Non-blocking beat

TODO.md

@@ -386,3 +386,6 @@ try: import uwsgi to know its running uwsgi
 # don't block on beat, and --report periodic tasks
 
 # Deprecate restart/start/stop services (do touch wsgi.py and fuck celery)
+
+
+# high perf apache sync, 

orchestra/bin/orchestra-beat

@@ -16,7 +16,7 @@ from datetime import datetime, timedelta
 
 from celery.schedules import crontab_parser as CrontabParser
 
-from orchestra.utils.sys import run, join
+from orchestra.utils.sys import run, join, LockFile
 
 
 class Setting(object):
@@ -126,13 +126,14 @@ def fire_pending_messages(settings, db):
 
 
 if __name__ == "__main__":
-    # TODO aquire lock
+    with LockFile('/dev/shm/beat.lock', expire=20):
         manage = sys.argv[1]
         procs = []
         settings = Setting(manage).get_settings()
         db = DB(settings)
         db.connect()
         try:
+            # Non-blocking loop, we need to finish this in time for the next minute.
             if 'orchestra.contrib.tasks' in settings['INSTALLED_APPS']:
                 if settings.get('TASKS_BACKEND', 'thread') in ('thread', 'process'):
                     for proc in fire_pending_tasks(manage, db):
@@ -140,13 +141,6 @@ if __name__ == "__main__":
             if 'orchestra.contrib.mailer' in settings['INSTALLED_APPS']:
                 for proc in fire_pending_messages(settings, db):
                     procs.append(proc)
-        exit_code = 0
-        for proc in procs:
-            result = join(proc)
-            sys.stdout.write(result.stdout.decode('utf8'))
-            sys.stderr.write(result.stderr.decode('utf8'))
-            if result.return_code != 0:
-                exit_code = result.return_code
         finally:
             db.close()
-    sys.exit(exit_code)
+    sys.exit(0)

orchestra/contrib/mailer/engine.py

@@ -4,6 +4,8 @@ from socket import error as SocketError
 from django.core.mail import get_connection
 from django.utils.encoding import smart_str
 
+from orchestra.utils.sys import LockFile
+
 from .models import Message
 
 
@@ -29,7 +31,7 @@ def send_message(message, num=0, connection=None, bulk=100):
 
 
 def send_pending(bulk=100):
-    # TODO aquire lock
+    with LockFile('/dev/shm/mailer.send_pending.lock'):
         connection = None
         num = 0
         for message in Message.objects.filter(state=Message.QUEUED).order_by('priority'):

orchestra/contrib/resources/tasks.py

@@ -1,14 +1,15 @@
 from orchestra.contrib.orchestration import Operation
 from orchestra.contrib.tasks import task
 from orchestra.models.utils import get_model_field_path
+from orchestra.utils.sys import LockFile
 
 from .backends import ServiceMonitor
 
 
 @task(name='resources.Monitor')
 def monitor(resource_id, ids=None, async=True):
+    with LockFile('/dev/shm/resources.monitor.lock', expire=60*60):
         from .models import ResourceData, Resource
-    
         resource = Resource.objects.get(pk=resource_id)
         resource_model = resource.content_type.model_class()
         logs = []

orchestra/contrib/webapps/backends/php.py

@@ -1,5 +1,6 @@
 import os
 import textwrap
+from collections import OrderedDict
 
 from django.template import Template, Context
 from django.utils.translation import ugettext_lazy as _
@@ -115,19 +116,27 @@ class PHPBackend(WebAppServiceMixin, ServiceController):
                 service php5-fpm reload
             fi
             # Coordinate apache restart with Apache2Backend
-            # FIXME race condition
-            locked=1
-            state="$(grep -v 'PHPBackend' /dev/shm/restart.apache2)" || locked=0
-            echo -n "$state" > /dev/shm/restart.apache2
+            restart=0
+            backend='PHPBackend'
+            mv /dev/shm/restart.apache2 /dev/shm/restart.apache2.locked || {
+                sleep 0.1
+                mv /dev/shm/restart.apache2 /dev/shm/restart.apache2.locked
+            }
+            state="$(grep -v $backend /dev/shm/restart.apache2.locked)" || restart=1
+            echo -n "$state" > /dev/shm/restart.apache2.locked
             if [[ $UPDATED_APACHE -eq 1 ]]; then
-                if [[ $locked == 0 ]]; then
+                if [[ $restart == 1 ]]; then
                     service apache2 status && service apache2 reload || service apache2 start
+                    rm /dev/shm/restart.apache2.locked
                 else
-                    echo "PHPBackend RESTART" >> /dev/shm/restart.apache2
+                    echo "$backend RESTART" >> /dev/shm/restart.apache2.locked
+                    mv /dev/shm/restart.apache2.locked /dev/shm/restart.apache2
                 fi
             elif [[ "$state" =~ .*RESTART$ ]]; then
-                rm /dev/shm/restart.apache2
+                rm /dev/shm/restart.apache2.locked
                 service apache2 status && service apache2 reload || service apache2 start
+            else
+                mv /dev/shm/restart.apache2.locked /dev/shm/restart.apache2
             fi
             """)
         )
@@ -135,7 +144,6 @@ class PHPBackend(WebAppServiceMixin, ServiceController):
     
