cluster-join

cluster-join.py

#!/usr/bin/python
#
# Copyright (c) 2015 Juniper Networks, Inc. All rights reserved.
#
"""Contrail Cluster Join Operations."""

# Example: python join.py --zk_index 2 --new_node 1.1.1.5 --roles database
#          python join.py --my_zk_index 2 --my_ip 1.1.1.5
#                         --other_nodes_ip 1.1.1.4 --other_nodes_name c1 --other_nodes_zk_indexes 1
#          python join.py --my_zk_index 3 --my_ip 1.1.1.6
#                         --other_nodes_ip 1.1.1.4,1.1.1.5 --other_nodes_name c1,c2 --other_nodes_zk_indexes 1,2
#                         --erlang_cookie 5ec9eedc-1c75-4ea0-9932-8ac8e0fd08e0

import os
import sys
import datetime
import shutil
import argparse
from distutils.dir_util import copy_tree

from fabric.api import local
from fabric.context_managers import settings

CASSANDRA_DIR = '/var/lib/cassandra'
ZK_DIR = '/var/lib/zookeeper'
RABBITMQ_DIR = '/var/lib/rabbitmq'

class ContrailClusterJoin(object):
    def __init__(self, args_str):
        self._parse_args(args_str)
        self._validate_args()
    # end __init__
# end class ContrailClusterJoin

class JoinRequest(ContrailClusterJoin):
    def _parse_args(self, args_str):
        '''
            Base parser.
        '''

        # Source any specified config/ini file
        # Turn off help, so we print all options in response to -h
        conf_parser = argparse.ArgumentParser(add_help=False)

        conf_parser.add_argument("-c", "--conf_file",
                                 help="Specify config file", metavar="FILE")
        args, remaining_argv = conf_parser.parse_known_args(args_str.split())

        defaults = {}
        if args.conf_file:
            config = ConfigParser.SafeConfigParser()
            config.read([args.conf_file])
            defaults.update(dict(config.items("DEFAULTS")))

        # Override with CLI options
        # Don't surpress add_help here so it will handle -h
        parser = argparse.ArgumentParser(
            # Inherit options from config_parser
            parents=[conf_parser],
            # print script description with -h/--help
            description=__doc__,
            # Don't mess with format of description
            formatter_class=argparse.RawDescriptionHelpFormatter,
        )
        parser.set_defaults(**defaults)
        parser.add_argument("-e", "--erlang_cookie",
            help="Erlang cookie of rabbitmq cluster")
        parser.add_argument("-i", "--my_zk_index",
            help="Index of this node in cluster (for zookeeper conf)")
        parser.add_argument("-a", "--my_ip",
            help="IP address of this node in cluster")
        parser.add_argument("-p", "--other_nodes_ip",
            help="Comma separated IP addresses of other nodes in cluster")
        parser.add_argument("-m", "--other_nodes_name",
            help="Comma separated hostnames of other nodes in cluster")
        parser.add_argument("-z", "--other_nodes_zk_indexes",
            help="Comma separated zookeeper indexes of other nodes in cluster")
        self._args_obj, remaining_argv = parser.parse_known_args(remaining_argv)
    # end _parse_args

    def _validate_args(self):
        # csv to list
        other_nodes_ip = self._args_obj.other_nodes_ip.split(',')
        other_nodes_name = self._args_obj.other_nodes_name.split(',')
        other_nodes_zk_indexes = self._args_obj.other_nodes_zk_indexes.split(',')
        if os.path.exists(ZK_DIR):
            if len(set(other_nodes_ip)) != len(set(other_nodes_zk_indexes)):
                raise Exception("Error: Unequal list lengths for other_nodes_ip vs their zookeeper indexes.")
            if not self._args_obj.my_zk_index:
                raise Exception("Error: my_zk_index has to be specified.")
            if not self._args_obj.my_ip:
                raise Exception("Error: my_ip has to be specified.")

        if os.path.exists(CASSANDRA_DIR):
            if not other_nodes_ip:
                raise Exception("Error: other_nodes_ip needs to be specified.")

