Support managed Python and remote Docker in CI; stop collector promptly

2026-10-17 13:56:26 +02:00
parent d1978abb05
commit afacb32b6a
4 changed files with 65 additions and 40 deletions
@@ -13,7 +13,8 @@ jobs:
      - uses: actions/checkout@v4
      - name: Install tools
        run: |
-          python3 -m pip install --user 'PyYAML==6.0.3'
+          python3 -m pip install --target "${RUNNER_TEMP}/python" 'PyYAML==6.0.3'
+          echo "PYTHONPATH=${RUNNER_TEMP}/python" >> "$GITHUB_ENV"
          curl -fsSL https://dl.k8s.io/release/v1.35.3/bin/linux/amd64/kubectl -o /usr/local/bin/kubectl
          chmod 0755 /usr/local/bin/kubectl
      - name: Test policy and boot the actual game configuration
@@ -36,7 +37,8 @@ jobs:
      - uses: actions/checkout@v4
      - name: Install tools
        run: |
-          python3 -m pip install --user 'PyYAML==6.0.3'
+          python3 -m pip install --target "${RUNNER_TEMP}/python" 'PyYAML==6.0.3'
+          echo "PYTHONPATH=${RUNNER_TEMP}/python" >> "$GITHUB_ENV"
          curl -fsSL https://dl.k8s.io/release/v1.35.3/bin/linux/amd64/kubectl -o /usr/local/bin/kubectl
          chmod 0755 /usr/local/bin/kubectl
      - name: Deploy using the dedicated Factorio service account
@@ -1,6 +1,7 @@
 #!/bin/sh
 # Only ordinary console commands: no /c, /sc, Lua, mods, or achievement changes.
 set -eu
+trap 'exit 0' TERM INT
 metrics_dir=${METRICS_DIR:-/metrics}
 mkdir -p "$metrics_dir"
 while :; do
@@ -1,43 +1,65 @@
-"""Boot the actual server image with the actual rendered configuration, then probe/save/restart it."""
-import json,os,secrets,subprocess,sys,tempfile,time
+"""Boot the rendered image/configuration via Docker, including remote Docker sockets."""
+import secrets
+import subprocess
+import sys
+import tempfile
+import time
 from pathlib import Path
 import yaml

-docs=list(yaml.safe_load_all(Path(sys.argv[1]).read_text()))
-d=next(d for d in docs if d['kind']=='Deployment')
-c=next(c for c in d['spec']['template']['spec']['containers'] if c['name']=='factorio')
-cms={d['metadata']['name']:d['data'] for d in docs if d['kind']=='ConfigMap'}
-run=lambda *a,**kw: subprocess.run(a,check=True,**kw)
-with tempfile.TemporaryDirectory() as directory:
-    root=Path(directory)
-    for name in ['scripts','settings','credentials','data']:(root/name).mkdir()
-    for folder,name in [('scripts','factorio-scripts'),('settings','factorio-settings')]:
-        for key,value in cms[name].items():
-            file=root/folder/key; file.write_text(value); file.chmod(0o755 if key.endswith('.sh') else 0o644)
-    (root/'credentials/rcon-password').write_text(secrets.token_urlsafe(32))
-    args=['docker','run','--detach','--name','factorio-smoke-'+secrets.token_hex(5)]
-    for env in c['env']:args+=['--env',env['name']+'='+env['value']]
-    for mount in c['volumeMounts']:
-        folder={'data':'data','scripts':'scripts','settings':'settings','credentials':'credentials'}[mount['name']]
-        args+=['--volume',f'{root/folder}:{mount["mountPath"]}'+(':ro' if mount.get('readOnly') else '')]
-    args+=['--entrypoint',c['command'][0],c['image'],*c['command'][1:]]
-    container=subprocess.check_output(args,text=True).strip()
-    def ready():
-        command=c['readinessProbe']['exec']['command']
-        for _ in range(90):
-            if subprocess.run(['docker','exec',container,*command],stdout=subprocess.DEVNULL,stderr=subprocess.DEVNULL,timeout=15).returncode==0:return
-            time.sleep(2)
-        raise RuntimeError('Server did not pass the production readiness probe')
-    try:
+docs = list(yaml.safe_load_all(Path(sys.argv[1]).read_text()))
+d = next(d for d in docs if d['kind'] == 'Deployment')
+c = next(c for c in d['spec']['template']['spec']['containers'] if c['name'] == 'factorio')
+cms = {d['metadata']['name']: d['data'] for d in docs if d['kind'] == 'ConfigMap'}
+run = lambda *a, **kw: subprocess.run(a, check=True, **kw)
+volume = 'factorio-smoke-' + secrets.token_hex(6)
+run('docker', 'volume', 'create', volume, stdout=subprocess.DEVNULL)
+container = None
+try:
+    with tempfile.TemporaryDirectory() as directory:
+        root = Path(directory)
+        for name in ['scripts', 'settings', 'credentials']:
+            (root / name).mkdir()
+        for folder, name in [('scripts', 'factorio-scripts'), ('settings', 'factorio-settings')]:
+            for key, value in cms[name].items():
+                file = root / folder / key
+                file.write_text(value)
+                file.chmod(0o755 if key.endswith('.sh') else 0o644)
+        (root / 'credentials/rcon-password').write_text(secrets.token_urlsafe(32))
+        args = ['docker', 'create']
+        for env in c['env']:
+            args += ['--env', env['name'] + '=' + env['value']]
+        data_mount = next(m['mountPath'] for m in c['volumeMounts'] if m['name'] == 'data')
+        args += ['--volume', f'{volume}:{data_mount}']
+        args += ['--entrypoint', c['command'][0], c['image'], *c['command'][1:]]
+        container = subprocess.check_output(args, text=True).strip()
+        # docker cp works when the daemon is outside the CI job container;
+        # bind-mounting the job container's /tmp would silently mount a different directory.
