Merge "Add an init phase to scheduler/web startup"

web/src/containers/component/ComponentTable.jsx

@@ -30,6 +30,7 @@ import {
   RunningIcon,
   QuestionIcon,
   StopCircleIcon,
+  HistoryIcon,
 } from '@patternfly/react-icons'
 
 import { IconProperty } from '../build/Misc'
@@ -39,6 +40,10 @@ const STATE_ICON_CONFIGS = {
     icon: RunningIcon,
     color: 'var(--pf-global--success-color--100)',
   },
+  INITIALIZING: {
+    icon: HistoryIcon,
+    color: 'var(--pf-global--warning-color--100)',
+  },
   PAUSED: {
     icon: PauseCircleIcon,
     color: 'var(--pf-global--warning-color--100)',

zuul/driver/gerrit/gerritconnection.py

@@ -181,6 +181,11 @@ class GerritEventConnector(threading.Thread):
         return not self._stopped
 
     def run(self):
+        # Wait for the scheduler to prime its config so that we have
+        # the full tenant list before we start moving events.
+        self.connection.sched.primed_event.wait()
+        if self._stopped:
+            return
         self.event_queue.registerEventWatch(self._onNewEvent)
         while not self._stopped:
             try:

zuul/driver/github/githubconnection.py

@@ -693,6 +693,11 @@ class GithubEventConnector:
         return not self._stopped
 
     def run_event_dispatcher(self):
+        # Wait for the scheduler to prime its config so that we have
+        # the full tenant list before we start moving events.
+        self.connection.sched.primed_event.wait()
+        if self._stopped:
+            return
         self.event_queue.registerEventWatch(self._onNewEvent)
         # Set the wake event so we get an initial run
         self._dispatcher_wake_event.set()

zuul/driver/gitlab/gitlabconnection.py

@@ -91,6 +91,11 @@ class GitlabEventConnector(threading.Thread):
         return not self._stopped
 
     def run(self):
+        # Wait for the scheduler to prime its config so that we have
+        # the full tenant list before we start moving events.
+        self.connection.sched.primed_event.wait()
+        if self._stopped:
+            return
         self.event_queue.registerEventWatch(self._onNewEvent)
         while not self._stopped:
             try:

zuul/driver/pagure/pagureconnection.py

@@ -156,6 +156,11 @@ class PagureEventConnector(threading.Thread):
         return not self._stopped
 
     def run(self):
+        # Wait for the scheduler to prime its config so that we have
+        # the full tenant list before we start moving events.
+        self.connection.sched.primed_event.wait()
+        if self._stopped:
+            return
         self.event_queue.registerEventWatch(self._onNewEvent)
         while not self._stopped:
             try:

zuul/scheduler.py

@@ -150,6 +150,7 @@ class Scheduler(threading.Thread):
         threading.Thread.__init__(self)
         self.daemon = True
         self.hostname = socket.getfqdn()
+        self.primed_event = threading.Event()
         self.wake_event = threading.Event()
         self.layout_lock = threading.Lock()
         self.run_handler_lock = threading.Lock()
@@ -282,7 +283,7 @@ class Scheduler(threading.Thread):
             target=self.startCleanup, name='cleanup start')
         self.start_cleanup_thread.daemon = True
         self.start_cleanup_thread.start()
-        self.component_info.state = self.component_info.RUNNING
+        self.component_info.state = self.component_info.INITIALIZING
 
     def stop(self):
         self.log.debug("Stopping scheduler")
@@ -301,6 +302,8 @@ class Scheduler(threading.Thread):
         self.log.debug("Stopping nodepool")
         self.nodepool.stop()
         self.log.debug("Stopping connections")
+        # Connections may be waiting on the primed event
+        self.primed_event.set()
         self.stopConnections()
         self.log.debug("Stopping stats thread")
         self.stats_election.cancel()
@@ -844,7 +847,9 @@ class Scheduler(threading.Thread):
         duration = round(time.monotonic() - start, 3)
         self.log.info("Config priming complete (duration: %s seconds)",
                       duration)
+        self.primed_event.set()
         self.wake_event.set()
+        self.component_info.state = self.component_info.RUNNING
 
     def reconfigure(self, config, smart=False):
         self.log.debug("Submitting reconfiguration event")

zuul/web/__init__.py

@@ -1792,6 +1792,7 @@ class ZuulWeb(object):
     def start(self):
         self.log.debug("ZuulWeb starting")
 
+        self.component_info.state = self.component_info.INITIALIZING
         # Wait for system config and layouts to be loaded
         while not self.system_config_cache.is_valid:
             self.system_config_cache_wake_event.wait()

zuul/zk/components.py

@@ -44,6 +44,7 @@ class BaseComponent(ZooKeeperBase):
     """
 
     # Component states
+    INITIALIZING = "initializing"
     RUNNING = "running"
     PAUSED = "paused"
     STOPPED = "stopped"