPOS:?REPOS, e.g. community-scripts/DevScripts}"
+EXTRA_LABELS=${EXTRA_LABELS:-}            # space-separated, e.g. for a test
+SNAPSHOT=${SNAPSHOT:-clean}
+TOKEN_CMD=${TOKEN_CMD:-/usr/local/sbin/ci-gh-token}  # prints a token with repo Administration: write
+PVE=${PVE:-}                              # prefix for qm, e.g. "ssh root@prox1" when not run on it
+GUEST_SSH=${GUEST_SSH:-"ssh -o BatchMode=yes -o ConnectTimeout=5 -o StrictHostKeyChecking=accept-new root@$VM_IP"}
+POLL=${POLL:-20}
+MAX_BUSY_MIN=${MAX_BUSY_MIN:-180}         # the test job's own limit is 90
+MAX_IDLE_MIN=${MAX_IDLE_MIN:-720}         # recycle an idle host twice a day
+ONE_CYCLE=${ONE_CYCLE:-}
+
+log() { echo "[$(date -u +%FT%TZ)] vm $VMID: $*"; }
+qm() { if [[ -n "$PVE" ]]; then $PVE qm "$@"; else command qm "$@"; fi; }
+
+# Not in a subshell, so the status and the cached token survive the call:
+# api METHOD PATH [JSON] -> body in $API_BODY, HTTP status in $API_STATUS.
+API_FILE=$(mktemp)
+trap 'rm -f "$API_FILE"' EXIT
+TOKEN="" TOKEN_AT=0
+api() {
+  local method=$1 path=$2 data=${3:-}
+  # Cached for 30 minutes (an App installation token lives 60).
+  if [[ -z "$TOKEN" || $(($(date +%s) - TOKEN_AT)) -gt 1800 ]]; then
+    TOKEN=$($TOKEN_CMD) || { log "no GitHub token"; API_STATUS=000 API_BODY=""; return 1; }
+    TOKEN_AT=$(date +%s)
+  fi
+  API_STATUS=$(curl -sS -o "$API_FILE" -w '%{http_code}' -X "$method" \
+    -H "Authorization: Bearer $TOKEN" -H 'Accept: application/vnd.github+json' \
+    -H 'X-GitHub-Api-Version: 2022-11-28' ${data:+-d "$data"} "https://api.github.com$path") || API_STATUS=000
+  API_BODY=$(<"$API_FILE")
+  [[ "$API_STATUS" != 000 ]]
+}
+
+reset_vm() {
+  log "rolling back to '$SNAPSHOT'"
+  [[ "$(qm status "$VMID")" == *running* ]] && qm stop "$VMID" --timeout 60 >/dev/null
+  qm rollback "$VMID" "$SNAPSHOT" >/dev/null
+  qm start "$VMID" >/dev/null
+  local i
+  for i in $(seq 1 60); do
+    $GUEST_SSH true 2>/dev/null && { log "booted"; return 0; }
+    sleep 5
+  done
+  log "no SSH after 5 minutes"
+  return 1
+}
+
+# Parallel arrays: one entry per runner registered in this cycle.
+R_REPO=() R_ID=() R_GONE=()
+
+register() { # register <owner/repo>
+  local repo=$1 name dir labels body cfg id
+  name="$HOST_NAME-${repo#*/}-$(date +%s)"
+  dir=/opt/actions-runner/${repo#*/}
+  labels=$(printf '%s\n' "$LABEL" $EXTRA_LABELS | jq -R . | jq -sc .)
+  api POST "/repos/$repo/actions/runners/generate-jitconfig" \
+    "$(jq -nc --arg n "$name" --argjson l "$labels" '{name: $n, runner_group_id: 1, labels: $l, work_folder: "_work"}')" || return 1
+  body=$API_BODY
+  [[ "$API_STATUS" == 201 ]] || { log "$repo: generate-jitconfig failed ($API_STATUS): $(jq -r .message <<<"$body" 2>/dev/null)"; return 1; }
+  id=$(jq -r .runner.id <<<"$body")
+  cfg=$(jq -r .encoded_jit_config <<<"$body")
+  # The config goes in through stdin, not the command line.
+  if ! $GUEST_SSH "umask 077; cat >/run/ci-jit-${repo#*/} && systemd-run --quiet --collect --unit=ci-jit-${repo#*/} \
+      -p WorkingDirectory=$dir -E RUNNER_ALLOW_RUNASROOT=1 \
+      /bin/sh -c 'cfg=\$(cat /run/ci-jit-${repo#*/}); rm -f /run/ci-jit-${repo#*/}; exec ./run.sh --jitconfig \"\$cfg\"'" <<<"$cfg"; then
+    log "$repo: could not start the runner on the VM"
+    api DELETE "/repos/$repo/actions/runners/$id" >/dev/null || true
+    return 1
+  fi
+  R_REPO+=("$repo") R_ID+=("$id") R_GONE+=(0)
+  log "$repo: runner $name (id $id) started"
+}
+
+drop() { # drop <index>: delete an idle runner; GitHub refuses a busy one
+  local i=$1
