Merge pull request #88 from dcos/ken/agent-fault-injection

restarting of nodes and agent or master processes

shakedown/dcos/agent.py

@@ -4,6 +4,7 @@
 import os
 from dcos import (marathon, mesos)
 
+
 def get_public_agents():
     """Provides a list of hostnames / IPs that are public agents in the cluster"""
     agent_list = []
@@ -15,6 +16,7 @@ def get_public_agents():
 
     return agent_list
 
+
 def get_private_agents():
     """Provides a list of hostnames / IPs that are private agents in the cluster"""
 
@@ -34,6 +36,7 @@ def get_private_agents():
 
     return agent_list
 
+
 def get_agents():
     """Provides a list of hostnames / IPs of all agents in the cluster"""
 
@@ -118,6 +121,7 @@ def restart_agent(
 
     run_command_on_agent(hostname, "sudo systemctl restart dcos-mesos-slave")
 
+
 def stop_agent(
     hostname
 ):
@@ -140,6 +144,13 @@ def start_agent(
     run_command_on_agent(hostname, "sudo systemctl start dcos-mesos-slave")
 
 
+def restart_agent_node(hostname):
+    """ Restarts the agent node
+    """
+
+    run_command_on_agent(hostname, "sudo /sbin/shutdown -r now")
+
+
 def delete_agent_log(
     hostname
 ):

shakedown/dcos/master.py

@@ -37,6 +37,19 @@ def reconnect_master():
     reconnect_node(shakedown.master_ip())
 
 
+def restart_master_node():
+    """ Restarts the master node
+    """
+
+    run_command_on_master("sudo /sbin/shutdown -r now")
+
+
+def systemctl_master(command='restart'):
+    """ Used to start, stop or restart the master process
+    """
+    run_command_on_master('sudo systemctl {} dcos-mesos-master'.format(command))
+
+
 @contextlib.contextmanager
 def disconnected_master(incoming=True, outgoing=True):