+        for mount in c['volumeMounts']:
+            if mount['name'] != 'data':
+                run('docker', 'cp', str(root / mount['name']), container + ':' + mount['mountPath'])
+        run('docker', 'start', container, stdout=subprocess.DEVNULL)
+
+        def ready():
+            command = c['readinessProbe']['exec']['command']
+            for _ in range(90):
+                if subprocess.run(['docker', 'exec', container, *command], stdout=subprocess.DEVNULL,
+                                  stderr=subprocess.DEVNULL, timeout=15).returncode == 0:
+                    return
+                time.sleep(2)
+            raise RuntimeError('Server did not pass the production readiness probe')
+
        ready()
-        run('docker','exec',container,'rcon','/server-save')
-        run('docker','stop','--time','180',container,stdout=subprocess.DEVNULL)
-        assert list((root/'data/saves').glob('*.zip')), 'No persistent save'
-        run('docker','start',container,stdout=subprocess.DEVNULL)
+        run('docker', 'exec', container, 'rcon', '/server-save')
+        run('docker', 'stop', '--time', str(d['spec']['template']['spec']['terminationGracePeriodSeconds']),
+            container, stdout=subprocess.DEVNULL)
+        run('docker', 'start', container, stdout=subprocess.DEVNULL)
        ready()
-        assert list((root/'data/upgrade-backups').glob('*/*.zip')), 'No pre-upgrade backup'
+        run('docker', 'exec', container, 'sh', '-c',
+            'test -n "$(find "$1/saves" -name "*.zip")" && test -n "$(find "$1/upgrade-backups" -name "*.zip")"',
+            'sh', data_mount)
        print('Actual image passed RCON readiness, persistent save, restart, and upgrade-backup checks')
-    finally:
-        run('docker','rm','--force',container,stdout=subprocess.DEVNULL)
-        # The upstream image owns its world files; cleanup through the same image.
-        run('docker','run','--rm','--volume',f'{root}/data:/cleanup','--entrypoint','sh',c['image'],'-c','rm -rf /cleanup/*',stdout=subprocess.DEVNULL)
+finally:
+    if container:
+        run('docker', 'rm', '--force', container, stdout=subprocess.DEVNULL)
+    run('docker', 'volume', 'rm', volume, stdout=subprocess.DEVNULL)
@@ -14,7 +14,7 @@ class CollectorTests(unittest.TestCase):
            fake=root/'rcon'
            fake.write_text('#!/bin/sh\n[ "$*" = "/players online" ] || exit 88\nprintf "%s\\n" "$RESPONSE"\nexit "$CODE"\n')
            fake.chmod(0o755)
-            proc=subprocess.Popen(['sh',str(SCRIPT)],env={**os.environ,'PATH':tmp+':'+os.environ['PATH'],'METRICS_DIR':tmp,'RESPONSE':response,'CODE':str(code)},start_new_session=True)
+            proc=subprocess.Popen(['sh',str(SCRIPT)],env={**os.environ,'PATH':tmp+':'+os.environ['PATH'],'METRICS_DIR':tmp,'RESPONSE':response,'CODE':str(code)},start_new_session=True,stderr=subprocess.DEVNULL)
            try:
                output=root/'factorio.prom'
                for _ in range(100):