From e9d7761e78b5b0e71460c879f7cfde9156ba49df Mon Sep 17 00:00:00 2001
From: Joshua Boniface <joshua@boniface.me>
Date: Sat, 29 Sep 2018 01:45:23 -0400
Subject: [PATCH] Add a dedicated primary key to keep master selection
 consistent

---
 router-daemon/pvcrd/Daemon.py         | 19 +++++++++++++++----
 router-daemon/pvcrd/RouterInstance.py | 17 ++++++++++++++---
 2 files changed, 29 insertions(+), 7 deletions(-)

diff --git a/router-daemon/pvcrd/Daemon.py b/router-daemon/pvcrd/Daemon.py
index ceaf12a5..04dca1ce 100644
--- a/router-daemon/pvcrd/Daemon.py
+++ b/router-daemon/pvcrd/Daemon.py
@@ -201,6 +201,14 @@ else:
     transaction.create('/routers/{}/ipmipassword'.format(myhostname), config['ipmi_password'].encode('ascii'))
     transaction.commit()
 
+# Check that the primary key exists, and create it with us as master if not
+current_primary = zkhandler.readdata(zk_conn, '/routers')
+if current_primary:
+    print('Current primary router is {}"{}"{}.'.format(ansiiprint.blue(), current_primary, ansiiprint.end()))
+else:
+    print('No primary router key found; creating with us as primary.')
+    zkhandler.writedata(zk_conn, { '/routers': myhostname })
+
 zkhandler.writedata(zk_conn, { '/routers/{}/daemonstate'.format(myhostname): 'init' })
 
 t_router = dict()
@@ -224,6 +232,13 @@ this_router = t_router[myhostname]
 update_zookeeper = this_router.update_zookeeper
 update_zookeeper()
 
+@zk_conn.DataWatch('/routers')
+def updateprimary(data, meta):
+    if data == myhostname:
+        this_router.set_primary()
+    else:
+        this_router.set_secondary()
+
 @zk_conn.ChildrenWatch('/networks')
 def updatenetworks(new_network_list):
     global network_list
@@ -239,10 +254,6 @@ def updatenetworks(new_network_list):
     network_list = new_network_list
     print(ansiiprint.blue() + 'Network list: ' + ansiiprint.end() + '{}'.format(' '.join(network_list)))
 
-# Ensure we force startup of interfaces if we're primary
-if this_router.getnetworkstate() == 'primary':
-    this_router.become_primary()
-
 # Create timer to update this router in Zookeeper
 def createKeepaliveTimer():
     interval = int(config['keepalive_interval'])
diff --git a/router-daemon/pvcrd/RouterInstance.py b/router-daemon/pvcrd/RouterInstance.py
index 037d2703..d4ef5cda 100644
--- a/router-daemon/pvcrd/RouterInstance.py
+++ b/router-daemon/pvcrd/RouterInstance.py
@@ -107,8 +107,13 @@ class RouterInstance():
             self.s_network[network].removeAddress()
 
     def set_secondary(self):
-        zkhandler.writedata(self.zk_conn, { '/routers/{}/networkstate'.format(self.name): 'secondary' })
-   
+        try:
+            zkhandler.writedata(self.zk_conn, {
+                '/routers/{}/networkstate'.format(self.name): 'secondary'
+            })
+        except:
+            time.sleep(1)
+
     def become_primary(self):
         ansiiprint.echo('Setting router {} to primary state.'.format(self.name), '', 'i')
         ansiiprint.echo('Network list: {}'.format(', '.join(self.network_list)), '', 'c')
@@ -120,7 +125,13 @@ class RouterInstance():
                     zkhandler.writedata(self.zk_conn, { '/routers/{}/networkstate'.format(self.t_router[router].getname()): 'secondary' })
 
     def set_primary(self):
-        zkhandler.writedata(self.zk_conn, { '/routers/{}/networkstate'.format(self.name): 'primary' })
+        try:
+            zkhandler.writedata(self.zk_conn, { 
+                '/routers/primary': self.name,
+                '/routers/{}/networkstate'.format(self.name): 'primary',
+            })
+        except:
+            time.sleep(1)
                 
 
     def update_zookeeper(self):