        if os.path.exists(RABBITMQ_DIR):
            if len(set(other_nodes_ip)) != len(set(other_nodes_name)):
                raise Exception("Error: Unequal list lengths for other_nodes_ip vs other_nodes_name")
            if not self._args_obj.erlang_cookie:
                raise Exception("Error: erlang_cookie needs to be specified.")
            if not other_nodes_name:
                raise Exception("Error: other_nodes_name needs to be specified.")
    # end _validate_args

    def request_cluster_join(self):
        saved_suffix = str(datetime.datetime.now()).replace(' ','_')

        # cassandra join
        if os.path.exists(CASSANDRA_DIR):
            local('service contrail-database stop')
            # create a directory for backup
            saved_dirname = 'cassandra_%s' %(saved_suffix)
            saved_dirpath = '%s/%s' %('/'.join(CASSANDRA_DIR.split('/')[:-1]),
                                      saved_dirname)
            local('mkdir %s' %(saved_dirpath))
            # backup config file
            local("cp /etc/cassandra/cassandra.yaml %s/" %(saved_dirpath))
            # edit config file
            quoted_list = '"%s"' %(self._args_obj.other_nodes_ip)
            local("sed -i 's/seeds: .*$/seeds: %s/' /etc/cassandra/cassandra.yaml" %(quoted_list))
            # backup data directory
            local('mv %s/* %s/' %(CASSANDRA_DIR, saved_dirpath))
            local('service contrail-database start')

        # zookeeper join
        if os.path.exists(ZK_DIR):
            # csv to list
            other_nodes_ip = self._args_obj.other_nodes_ip.split(',')
            other_nodes_zk_indexes = self._args_obj.other_nodes_zk_indexes.split(',')
            with settings(warn_only=True):
                local('service zookeeper stop')
            # create a directory for backup
            saved_dirname = 'zookeeper_%s' %(saved_suffix)
            saved_dirpath = '%s/%s' %('/'.join(ZK_DIR.split('/')[:-1]),
                                      saved_dirname)
            local('mkdir %s' %(saved_dirpath))
            # backup config file
            local("cp -R /etc/zookeeper/* %s/" %(saved_dirpath))
            # edit config file
            local("echo %s > /etc/zookeeper/conf/myid" %(self._args_obj.my_zk_index))
            # clobber all server entries and generate new list
            local("sed -i 's/^server.*$//' /etc/zookeeper/conf/zoo.cfg")
            local("echo server.%s=%s:2888:3888 >> /etc/zookeeper/conf/zoo.cfg"
                  %(self._args_obj.my_zk_index, self._args_obj.my_ip))
            for i in range(len(other_nodes_ip)):
                ip = other_nodes_ip[i]
                idx = other_nodes_zk_indexes[i]
                local("echo server.%s=%s:2888:3888 >> /etc/zookeeper/conf/zoo.cfg"
                      %(idx, ip))
            # backup data directory
            #local('cp -R %s/* %s/' %(ZK_DIR, saved_dirpath))
            local('mv %s/* %s/' %(ZK_DIR, saved_dirpath))
            local('ln -s /etc/zookeeper/conf/myid %s/myid' %(ZK_DIR))
            local('service zookeeper start')

        # rabbitmq join
        if os.path.exists(RABBITMQ_DIR):
            # csv to list
            other_nodes_ip = self._args_obj.other_nodes_ip.split(',')
            other_nodes_name = self._args_obj.other_nodes_name.split(',')
            with settings(warn_only=True):
                local('rabbitmqctl stop_app')
            # create a directory for backup
            saved_dirname = 'rabbitmq_%s' %(saved_suffix)
            saved_dirpath = '%s/%s' %('/'.join(RABBITMQ_DIR.split('/')[:-1]),
                                      saved_dirname)
            local('mkdir %s' %(saved_dirpath))
            # backup data directory
            local('cp /var/lib/rabbitmq/.erlang.cookie %s/' %(saved_dirpath))
            local('echo "%s" > /var/lib/rabbitmq/.erlang.cookie' %(self._args_obj.erlang_cookie))

            # create hosts entry if not resolvable
            for i in range(len(other_nodes_ip)):
                ip = other_nodes_ip[i]
                name = other_nodes_name[i]
                with settings(warn_only=True):
                    out = local("ping -c 1 %s" %(name), capture=True)
                    if not out.succeeded:
                        local("echo %s %s %s-ctrl >> /etc/hosts"
                              %(ip, name, name))

            local('rabbitmqctl join_cluster %s@%s' %(other_nodes_name[0],
                                                     other_nodes_name[0]))
            local('rabbitmqctl start_app')
    # end request_cluster_join
# class JoinRequest


class JoinAdmit(ContrailClusterJoin):
    def _parse_args(self, args_str):
        '''
            Base parser.
        '''

        # Source any specified config/ini file
        # Turn off help, so we print all options in response to -h
        conf_parser = argparse.ArgumentParser(add_help=False)

        conf_parser.add_argument("-c", "--conf_file",
                                 help="Specify config file", metavar="FILE")
        args, remaining_argv = conf_parser.parse_known_args(args_str.split())

        defaults = {}
        if args.conf_file:
            config = ConfigParser.SafeConfigParser()
            config.read([args.conf_file])
            defaults.update(dict(config.items("DEFAULTS")))

        # Override with CLI options
        # Don't surpress add_help here so it will handle -h
        parser = argparse.ArgumentParser(
            # Inherit options from config_parser
            parents=[conf_parser],
            # print script description with -h/--help
            description=__doc__,
            # Don't mess with format of description
            formatter_class=argparse.RawDescriptionHelpFormatter,
        )
        parser.set_defaults(**defaults)
        parser.add_argument("-i", "--zk_index",
            help="Zookeeper index of new node joining cluster")
        parser.add_argument("-n", "--new_node",
            help="IP address of new node joining cluster")
        parser.add_argument("-r", "--roles",
            help="Comma separated roles of new node joining cluster")
        self._args_obj, remaining_argv = parser.parse_known_args(remaining_argv)
    # end _parse_args

    def _validate_args(self):
        if not self._args_obj.new_node or not self._args_obj.roles:
            raise Exception("Error: New node ip or roles has not been specified.")

        if os.path.exists(ZK_DIR):
            if 'database' in roles and not self._args_obj.zk_index:
                raise Exception("Error: New node has database role but zookeeper index not specified.")
    # end _validate_args

    def admit_to_cluster(self):
        # csv to list
        roles = self._args_obj.roles.split(',')

        if os.path.exists(ZK_DIR):
            with settings(warn_only=True):
                local('service zookeeper stop')
            # create a directory for backup
            saved_dirname = 'zookeeper_%s' %(str(datetime.datetime.now()).replace(' ','_'))
            saved_dirpath = '%s/%s' %('/'.join(ZK_DIR.split('/')[:-1]),
                                      saved_dirname)
            local('mkdir %s' %(saved_dirpath))
            # backup config file
            local("cp -R /etc/zookeeper/* %s/" %(saved_dirpath))
            # edit config file
            local("echo server.%s=%s:2888:3888 >> /etc/zookeeper/conf/zoo.cfg"
                  %(self._args_obj.zk_index,
                    self._args_obj.new_node))
            # backup data directory
            local('cp -R %s/* %s/' %(ZK_DIR, saved_dirpath))
            #local('mv %s/* %s/' %(ZK_DIR, saved_dirpath))
            local('service zookeeper start')
    # end admit_to_cluster
# class JoinAdmit


def cluster_join_request(args_str=None):
    jro = JoinRequest(args_str)
    jro.request_cluster_join()
# end cluster_join_request

def cluster_join_admit(args_str=None):
    jao = JoinAdmit(args_str)
    jao.admit_to_cluster()
# end cluster_join_admit

if __name__ == "__main__":
    cluster_join_request(' '.join(sys.argv[1:]))