+  api DELETE "/repos/${R_REPO[$i]}/actions/runners/${R_ID[$i]}" >/dev/null || return 1
+  if [[ "$API_STATUS" == 204 || "$API_STATUS" == 404 ]]; then R_GONE[$i]=1; return 0; fi
+  return 1
+}
+
+watch_cycle() {
+  local start busy_since="" i body status busy any_busy left offline_since=""
+  start=$(date +%s)
+  while :; do
+    sleep "$POLL"
+    any_busy=0 left=0 offline=0
+    for i in "${!R_ID[@]}"; do
+      [[ "${R_GONE[$i]}" == 1 ]] && continue
+      api GET "/repos/${R_REPO[$i]}/actions/runners/${R_ID[$i]}" || { left=1; continue; }
+      body=$API_BODY
+      if [[ "$API_STATUS" == 404 ]]; then
+        R_GONE[$i]=1
+        log "${R_REPO[$i]}: runner ${R_ID[$i]} is gone (job done or removed)"
+        continue
+      fi
+      left=1
+      busy=$(jq -r .busy <<<"$body"); status=$(jq -r .status <<<"$body")
+      [[ "$busy" == true ]] && any_busy=1
+      [[ "$status" == offline ]] && offline=1
+    done
+    if [[ "$any_busy" == 1 && -z "$busy_since" ]]; then
+      busy_since=$(date +%s)
+      log "a job started; withdrawing the idle runners"
+    fi
+    # A host that has started a job takes no other one: it will be reset.
+    if [[ -n "$busy_since" ]]; then
+      for i in "${!R_ID[@]}"; do
+        [[ "${R_GONE[$i]}" == 1 ]] && continue
+        api GET "/repos/${R_REPO[$i]}/actions/runners/${R_ID[$i]}" || continue
+        body=$API_BODY
+        [[ "$API_STATUS" == 200 && "$(jq -r .busy <<<"$body")" == false ]] && drop "$i" &&
+          log "${R_REPO[$i]}: idle runner ${R_ID[$i]} withdrawn"
+      done
+    fi
+    [[ "$left" == 0 ]] && { log "cycle over"; return 0; }
+
+    local now; now=$(date +%s)
+    if [[ -n "$busy_since" && $((now - busy_since)) -gt $((MAX_BUSY_MIN * 60)) ]]; then
+      log "job still running after $MAX_BUSY_MIN minutes; resetting anyway"
+      break
+    fi
+    if [[ -z "$busy_since" && $((now - start)) -gt $((MAX_IDLE_MIN * 60)) ]]; then
+      log "idle for $MAX_IDLE_MIN minutes; recycling"
+      break
+    fi
+    # A runner that never comes online (or dies while idle) holds the host
+    # forever otherwise.
+    if [[ -z "$busy_since" && "$offline" == 1 ]]; then
+      [[ -z "$offline_since" ]] && offline_since=$now
+      if [[ $((now - offline_since)) -gt 300 ]]; then log "a runner is offline for 5 minutes; resetting"; break; fi
+    else
+      offline_since=""
+    fi
+  done
+  for i in "${!R_ID[@]}"; do [[ "${R_GONE[$i]}" == 1 ]] || drop "$i" || true; done
+}
+
+# Registrations of this host that a stopped or crashed controller left behind:
+# their VM was reset, so they are offline and never take a job, but they
+# clutter the runner list. Busy or online ones are left alone.
+cleanup_stale() {
+  local repo ids id
+  for repo in $REPOS; do
+    api GET "/repos/$repo/actions/runners?per_page=100" || continue
+    [[ "$API_STATUS" == 200 ]] || continue
+    ids=$(jq -r --arg p "$HOST_NAME-${repo#*/}-" \
+      '.runners[] | select((.name | startswith($p)) and .status == "offline" and (.busy | not)) | .id' <<<"$API_BODY")
+    for id in $ids; do
+      api DELETE "/repos/$repo/actions/runners/$id" && log "$repo: stale runner $id removed"
+    done
+  done
+}
+
+while :; do
+  R_REPO=() R_ID=() R_GONE=()
+  if reset_vm; then
+    cleanup_stale
+    ok=0
+    for repo in $REPOS; do register "$repo" && ok=1 || true; done
+    if [[ "$ok" == 1 ]]; then watch_cycle; else log "no runner could be registered"; sleep 60; fi
+  else
+    sleep 60
+  fi
+  [[ -n "$ONE_CYCLE" ]] && { log "ONE_CYCLE: stopping; the VM is left as the job left it"; break; }
+done

