From 5a77648aa2ba2b5c54343304630abcce7e429131 Mon Sep 17 00:00:00 2001
From: Christophe Monniez <moc@odoo.com>
Date: Fri, 5 Jun 2026 11:08:08 +0200
Subject: [PATCH] [IMP] runbot: restart nginx when needed

When the main nginx process is restarted e.g. during unattended
upgrades, it happens that the nginx process handled by the runbot
builder needs to be restarted. Otherwise we lose conections with
running instances.

With this commit, the runbot builder will restart the runbot controlled
nginx if the main nginx process pid changes.
---
 runbot/models/runbot.py | 52 ++++++++++++++++++++++++++++++++++-------
 1 file changed, 44 insertions(+), 8 deletions(-)

diff --git a/runbot/models/runbot.py b/runbot/models/runbot.py
index 195efe4a7..884d07a9a 100644
--- a/runbot/models/runbot.py
+++ b/runbot/models/runbot.py
@@ -152,6 +152,48 @@ def _allocate_builds(self, host, nb_slots, domain=None):
                     RETURNING id""", host.name, query.select(), nb_slots))
         return self.env.cr.fetchall()
 
+    def _get_main_nginx_pid(self):
+        # not under an Odoo root, so plain open() rather than file_open()
+        try:
+            with open('/var/run/nginx/nginx.pid', encoding='utf-8') as f:
+                return int(f.read().strip())
+        except (FileNotFoundError, ValueError):
+            return None
+
+    def _start_nginx(self):
+        nginx_dir = self.env['runbot.runbot']._path('nginx')
+        _logger.info('start nginx')
+        main_nginx_pid = self._get_main_nginx_pid()
+        with file_open(self.env['runbot.runbot']._path('nginx', 'main_nginx.pid'), 'w') as f:
+            f.write(str(main_nginx_pid) if main_nginx_pid is not None else '')
+
+        if subprocess.call(['/usr/sbin/nginx', '-p', nginx_dir, '-c', 'nginx.conf']):
+            # obscure nginx bug leaving orphan worker listening on nginx port
+            if not subprocess.call(['pkill', '-f', '-P1', 'nginx: worker']):
+                _logger.warning('failed to start nginx - orphan worker killed, retrying')
+                subprocess.call(['/usr/sbin/nginx', '-p', nginx_dir, '-c', 'nginx.conf'])
+            else:
+                _logger.warning('failed to start nginx - failed to kill orphan worker - oh well')
+
+    def _check_main_nginx(self):
+        try:
+            with file_open(self.env['runbot.runbot']._path('nginx', 'main_nginx.pid')) as f:
+                known_main_nginx_pid = int(f.read().strip())
+        except (FileNotFoundError, ValueError):
+            self._start_nginx()
+            return
+
+        current_main_nginx_pid = self._get_main_nginx_pid()
+        if current_main_nginx_pid is not None and known_main_nginx_pid != current_main_nginx_pid:
+            _logger.info('main nginx restarted (%s -> %s), restarting runbot nginx',
+                         known_main_nginx_pid, current_main_nginx_pid)
+            try:
+                with file_open(self.env['runbot.runbot']._path('nginx', 'nginx.pid')) as f:
+                    os.kill(int(f.read().strip()), signal.SIGTERM)
+            except (FileNotFoundError, ValueError, ProcessLookupError):
+                pass
+            self._start_nginx()
+
     def _reload_nginx(self):
         env = self.env
         settings = {}
@@ -182,14 +224,8 @@ def _reload_nginx(self):
                 pid = int(file_open(self.env['runbot.runbot']._path('nginx', 'nginx.pid')).read().strip(' \n'))
                 os.kill(pid, signal.SIGHUP)
             except Exception:
-                _logger.info('start nginx')
-                if subprocess.call(['/usr/sbin/nginx', '-p', nginx_dir, '-c', 'nginx.conf']):
-                    # obscure nginx bug leaving orphan worker listening on nginx port
-                    if not subprocess.call(['pkill', '-f', '-P1', 'nginx: worker']):
-                        _logger.warning('failed to start nginx - orphan worker killed, retrying')
-                        subprocess.call(['/usr/sbin/nginx', '-p', nginx_dir, '-c', 'nginx.conf'])
-                    else:
-                        _logger.warning('failed to start nginx - failed to kill orphan worker - oh well')
+                self._start_nginx()
+        self._check_main_nginx()
 
     def _fetch_loop_turn(self, host, pull_info_failures, default_sleep=1):
         with self._manage_host_exception(host) as manager: