csit/libraries/ClusterStateLibrary.py

   1 import SettingsLibrary
   2 from time import sleep
   3 import UtilLibrary
   4 import json
   5 import sys
   6
   7
   8 __author__ = "Basheeruddin Ahmed"
   9 __copyright__ = "Copyright(c) 2014, Cisco Systems, Inc."
  10 __license__ = "New-style BSD"
  11 __email__ = "syedbahm@cisco.com"
  12
  13
  14 def getClusterRoles(
  15     shardName, numOfShards=3, numOfTries=3, sleepBetweenRetriesInSecs=3, port=8181, *ips
  16 ):
  17     """Given a shardname (e.g. shard-inventory-config), number of shards and bunch of ips
  18
  19     determines what role each ip has in an Akka (Raft based) cluster
  20     result would look like
  21     {'10.194.126.118':'Leader', '10.194.126.118':'Follower', '10.194.126.117': None}
  22     """
  23     dict = {}
  24     for ip in ips:
  25         i = 1
  26         dict[ip] = None
  27         print("numOfShards => ", str(numOfShards))
  28         while i <= numOfShards:
  29             shardMemberName = "member-" + str(i) + "-" + shardName
  30             j = 1
  31             print("j => ", str(j))
  32             print("numOfTries => ", str(numOfTries))
  33             while int(j) <= int(numOfTries):
  34                 print("Try number " + str(j))
  35                 try:
  36                     print(
  37                         "getting role of " + ip + "  for shardName = " + shardMemberName
  38                     )
  39                     url = SettingsLibrary.getJolokiaURL(
  40                         ip, str(port), str(i), shardName
  41                     )
  42                     print(url)
  43                     resp = UtilLibrary.get(url)
  44                     print(resp)
  45                     if resp.status_code != 200:
  46                         sleep(sleepBetweenRetriesInSecs)
  47                         continue
  48                     print(resp.text)
  49                     data = json.loads(resp.text)
  50                     if "value" in data:
  51                         dataValue = data["value"]
  52                         print("datavalue RaftState is", dataValue["RaftState"])
  53                         dict[ip] = dataValue["RaftState"]
  54                 except Exception:
  55                     e = sys.exc_info()[0]
  56                     print(
  57                         "Try"
  58                         + str(j)
  59                         + ":An error occurred when finding leader on"
  60                         + ip
  61                         + " for shardName:"
  62                         + shardMemberName
  63                     )
  64                     print(e)
  65                     sleep(sleepBetweenRetriesInSecs)
  66                     continue
  67                 finally:
  68                     j = j + 1
  69             if dict[ip] is not None:
  70                 break
  71             i = i + 1
  72     return dict
  73
  74
  75 def isRole(
  76     role, shardName, ipAddress, numOfShards=3, numOfRetries=1, sleepFor=3, port=8181
  77 ):
  78     """Given a role (Leader, Follower, Candidate, or IsolatedLeader),
  79     shardname (e.g. shard-inventory-config), controller IP address,
  80     and number of shards on the controller,this function determines if the controller,
  81     has that role for the specified shard.
  82     """
  83     ip = getClusterRoles(
  84         shardName, numOfShards, numOfRetries, sleepFor, port, ipAddress
  85     )
  86     print(ip)
  87     if ip[ipAddress] == role:
  88         return True
  89     return False
  90
  91
  92 def getLeader(
  93     shardName, numOfShards=3, numOfTries=3, sleepBetweenRetriesInSecs=1, port=8181, *ips
  94 ):
  95     """Returns the leader of the shard given a set of IPs Or None"""
  96     for i in range(3):  # Try 3 times to find a leader
  97         dict = getClusterRoles(
  98             shardName, numOfShards, numOfTries, sleepBetweenRetriesInSecs, port, *ips
  99         )
 100         for ip in dict.keys():
 101             if dict[ip] == "Leader":
 102                 return ip
 103     return None
 104
 105
 106 def getFollowers(
 107     shardName, numOfShards=3, numOfTries=3, sleepBetweenRetriesInSecs=1, port=8181, *ips
 108 ):
 109     """Returns the follower list of a shard given a set of IPs Or []"""
 110     for i in range(6):  # Try 6 times to find all followers
 111         dict = getClusterRoles(
 112             shardName, numOfShards, numOfTries, sleepBetweenRetriesInSecs, port, *ips
 113         )
 114         result = []
 115
 116         for ip in dict.keys():
 117             if dict[ip] == "Follower":
 118                 result.append(ip)
 119         print("i=%s result=%s" % (i, result))
 120         if len(result) == (len(ips) - 1):
 121             break
 122         sleep(1)
 123     return result
 124
 125
 126 def testGetClusterRoles():
 127     dict = getClusterRoles(
 128         "shard-inventory-config",
 129         3,
 130         1,
 131         1,
 132         8181,
 133         "10.194.126.116",
 134         "10.194.126.117",
 135         "10.194.126.118",
 136     )
 137     print(dict)
 138
 139     for ip in dict.keys():
 140         if isRole("Leader", "shard-inventory-config", 3, 1, 1, 8181, ip):
 141             print(ip + " is Leader")
 142         elif isRole("Follower", "shard-inventory-config", 3, 1, 1, 8181, ip):
 143             print(ip + " is follower")
 144         else:
 145             print(ip + " seems to have value " + str(dict[ip]))
 146
 147
 148 def testGetLeader():
 149     leader = getLeader(
 150         "shard-inventory-config",
 151         3,
 152         1,
 153         1,
 154         8181,
 155         "10.194.126.116",
 156         "10.194.126.117",
 157         "10.194.126.118",
 158     )
 159     print(leader)
 160     return leader
 161
 162
 163 def testGetFollowers():
 164     followers = getFollowers(
 165         "shard-inventory-config",
 166         3,
 167         1,
 168         1,
 169         8181,
 170         "10.194.126.116",
 171         "10.194.126.117",
 172         "10.194.126.118",
 173     )
 174     print(followers)
 175     return followers
 176
 177
 178 # testGetClusterRoles()
 179 # testGetLeader()
 180 # testGetFollowers()