.github/workflows/scripts/incus-ci-host-setup.sh

@@ -6,6 +6,8 @@
 #
 #   incus-ci-host-setup.sh                       host only
 #   incus-ci-host-setup.sh <owner/repo> <token>  host + a runner for that repo
+#   incus-ci-host-setup.sh <owner/repo> jit      host + runner software only, for
+#                                                ci-vm-controller.sh (single-use host)
 #
 # The token is a runner *registration* token (valid one hour):
 #   gh api -X POST repos/<owner/repo>/actions/runners/registration-token --jq .token
@@ -176,6 +178,13 @@ if [[ $# -eq 2 ]]; then
     curl -fsSL "https://github.com/actions/runner/releases/download/v${RUNNER_VERSION}/actions-runner-linux-x64-${RUNNER_VERSION}.tar.gz" | tar -xz
     ./bin/installdependencies.sh >/dev/null
   fi
+  # `jit`: a single-use host. ci-vm-controller.sh on the hypervisor registers
+  # a fresh runner for every job and resets the VM afterwards, so only the
+  # runner software goes on it, and the 'clean' snapshot is taken after this.
+  if [[ "$token" == jit ]]; then
+    echo "runner software for ${repo} installed; ci-vm-controller.sh registers a runner per job"
+    exit 0
+  fi
   if [[ -f .runner ]]; then
     ./svc.sh stop >/dev/null 2>&1 || true
     ./svc.sh uninstall >/dev/null 2>&1 || true

.github/workflows/scripts/pve-ci-host-setup.sh

@@ -5,6 +5,8 @@
 #
 #   pve-ci-host-setup.sh                       host only
 #   pve-ci-host-setup.sh <owner/repo> <token>  host + a runner for that repo
+#   pve-ci-host-setup.sh <owner/repo> jit      host + runner software only, for
+#                                              ci-vm-controller.sh (single-use host)
 #
 # The token is a runner *registration* token (valid one hour):
 #   gh api -X POST repos/<owner/repo>/actions/runners/registration-token --jq .token
@@ -196,6 +198,13 @@ if [[ $# -eq 2 ]]; then
     curl -fsSL "https://github.com/actions/runner/releases/download/v${RUNNER_VERSION}/actions-runner-linux-x64-${RUNNER_VERSION}.tar.gz" | tar -xz
     ./bin/installdependencies.sh >/dev/null
   fi
+  # `jit`: a single-use host. ci-vm-controller.sh on the hypervisor registers
+  # a fresh runner for every job and resets the VM afterwards, so only the
+  # runner software goes on it, and the 'clean' snapshot is taken after this.
+  if [[ "$token" == jit ]]; then
+    echo "runner software for ${repo} installed; ci-vm-controller.sh registers a runner per job"
+    exit 0
+  fi
   # A registered runner is reconfigured from scratch: stop and remove the old
   # registration (the token works for removal too).
   if [[ -f .runner ]]; then

.github/workflows/scripts/pve-run-script.sh

@@ -48,6 +48,16 @@ URL_WAIT="${URL_WAIT:-120}"
 # script that leaves var_arm64 unset; the ARM run builds those anyway, to see.
 VAR_ARM64="${VAR_ARM64:-}"
 [[ -z "$VAR_ARM64" || "$VAR_ARM64" == "yes" ]] || { echo "bad VAR_ARM64" >&2; exit 2; }
+# upgrade: install the script as it is on the base branch (BASE_SHA of
+# UPSTREAM), then update with the PR. That is what a user does who installed
+# before the PR is merged and runs `update` after it. Empty: install and
+# update both from the PR.
+TEST_MODE="${TEST_MODE:-}"
+[[ -z "$TEST_MODE" || "$TEST_MODE" == "upgrade" ]] || { echo "bad TEST_MODE" >&2; exit 2; }
+if [[ -n "$TEST_MODE" ]]; then
+  [[ "${BASE_SHA:-}" =~ ^[0-9a-f]{40}$ ]] || { echo "bad BASE_SHA" >&2; exit 2; }
+  [[ "$SCRIPT" == ct/* ]] || { echo "upgrade tests are for ct/ scripts" >&2; exit 2; }
+fi
 [[ "$PLATFORM" == "pve" || "$SCRIPT" == ct/* ]] || { echo "vm/ scripts run on pve only" >&2; exit 2; }
 [[ "$CTID_BASE" =~ ^[0-9]+$ ]] || { echo "bad CTID_BASE" >&2; exit 2; }
 