     def get_options(self, webapp):
         kwargs = {}
-        print(webapp.data)
         if self.MERGE:
             kwargs = {
                 'webapp__account': webapp.account,
@@ -195,10 +203,10 @@ class PHPBackend(WebAppServiceMixin, ServiceController):
     
     def get_fcgid_cmd_options(self, webapp, context):
         options = self.get_options(webapp)
-        maps = {
-            'MaxProcesses': options.get('processes', None),
-            'IOTimeout': options.get('timeout', None),
-        }
+        maps = OrderedDict(
+            MaxProcesses=options.get('processes', None),
+            IOTimeout=options.get('timeout', None),
+        )
         cmd_options = []
         for directive, value in maps.items():
             if value:

orchestra/contrib/websites/backends/apache.py

@@ -137,18 +137,28 @@ class Apache2Backend(ServiceController):
     def commit(self):
         """ reload Apache2 if necessary """
         self.append(textwrap.dedent("""\
-            locked=1
-            state="$(grep -v 'Apache2Backend' /dev/shm/restart.apache2)" || locked=0
-            echo -n "$state" > /dev/shm/restart.apache2
-            if [[ $UPDATED == 1 ]]; then
-                if [[ $locked == 0 ]]; then
+            # Coordinate apache restart with Apache2Backend
+            restart=0
+            backend='Apache2Backend'
+            mv /dev/shm/restart.apache2 /dev/shm/restart.apache2.locked || {
+                sleep 0.1
+                mv /dev/shm/restart.apache2 /dev/shm/restart.apache2.locked
+            }
+            state="$(grep -v $backend /dev/shm/restart.apache2.locked)" || restart=1
+            echo -n "$state" > /dev/shm/restart.apache2.locked
+            if [[ $UPDATED_APACHE -eq 1 ]]; then
+                if [[ $restart == 1 ]]; then
                     service apache2 status && service apache2 reload || service apache2 start
+                    rm /dev/shm/restart.apache2.locked
                 else
-                    echo "Apache2Backend RESTART" >> /dev/shm/restart.apache2
+                    echo "$backend RESTART" >> /dev/shm/restart.apache2.locked
+                    mv /dev/shm/restart.apache2.locked /dev/shm/restart.apache2
                 fi
             elif [[ "$state" =~ .*RESTART$ ]]; then
-                rm /dev/shm/restart.apache2
+                rm /dev/shm/restart.apache2.locked
                 service apache2 status && service apache2 reload || service apache2 start
+            else
+                mv /dev/shm/restart.apache2.locked /dev/shm/restart.apache2
             fi""")
         )
         super(Apache2Backend, self).commit()

orchestra/utils/sys.py

@@ -156,3 +156,35 @@ def get_default_celeryd_username():
         if user is None:
             raise CommandError("Can not find the default celeryd username")
     return user
+
+
+class LockFile(object):
+    """ File-based lock mechanism used for preventing concurrency problems """
+    def __init__(self, lockfile, expire=5*60, unlocked=False):
+        """ /dev/shm/ can be a good place for storing locks ;) """
+        self.lockfile = lockfile
+        self.expire = expire
+        self.unlocked = unlocked
+    
+    def acquire(self):
+        if os.path.exists(self.lockfile):
+            lock_time = os.path.getmtime(self.lockfile)
+            # lock expires to avoid starvation
+            if time.time()-lock_time < self.expire:
+                return False
+        touch(self.lockfile)
+        return True
+    
+    def release(self):
+        os.remove(self.lockfile)
+    
+    def __enter__(self):
+        if not self.unlocked:
+            if not self.acquire():
+                raise OperationLocked('%s lock file exists and its mtime is less '
+                    'than %s seconds' % (self.lockfile, self.expire))
+        return True
+    
+    def __exit__(self, type, value, traceback):
+        if not self.unlocked:
+            self.release()