Repair / Reconfigure Sentinel Script

sentinel_reconfig.py

#!/usr/bin/python

# This was written to help with common sentinel misconfigurations.
# This script will loop through each of your sentinels nodes and properly reconfigure them to monitor the master node.

# Instructions
# 1. Modify the instance variable values to reflect your setup.
# 2. Run the script.

import sys
import os
import subprocess
import shlex
import time
import signal
import datetime

class SentinelRepair():

    def __init__(self):
        # List all redis node urls ip:port
        self.redis_hosts = ['10.0.96.66:6379', '10.0.96.67:6379']
        # List all sentinel node urls. ip:port
        self.sentinel_hosts = ['10.0.96.66:26379', '10.0.96.67:26379', '10.0.96.68:26379']
        # Your cluster name.
        self.cluster = "scale-testing"
        # Quorum value
        self.quorum=2
        # Your cluster password
        self.password = "your-password"

    def run_command(self, hostname, command):
        host, port = hostname.split(':', 2)
        command = "redis-cli -a {password} -h {host} -p {port} {command}".format(password=self.password, host=host, port=str(port), command=command)
        start = datetime.datetime.now()
        # Hacking in a timeout.  Allows anyone using earlier versions of Python to run this without having to backport subprocess.
        process = subprocess.Popen(shlex.split(command), stdout=subprocess.PIPE, stderr=subprocess.PIPE)
        while process.poll() is None:
            time.sleep(0.1)
            now = datetime.datetime.now()
            if (now - start).seconds > 5:
                os.kill(process.pid, signal.SIGKILL)
                os.waitpid(-1, os.WNOHANG)
                raise Exception("Timeout occured when connecting to {hostname}".format(hostname=hostname))
        return process.stdout.read()

    def is_master(self, hostname):
        result = self.run_command(hostname, "INFO replication").split('\r\n')
        return "role:master" in result

    def discover_master_hostname(self):
        for hostname in self.redis_hosts:
            if self.is_master(hostname):
                print "Discovered master node at {hostname}".format(hostname=hostname)
                return hostname
        sys.stderr.write('Unable to discover master node. Exiting.')
        sys.exit(1)

    def start(self):
        self.master_host, self.master_port = self.discover_master_hostname().split(':', 2)
        print "Reconfiguring sentinels"
        for hostname in self.sentinel_hosts:
            print "Repairing sentinel {hostname}".format(hostname=hostname)
            self.run_command(hostname, "SENTINEL REMOVE {cluster}".format(cluster=self.cluster))
            self.run_command(hostname, "SENTINEL MONITOR {cluster} {master_host} {master_port} {quorum}".format(cluster=self.cluster, master_host=self.master_host, master_port=self.master_port, quorum=self.quorum))
            # Add any other configurations your cluster relies on below. 
            self.run_command(hostname, "SENTINEL SET {cluster} auth-pass {password}".format(cluster=self.cluster, password=self.password))
            self.run_command(hostname, "SENTINEL FLUSHCONFIG")

if __name__ == '__main__':
    SentinelRepair().start()