@@ -59,12 +69,17 @@ INSTALL_LOG_OUT="$OUT_DIR/install.log"
 UPDATE_RAW="$OUT_DIR/update.raw.log"
 UPDATE_LOG="$OUT_DIR/update.log"
 UPDATE_TIMEOUT="${UPDATE_TIMEOUT:-30m}"
+UPDATE2_RAW="$OUT_DIR/update2.raw.log"
+UPDATE2_LOG="$OUT_DIR/update2.log"
+# How long the app has to keep running without a service restarting.
+STABLE_WAIT="${STABLE_WAIT:-60}"
 
 kind="${SCRIPT%%/*}" # ct or vm
 app=$(basename "$SCRIPT" .sh)
-job="${GITHUB_RUN_ID:-local}-${GITHUB_RUN_ATTEMPT:-1}-${PLATFORM}-${kind}-${app}-${OS:-default}"
+job="${GITHUB_RUN_ID:-local}-${GITHUB_RUN_ATTEMPT:-1}-${PLATFORM}-${kind}-${app}-${OS:-default}${TEST_MODE:+-$TEST_MODE}"
 tree="$SERVE_ROOT/$job"
 base="$SERVE_URL/$job"
+src_tree="$tree" # what gets installed; the base branch's tree in an upgrade test
 started=$(date +%s)
 
 status="fail"
@@ -102,6 +117,16 @@ failed_units=""
 # fail: the update broke; empty: not attempted (the install failed).
 update_status=""
 update_step=""
+# install: building and checking what was installed; update: from the first
+# update on. In an upgrade test a failure while installing is the base
+# branch's, not the PR's.
+phase="install"
+# Findings that do not fail the build, shown under the table on the PR.
+notes=()
+# Disk, RAM and OOM kills of the container, measured at the end.
+resources=""
+# Second `update` right after the first: no-update, updated-again, or empty.
+update2_status=""
 
 # Spinner frames: the engine redraws "⠋ Starting VM" a few times a second
 # without a tty too. An alternation, not a bracket: these are multibyte and
@@ -163,13 +188,61 @@ classify_cpu_failure() {
   step="needs a newer CPU than the test host has (no ${feature}): ${line}"
 }
 
