Add some configuration sanity checking at startup

zoe_api/entrypoint.py

@@ -34,6 +34,13 @@ log = logging.getLogger("zoe_api")
 LOG_FORMAT = '%(asctime)-15s %(levelname)s %(threadName)s->%(name)s: %(message)s'
 
 
+def _check_configuration_sanity():
+    if config.get_conf().auth_type == 'ldap' and not zoe_api.auth.ldap.LDAP_AVAILABLE:
+        log.error("LDAP authentication requested, but 'pyldap' module not installed.")
+        return 1
+    return 0
+
+
 def zoe_web_main() -> int:
     """
     This is the entry point for the Zoe Web script.
@@ -46,9 +53,9 @@ def zoe_web_main() -> int:
     else:
         logging.basicConfig(level=logging.INFO, format=LOG_FORMAT)
 
-    if config.get_conf().auth_type == 'ldap' and not zoe_api.auth.ldap.LDAP_AVAILABLE:
-        log.error("LDAP authentication requested, but 'pyldap' module not installed.")
-        return 1
+    ret = _check_configuration_sanity()
+    if ret != 0:
+        return ret
 
     zoe_api.db_init.init()
 

zoe_master/backends/common.py

@@ -48,7 +48,7 @@ def gen_environment(service: Service, execution: Execution):
 
 
 def _create_logs_directories(exec_id, service_name):
-    path = get_conf().logs_base_path + '/' + get_conf().deployment_name + str(exec_id) + '/' + service_name
+    path = get_conf().logs_base_path + '/' + get_conf().deployment_name + '/' + str(exec_id) + '/' + service_name
     try:
         os.makedirs(path)
     except OSError as e:

zoe_master/backends/interface.py

@@ -23,8 +23,10 @@ from zoe_lib.state import Execution, Service
 
 from zoe_master.backends.base import BaseBackend
 from zoe_master.backends.old_swarm.backend import OldSwarmBackend
+import zoe_master.backends.old_swarm.api_client
 from zoe_master.backends.old_swarm_new_api.backend import OldSwarmNewAPIBackend
-from zoe_master.exceptions import ZoeStartExecutionFatalException, ZoeStartExecutionRetryException
+import zoe_master.backends.old_swarm_new_api.api_client
+from zoe_master.exceptions import ZoeStartExecutionFatalException, ZoeStartExecutionRetryException, ZoeException
 
 log = logging.getLogger(__name__)
 
@@ -33,8 +35,12 @@ def _get_backend() -> BaseBackend:
     """Return the right backend instance by reading the global configuration."""
     backend_name = get_conf().backend
     if backend_name == 'OldSwarm':
+        if not zoe_master.backends.old_swarm.api_client.AVAILABLE:
+            raise ZoeException('The OldSwarm backend requires docker-py version <= 1.10.2')
         return OldSwarmBackend(get_conf())
     elif backend_name == 'OldSwarmNewAPI':
+        if not zoe_master.backends.old_swarm_new_api.api_client.AVAILABLE:
+            raise ZoeException('The OldSwarmNewAPI backend requires docker python version >= 2.0.2')
         return OldSwarmNewAPIBackend(get_conf())
     else:
         log.error('Unknown backend selected')

zoe_master/backends/old_swarm/api_client.py

@@ -32,12 +32,17 @@ try:
 except ImportError:
     KazooClient = None
 
+AVAILABLE = True
 try:
     import docker
     import docker.errors
     import docker.utils
 except ImportError:
-    pass
+    AVAILABLE = False
+try:
+    docker.Client()
+except:
+    AVAILABLE = False
 
 import requests.packages
 
@@ -148,9 +153,13 @@ class SwarmClient:
         self.opts = opts
         url = opts.swarm
         if 'zk://' in url:
+            if KazooClient is None:
+                raise ZoeLibException('ZooKeeper URL for Swarm, but the kazoo package is not installed')
             url = url[len('zk://'):]
             manager = zookeeper_swarm(url, opts.backend_swarm_zk_path)
         elif 'consul://' in url:
+            if Consul is None:
+                raise ZoeLibException('Consul URL for Swarm, but the consul package is not installed')
             url = url[len('consul://'):]
             manager = consul_swarm(url)
         elif 'http://' or 'https://' in url:

