Merge "Add zookeeper-timeout connection config"

doc/source/configuration.rst

@@ -162,6 +162,11 @@ Options
 
       The path to the PEM encoded CA certificate.
 
+.. attr:: zookeeper-timeout
+   :type: float
+   :default: 10.0
+
+   The ZooKeeper session timeout, in seconds.
 
 .. attr:: labels
    :type: list

nodepool/builder.py

@@ -1397,7 +1397,9 @@ class NodePoolBuilder(object):
                 list(self._config.zookeeper_servers.values()),
                 tls_cert=self._config.zookeeper_tls_cert,
                 tls_key=self._config.zookeeper_tls_key,
-                tls_ca=self._config.zookeeper_tls_ca)
+                tls_ca=self._config.zookeeper_tls_ca,
+                timeout=self._config.zookeeper_timeout,
+            )
 
             self.log.debug('Starting listener for build jobs')
 

nodepool/cmd/config_validator.py

@@ -81,6 +81,7 @@ class ConfigValidator:
                 'chroot': str,
             }],
             'zookeeper-tls': zk_tls,
+            'zookeeper-timeout': float,
             'providers': list,
             'labels': [label],
             'diskimages': [diskimage],

nodepool/config.py

@@ -38,6 +38,7 @@ class Config(ConfigValue):
         self.providers = {}
         self.provider_managers = {}
         self.zookeeper_servers = {}
+        self.zookeeper_timeout = 10.0
         self.zookeeper_tls_cert = None
         self.zookeeper_tls_key = None
         self.zookeeper_tls_ca = None
@@ -56,6 +57,7 @@ class Config(ConfigValue):
                     self.providers == other.providers and
                     self.provider_managers == other.provider_managers and
                     self.zookeeper_servers == other.zookeeper_servers and
+                    self.zookeeper_timeout == other.zookeeper_timeout and
                     self.elements_dir == other.elements_dir and
                     self.images_dir == other.images_dir and
                     self.build_log_dir == other.build_log_dir and
@@ -109,6 +111,9 @@ class Config(ConfigValue):
             name = z.host + '_' + str(z.port)
             self.zookeeper_servers[name] = z
 
+    def setZooKeeperTimeout(self, timeout):
+        self.zookeeper_timeout = float(timeout)
+
     def setDiskImages(self, diskimages_cfg):
         if not diskimages_cfg:
             return
@@ -368,6 +373,7 @@ def loadConfig(config_path, env=os.environ):
     newconfig.setMaxHoldAge(config.get('max-hold-age'))
     newconfig.setWebApp(config.get('webapp'))
     newconfig.setZooKeeperServers(config.get('zookeeper-servers'))
+    newconfig.setZooKeeperTimeout(config.get('zookeeper-timeout', 10.0))
     newconfig.setDiskImages(config.get('diskimages'))
     newconfig.setLabels(config.get('labels'))
     newconfig.setProviders(config.get('providers'))

nodepool/launcher.py

@@ -946,7 +946,9 @@ class NodePool(threading.Thread):
             self.zk.connect(configured,
                             tls_cert=config.zookeeper_tls_cert,
                             tls_key=config.zookeeper_tls_key,
-                            tls_ca=config.zookeeper_tls_ca)
+                            tls_ca=config.zookeeper_tls_ca,
+                            timeout=config.zookeeper_timeout,
+                            )
         else:
             self.log.debug("Detected ZooKeeper server changes")
             self.zk.resetHosts(configured)

nodepool/tests/fixtures/node.yaml

@@ -13,6 +13,8 @@ zookeeper-tls:
   cert: {zookeeper_cert}
   key: {zookeeper_key}
 
+zookeeper-timeout: 20.0
+
 labels:
   - name: fake-label
     min-ready: 1

nodepool/zk.py

@@ -1012,7 +1012,7 @@ class ZooKeeper(object):
         self._became_lost = False
 
     def connect(self, host_list, read_only=False, tls_cert=None,
-                tls_key=None, tls_ca=None):
+                tls_key=None, tls_ca=None, timeout=10.0):
         '''
         Establish a connection with ZooKeeper cluster.
 
@@ -1030,8 +1030,11 @@ class ZooKeeper(object):
         '''
         if self.client is None:
             hosts = buildZooKeeperHosts(host_list)
-            args = dict(hosts=hosts,
-                        read_only=read_only)
+            args = dict(
+                hosts=hosts,
+                read_only=read_only,
+                timeout=timeout,
+            )
 
             args['use_ssl'] = True
             if not (tls_key and tls_cert and tls_ca):

releasenotes/notes/zk-timeout-bede88920e248c96.yaml

@@ -0,0 +1,5 @@
+---
+features:
+  - |
+    It is now possible to configure the :attr:`zookeeper-timeout`. which can
+    help to avoid zookeeper session losses on busy systems.