+# Disk used against var_disk, RAM in use at rest (anonymous memory, without
+# the page cache that fills any limit) against the limit, and how often the
+# kernel killed something for lack of memory, over the container's whole life,
+# the build included. A dir-backed Incus root has no size of its own: df shows
+# the host's disk. Then what the container holds is counted with du, and the
+# script's var_disk is the size to compare with.
+measure_resources() {
+  local out disk_size disk_used anon limit oom want_disk want_ram line=()
+  out=$(guest_exec sh -c 'df -Pm / | awk "NR==2 {print \$2, \$3}"
+    awk "/^anon / {print int(\$2 / 1048576)}" /sys/fs/cgroup/memory.stat 2>/dev/null || echo -
+    cat /sys/fs/cgroup/memory.max 2>/dev/null || echo -
+    awk "/^oom_kill / {print \$2}" /sys/fs/cgroup/memory.events 2>/dev/null || echo -' 2>/dev/null) || return 0
+  read -r disk_size disk_used <<<"$(sed -n 1p <<<"$out")"
+  anon=$(sed -n 2p <<<"$out"); limit=$(sed -n 3p <<<"$out"); oom=$(sed -n 4p <<<"$out")
+  want_disk=$(grep -m1 -oE 'var_disk="\$\{var_disk:-[0-9]+\}"' "$src_tree/$SCRIPT" | grep -oE '[0-9]+' | head -n1)
+  want_ram=$(grep -m1 -oE 'var_ram="\$\{var_ram:-[0-9]+\}"' "$src_tree/$SCRIPT" | grep -oE '[0-9]+' | head -n1)
+  [[ "$disk_size" =~ ^[0-9]+$ && "$disk_used" =~ ^[0-9]+$ ]] || return 0
+  if [[ -n "$want_disk" ]] && ((disk_size > want_disk * 1024 * 3 / 2)); then
+    disk_size=$((want_disk * 1024))
+    disk_used=$(guest_exec sh -c 'du -sxm / 2>/dev/null | cut -f1' 2>/dev/null)
+    [[ "$disk_used" =~ ^[0-9]+$ ]] || return 0
+  fi
+  line+=("disk $(awk -v u="$disk_used" 'BEGIN {printf "%.1f", u / 1024}') of $(awk -v s="$disk_size" 'BEGIN {printf "%.0f", s / 1024}') GB")
+  if ((disk_used * 100 >= disk_size * 90)); then
+    notes+=("disk ${disk_used} of ${disk_size} MB used after the install: var_disk may be too small")
+  fi
+  [[ "$limit" =~ ^[0-9]+$ ]] && limit=$((limit / 1048576)) || limit="${want_ram:-}"
+  if [[ "$anon" =~ ^[0-9]+$ ]]; then
+    line+=("RAM at rest ${anon}${limit:+ of ${limit}} MB")
+    if [[ "$limit" =~ ^[0-9]+$ ]] && ((anon * 100 >= limit * 80)); then
+      notes+=("${anon} of ${limit} MB RAM in use at rest: var_ram may be too small")
+    fi
+  fi
+  if [[ "$oom" =~ ^[0-9]+$ ]]; then
+    line+=("OOM kills ${oom}")
+    ((oom > 0)) && notes+=("the kernel killed ${oom} process(es) for lack of memory: var_ram is too small for the build or the app")
+  fi
+  resources=$(
+    IFS='|'
+    echo "${line[*]}"
+  )
+  resources=${resources//|/ · }
+}
+
 finish() {
   [[ -f "$RAW_LOG" ]] && strip_ansi <"$RAW_LOG" >"$LOG"
   [[ -f "$RAW_LOG" ]] && collect_install_log
   [[ -f "$UPDATE_RAW" ]] && strip_ansi <"$UPDATE_RAW" >"$UPDATE_LOG"
-  rm -f "$UPDATE_RAW"
+  [[ -f "$UPDATE2_RAW" ]] && strip_ansi <"$UPDATE2_RAW" >"$UPDATE2_LOG"
+  rm -f "$UPDATE_RAW" "$UPDATE2_RAW"
+  [[ "$kind" == ct && -n "$ctid" && -z "$resources" ]] && guest_running && measure_resources
   jq -n \
     --arg script "$SCRIPT" --arg kind "$kind" --arg platform "$PLATFORM" --arg os "${OS:-default}" --arg status "$status" \
+    --arg mode "$TEST_MODE" --arg resources "$resources" --arg update2_status "$update2_status" \
+    --arg notes "$(printf '%s\n' "${notes[@]}")" \
     --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 "$failed_units" --arg ip "$ct_ip" --arg arch "$(uname -m)" \
@@ -179,6 +252,8 @@ finish() {
     '{script: $script, kind: $kind, platform: $platform, arch: $arch, os: $os, status: $status, step: $step, exit_code: $exit_code,
       sha: $sha, ctid: $ctid, ip: $ip, url: $url, http_code: $http_code,
       failed_units: $failed_units, update_status: $update_status, update_step: $update_step,
+      mode: $mode, resources: $resources, update2_status: $update2_status,
+      notes: ($notes | split("\n") | map(select(. != ""))),
       duration: $duration, log_tail: $tail}' >"$RESULT"
   echo "::group::result"
   jq 'del(.log_tail)' "$RESULT"
@@ -262,7 +337,7 @@ cleanup() {
     done
   fi
   rm -f "$RAW_LOG.start"
-  rm -rf -- "$tree"
+  rm -rf -- "$tree" "$tree-base"
 }
 trap cleanup EXIT
 # A cancelled job sends TERM; without this the EXIT trap would not run.
@@ -281,6 +356,10 @@ fail() {
   step="$1"
   status="${2:-fail}"
   [[ "$status" == "unsupported" ]] || status="fail"
+  if [[ "$TEST_MODE" == "upgrade" && "$phase" == "install" && "$status" == "fail" ]]; then
+    status="error"
+    step="the install before this PR (${UPSTREAM}@${BASE_SHA:0:7}) failed, so the upgrade was not tested: $1"
+  fi
   finish
   echo "::warning title=${SCRIPT}::${step}"
   exit 0
@@ -335,20 +414,37 @@ upstream_re="https://raw\.githubusercontent\.com/${UPSTREAM//./\\.}/(refs/heads/
 grep -rlE "$upstream_re" "$tree" --include='*.sh' --include='*.func' |
   xargs -r sed -i -E "s#${upstream_re}#${base}#g"
 
+# What gets installed: the PR, or in an upgrade test the base branch.
+src_tree="$tree"
+src_base="$base"
+if [[ "$TEST_MODE" == "upgrade" ]]; then
+  src_tree="$tree-base"
+  src_base="$base-base"
+  mkdir -p "$src_tree"
+  curl -fsSL --retry 3 "https://codeload.github.com/${UPSTREAM}/tar.gz/${BASE_SHA}" |
+    tar -xz -C "$src_tree" --strip-components=1 || setup_error "could not download ${UPSTREAM}@${BASE_SHA:0:7}"
+  [[ -f "$src_tree/$SCRIPT" ]] || fail "${SCRIPT} does not exist at ${UPSTREAM}@${BASE_SHA:0:7}"
+  grep -rlE "$upstream_re" "$src_tree" --include='*.sh' --include='*.func' |
+    xargs -r sed -i -E "s#${upstream_re}#${src_base}#g"
+fi
+
 if [[ "$kind" == "vm" ]]; then
   vm_build_and_check
   exit 0
 fi
 