zoe_master/backends/old_swarm_new_api/api_client.py

@@ -32,12 +32,18 @@ try:
 except ImportError:
     KazooClient = None
 
+AVAILABLE = True
 try:
     import docker
     import docker.errors
     import docker.utils
 except ImportError:
-    pass
+    AVAILABLE = False
+else:
+    try:
+        docker.DockerClient()
+    except:
+        AVAILABLE = False
 
 import requests.packages
 
@@ -290,6 +296,9 @@ class SwarmClient:
         elif container.status == 'OOMKilled' or container.status == 'exited':
             info["state"] = "killed"
             info["running"] = False
+        elif container.status == 'created':
+            info["state"] = 'created'
+            info["running"] = False
         else:
             log.error('Unknown container status: {}'.format(container.status))
             info["state"] = "unknown"

zoe_master/backends/old_swarm_new_api/backend.py

@@ -20,11 +20,11 @@ import logging
 from zoe_lib.config import get_conf
 from zoe_lib.exceptions import ZoeLibException, ZoeNotEnoughResourcesException
 from zoe_lib.state import Execution, Service
-from zoe_master.backends.old_swarm.api_client import DockerContainerOptions, SwarmClient
+from zoe_master.backends.old_swarm_new_api.api_client import DockerContainerOptions, SwarmClient
 from zoe_master.exceptions import ZoeStartExecutionRetryException, ZoeStartExecutionFatalException, ZoeException
 import zoe_master.backends.common
 import zoe_master.backends.base
-from zoe_master.backends.old_swarm.threads import SwarmMonitor, SwarmStateSynchronizer
+from zoe_master.backends.old_swarm_new_api.threads import SwarmMonitor, SwarmStateSynchronizer
 from zoe_master.stats import NodeStats, ClusterStats  # pylint: disable=unused-import
 
 log = logging.getLogger(__name__)

zoe_master/backends/old_swarm_new_api/threads.py

@@ -21,7 +21,7 @@ import time
 
 from zoe_lib.config import get_conf
 from zoe_lib.state import SQLManager, Service
-from zoe_master.backends.old_swarm.api_client import SwarmClient
+from zoe_master.backends.old_swarm_new_api.api_client import SwarmClient
 
 log = logging.getLogger(__name__)
 

zoe_master/entrypoint.py

@@ -18,6 +18,7 @@
 """Zoe Master main entrypoint."""
 
 import logging
+import os
 
 import zoe_lib.config as config
 from zoe_lib.metrics.influxdb import InfluxDBMetricSender
@@ -28,11 +29,19 @@ import zoe_master.scheduler
 import zoe_master.backends.interface
 from zoe_master.preprocessing import restart_resubmit_scheduler
 from zoe_master.master_api import APIManager
+from zoe_master.exceptions import ZoeException
 
 log = logging.getLogger("main")
 LOG_FORMAT = '%(asctime)-15s %(levelname)s %(threadName)s->%(name)s: %(message)s'
 
 
+def _check_configuration_sanity():
+    if not os.access(config.get_conf().logs_base_path, os.W_OK):
+        log.error('Logs directory {} is not writable'.format(config.get_conf().logs_base_path))
+        return 1
+    return 0
+
+
 def main():
     """
     The entrypoint for the zoe-master script.
@@ -42,10 +51,13 @@ def main():
     args = config.get_conf()
     if args.debug:
         logging.basicConfig(level=logging.DEBUG, format=LOG_FORMAT)
-
     else:
         logging.basicConfig(level=logging.INFO, format=LOG_FORMAT)
 
+    ret = _check_configuration_sanity()
+    if ret != 0:
+        return ret
+
     if config.get_conf().influxdb_enable:
         metrics = InfluxDBMetricSender(config.get_conf().deployment_name, config.get_conf().influxdb_url, config.get_conf().influxdb_dbname)
     else:
@@ -57,7 +69,11 @@ def main():
     log.info("Initializing scheduler")
     scheduler = getattr(zoe_master.scheduler, config.get_conf().scheduler_class)(state, config.get_conf().scheduler_policy)
 
-    zoe_master.backends.interface.initialize_backend(state)
+    try:
+        zoe_master.backends.interface.initialize_backend(state)
+    except ZoeException as e:
+        log.error('Cannot initialize backend: {}'.format(e.message))
+        return 1
 
     restart_resubmit_scheduler(state, scheduler)