 # --- Build ----------------------------------------------------------------------
-echo "Building ${SCRIPT} (${OS:-default OS}) on ${PLATFORM} as CT ${ctid} from ${PR_REPO}@${PR_SHA:0:7}"
+if [[ "$TEST_MODE" == "upgrade" ]]; then
+  echo "Upgrade test: installing ${SCRIPT} from ${UPSTREAM}@${BASE_SHA:0:7} (before this PR), then updating with ${PR_REPO}@${PR_SHA:0:7}"
+fi
+echo "Building ${SCRIPT} (${OS:-default OS}) on ${PLATFORM} as CT ${ctid} from $([[ "$TEST_MODE" == upgrade ]] && echo "${UPSTREAM}@${BASE_SHA:0:7}" || echo "${PR_REPO}@${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_brg="$BRIDGE"
-  "dev_mode=timing,keep" COMMUNITY_SCRIPTS_URL="$base"
+  "dev_mode=timing,keep" COMMUNITY_SCRIPTS_URL="$src_base"
 )
 if [[ "$PLATFORM" == "incus" ]]; then
   build_env+=(var_container_storage=default)
@@ -387,7 +483,7 @@ ci_value() {
   esac
 }
 declare -A answers=()
-json="$tree/json/${app}.json"
+json="$src_tree/json/${app}.json"
 if [[ -f "$json" ]]; then
   while IFS=$'\t' read -r name default; do
     answers[$name]="$default"
@@ -419,8 +515,8 @@ expand_templates() {
 
 # 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/$SCRIPT"; do
+nsapp=$(grep -m1 -oE '^APP="[^"]+"' "$src_tree/$SCRIPT" | cut -d'"' -f2 | tr -d ' ' | tr '[:upper:]' '[:lower:]')
+for f in "$src_tree/install/${app}-install.sh" "$src_tree/install/${nsapp}-install.sh" "$src_tree/$SCRIPT"; do
   [[ -f "$f" ]] || continue
   while read -r name; do
     [[ -v "answers[$name]" ]] || answers[$name]=""
@@ -443,7 +539,7 @@ done
 # In the background so that a TERM from a cancelled job is handled at once:
 # bash defers traps until a foreground command returns.
 env -i "${build_env[@]}" timeout --kill-after=1m "$BUILD_TIMEOUT" \
-  bash -c 'bash -c "$(curl -fsSL "$1")"' _ "$base/$SCRIPT" </dev/null >"$RAW_LOG" 2>&1 &
+  bash -c 'bash -c "$(curl -fsSL "$1")"' _ "$src_base/$SCRIPT" </dev/null >"$RAW_LOG" 2>&1 &
 build_pid=$!
 start_streams
 wait "$build_pid"
@@ -522,8 +618,12 @@ take_screenshot() {
 # open, e.g. http://${IP}:1337/admin, answers 200 (after redirects, so a jump
 # to /login is fine). A script that prints no URL has no web UI to check.
 # Run after the install and again after the update. Sets health_problem.
-pattern=$(grep -oE 'https?://\$\{IP\}(:[0-9]+)?(/[A-Za-z0-9._~/?=&%-]*)?' "$tree/$SCRIPT" | head -n1)
-[[ -n "$pattern" && -n "$ct_ip" ]] && url="${pattern/\$\{IP\}/$ct_ip}"
+app_url() { # the URL the script at $1 tells the user to open
+  local pattern
+  pattern=$(grep -oE 'https?://\$\{IP\}(:[0-9]+)?(/[A-Za-z0-9._~/?=&%-]*)?' "$1" | head -n1)
+  [[ -n "$pattern" && -n "$ct_ip" ]] && echo "${pattern/\$\{IP\}/$ct_ip}"
+}
+url=$(app_url "$src_tree/$SCRIPT")
 health_problem=""
 check_health() {
   health_problem=""
@@ -555,7 +655,9 @@ check_health screenshot.png || fail "$health_problem"
 # only say "No update available" and never run update_script's real path. The
 # version files the install wrote (~/.<app>, one per fetch_and_deploy_gh_release)
 # that update_script checks are set back to 0.0.0-ci first; check_for_gh_release then sees an update and
-# runs it all. `update` is the /usr/bin/update the install generated: it pulls
+# runs it all. Apps deployed from a branch (fetch_and_deploy_gh_branch, checked
+# with check_for_gh_branch) store a commit hash instead of a version; the
+# engine compares both as plain strings, so 0.0.0-ci works for them too. `update` is the /usr/bin/update the install generated: it pulls
 # ct/<app>.sh from the URL this container was built from, i.e. the PR's tree.
 UPDATE_FAKE="0.0.0-ci"
 update_fail() {
@@ -568,7 +670,7 @@ version_files=$(guest_exec sh -c '
   for f in /root/.*; do
     [ -f "$f" ] || continue
     case "$f" in */.install-* | *.pending | *.bak | */.bash* | */.profile | */.hushlogin | */.viminfo | */.lesshst | */.wget-hsts | */.selected_editor) continue ;; esac
-    [ "$(wc -l <"$f")" -le 1 ] && grep -qE "^v?[0-9]+(\.[0-9A-Za-z_+-]+)+$" "$f" && echo "$f"
+    [ "$(wc -l <"$f")" -le 1 ] && grep -qE "^(v?[0-9]+(\.[0-9A-Za-z_+-]+)+|[0-9a-f]{7,40})$" "$f" && echo "$f"
   done' 2>/dev/null | grep -E '^/root/\.[A-Za-z0-9._-]+$')
 # Only set back what update_script checks: the names it passes to
 # check_for_*release. The install can also deploy helpers the update never
@@ -585,24 +687,41 @@ for f in $version_files; do
 done
 [[ -n "$version_files" ]] && echo "Update test: set back $(tr '\n' ' ' <<<"$version_files")to ${UPDATE_FAKE}"
 
-: >"$UPDATE_RAW"
-update_token=$(head -c 16 /dev/urandom | od -An -tx1 | tr -d ' \n')
-echo "──────── update test (live) ────────"
-echo "::stop-commands::${update_token}"
-# timeout runs a program, not a shell function, so spell out the exec here.
-if [[ "$PLATFORM" == "incus" ]]; then guest_cmd=(incus exec "$ctid" --); else guest_cmd=(pct exec "$ctid" --); fi
-timeout --kill-after=1m "$UPDATE_TIMEOUT" "${guest_cmd[@]}" env PHS_SILENT=1 mode=default MODE=default \
-  TERM=dumb DIAGNOSTICS=no var_ignore_breaking_changes=yes ${GH_API_TOKEN:+GITHUB_TOKEN="$GH_API_TOKEN"} update </dev/null \
-  > >(tee "$UPDATE_RAW" | live_filter) 2>&1 &
-update_pid=$!
-wait "$update_pid"
-update_rc=$?
-update_pid=""
-sleep 1
-echo "::${update_token}::"
+phase="update"
+# Upgrade test: the container pulls its update from where it was installed
+# from (COMMUNITY_SCRIPTS_URL in /usr/bin/update); point that at the PR, as
+# merging it would.
+if [[ "$TEST_MODE" == "upgrade" ]]; then
+  guest_exec sed -i -E "s#^export COMMUNITY_SCRIPTS_URL=.*#export COMMUNITY_SCRIPTS_URL=\"${base}\"#" /usr/bin/update &&
+    guest_exec grep -qxF "export COMMUNITY_SCRIPTS_URL=\"${base}\"" /usr/bin/update ||
+    setup_error "could not point /usr/bin/update at the PR"
+  echo "Upgrade test: /usr/bin/update now pulls from ${PR_REPO}@${PR_SHA:0:7}"
+fi
+
+# run_update <raw log> <title>: runs `update` in the container, live in the
+# job log; sets update_rc.
+run_update() {
+  local raw=$1 token
+  : >"$raw"
+  token=$(head -c 16 /dev/urandom | od -An -tx1 | tr -d ' \n')
+  echo "──────── $2 (live) ────────"
+  echo "::stop-commands::${token}"
+  # timeout runs a program, not a shell function, so spell out the exec here.
+  if [[ "$PLATFORM" == "incus" ]]; then guest_cmd=(incus exec "$ctid" --); else guest_cmd=(pct exec "$ctid" --); fi
+  timeout --kill-after=1m "$UPDATE_TIMEOUT" "${guest_cmd[@]}" env PHS_SILENT=1 mode=default MODE=default \
+    TERM=dumb DIAGNOSTICS=no var_ignore_breaking_changes=yes ${GH_API_TOKEN:+GITHUB_TOKEN="$GH_API_TOKEN"} update </dev/null \
+    > >(tee "$raw" | live_filter) 2>&1 &
+  update_pid=$!
+  wait "$update_pid"
+  update_rc=$?
+  update_pid=""
+  sleep 1
+  echo "::${token}::"
+}
+run_update "$UPDATE_RAW" "update test"
 
 update_failure_line() {
-  strip_ansi <"$UPDATE_RAW" | grep -E '✖|\[ERROR\]|error:' | tail -n 1 | sed -E 's/^\s*(✖️?|\s)*//' | cut -c1-180
+  strip_ansi <"${1:-$UPDATE_RAW}" | grep -E '✖|\[ERROR\]|error:' | tail -n 1 | sed -E 's/^\s*(✖️?|\s)*//' | cut -c1-180
 }
 if ((update_rc == 124 || update_rc == 137)); then
   update_fail "timed out after ${UPDATE_TIMEOUT}"
@@ -637,7 +756,69 @@ else
   update_step=$(strip_ansi <"$UPDATE_RAW" | grep -iE 'no update|up to date|up-to-date|already' | tail -n 1 | sed -E 's/^\s*(✔️?|ℹ️?|\s)*//' | cut -c1-160)
 fi
 
+url=$(app_url "$tree/$SCRIPT")
 check_health screenshot-update.png || update_fail "after the update: $health_problem"
 
+# --- Second update ------------------------------------------------------------------
+# Right after an update, `update` has nothing to do: it must say so and leave
+# the app alone. One that breaks on a second run breaks for every user on
+# their next update; one that updates again although nothing changed
+# rebuilds on every run (often a version file it never writes).
+if [[ "$update_status" == "updated" ]]; then
+  run_update "$UPDATE2_RAW" "second update (must have nothing to do)"
+  if ((update_rc == 124 || update_rc == 137)); then
+    update_fail "second run timed out after ${UPDATE_TIMEOUT}"
+  elif ((update_rc != 0)); then
+    last=$(update_failure_line "$UPDATE2_RAW")
+    update_fail "second run exited ${update_rc}${last:+: $last}"
+  fi
+  update_error=$(strip_ansi <"$UPDATE2_RAW" | grep -E '✖|\[ERROR\]' | tail -n 1 | sed -E 's/^\s*(✖️?|\s)*//' | cut -c1-180)
+  if [[ -n "$update_error" && "${update_error,,}" != *"rate limit"* ]]; then
+    update_fail "second run: $update_error"
+  fi
+  if strip_ansi <"$UPDATE2_RAW" | grep -qi 'updated successfully'; then
+    update2_status="updated-again"
+    notes+=("a second \`update\` right after the first updated again: update_script does not notice it is current (e.g. a version file it never writes)")
+  else
+    update2_status="no-update"
+  fi
+  check_health screenshot-update.png || update_fail "after the second update: $health_problem"
+fi
+
+# --- Does it stay up? -----------------------------------------------------------
+# A service that crash-loops can be "active" at every single look. Count the
+# restarts of every service (and Docker container) over STABLE_WAIT seconds:
+# any that went up was restarted while nobody touched the container.
+restart_counts() {
+  guest_exec sh -c '
+    if command -v systemctl >/dev/null; then
+      systemctl list-units --type=service --all --no-legend --plain | awk "{print \$1}" |
+        while read -r u; do echo "$u $(systemctl show -p NRestarts --value "$u" 2>/dev/null)"; done
+    fi
+    if command -v docker >/dev/null; then
+      docker ps -aq | xargs -r docker inspect -f "docker:{{.Name}} {{.RestartCount}}" 2>/dev/null
+    fi' 2>/dev/null | awk '$2 ~ /^[0-9]+$/' | sort
+}
+before=$(restart_counts)
+sleep "$STABLE_WAIT"
+after=$(restart_counts)
+restarting=$(join <(echo "$before") <(echo "$after") | awk '$3 > $2 {printf "%s%s (+%d)", sep, $1, $3 - $2; sep=", "}')
+[[ -z "$restarting" ]] || fail "not stable: restarted within ${STABLE_WAIT}s without anyone touching it: ${restarting}"
+check_health screenshot-update.png || fail "not stable: ${STABLE_WAIT}s later, ${health_problem}"
+
+# Error lines in the journal since boot: often harmless, sometimes the reason
+# a feature is broken. Information only.
+if guest_exec sh -c 'command -v journalctl' &>/dev/null; then
+  # Without the noise every container produces: udev may not trigger devices
+  # in an unprivileged container, and logind finds no VTs.
+  journal_errors=$(guest_exec journalctl -b -p err -q -o cat --no-pager 2>/dev/null | grep -vE '^\s*$' |
+    grep -vE "Failed to write 'add' to '/sys/.*/uevent'|Seat has no VTs but VT number not 0" | sort | uniq -c | sort -rn)
+  if [[ -n "$journal_errors" ]]; then
+    n=$(awk '{s += $1} END {print s}' <<<"$journal_errors")
+    top=$(head -n 2 <<<"$journal_errors" | sed -E 's/^\s*[0-9]+\s+//' | cut -c1-120 | paste -sd'|' - | sed 's/|/; /g')
+    notes+=("${n} error line(s) in the journal since boot, e.g. ${top}")
+  fi
+fi
+
 status="pass"
 finish

CHANGELOG.md

@@ -593,6 +593,10 @@ Exercise vigilance regarding copycat or coat-tailing sites that seek to exploit
 
     - Whisparr-eros: set the first login and save it [@MickLesk](https://github.com/MickLesk) ([#18617](https://github.com/community-scripts/ProxmoxVE/pull/18617))
 
+### 📚 Documentation
+
+  - pocketbase-bot: apply website metadata requests from Accept/Decline boxes [@MickLesk](https://github.com/MickLesk) ([#18631](https://github.com/community-scripts/ProxmoxVE/pull/18631))
+
 ## 2026-11-11
 
 ### 🚀 Updated Scripts

ct/dotnetaspwebapi.sh

@@ -11,8 +11,8 @@ var_tags="${var_tags:-web}"
 var_cpu="${var_cpu:-1}"
 var_ram="${var_ram:-1024}"
 var_disk="${var_disk:-8}"
-var_os="${var_os:-Debian}"
-var_version="${var_version:-13}"
+var_os="${var_os:-ubuntu}"
+var_version="${var_version:-24.04}"
 var_arm64="${var_arm64:-yes}"
 var_unprivileged="${var_unprivileged:-0}"