Automatic process monitoring/management with Python

Question

Right, so I have a python process which is running constantly, maybe even on Supervisor. What is the best way to achieve the following monitoring?

• Send an alert and restart if the process has crashed. I'd like to automatically receive a signal every time the process crashes and auto restart it.
• Send an alert and restart if the process has gone stale, ie hasn't crunched anything for say 1 minute.
• Restart on demand

I'd like the achieve all of the above through Python. I know Supervisord will do most of it, but I want to see if it can be done through Python itself.

Answer 1

I think what you are looking for is, Supervisor Events. http://supervisord.org/events.html

Also look at Superlance, its a package of plugin utilities for monitoring and controlling processes that run under supervisor. [ https://superlance.readthedocs.org/en/latest/]

You can configure stuff like Crash emails, Crash SMS, Memory consumption alerts, HTTP hooks etc.

Answer 2

Well, if you want a homegrown solution, this is what I could come up with.

Maintain the process state both actual and expected in redis. You can monitor it the way you want by making a web interface to check the actual state and change the expected state.

Run the python script in crontab to check for state and take appropriate action when required. Here I have checked for every 3 seconds and used SES to alert admins via email.

DISCLAIMER: The code has not been run or tested. I just wrote it now, so prone to errors.

open crontab file:

$crontab -e

Add this line at the end of it, to make the run_process.sh run every minute.

#Runs this process every 1 minute.
*/1 * * * * bash ~/path/to/run_monitor.sh

run_moniter.sh runs the python script. It runs in a for loop every 3 second.

This is done because crontab gives minimum time interval of 1 minute. We want to check for the process every 3 second, 20 times (3sec * 20 = 1 minute). So it will run for a minute before crontab runs it again.

run_monitor.sh

for count in {0..20}
do
    cd '/path/to/check_status'
    /usr/local/bin/python check_status.py "myprocessname" "python startcommand.py"
    sleep 3 #check every 3 seconds.
done

Here I have assumed:

*state 0 = stop or stopped (expected vs. actual)

*state -1 = restart

*state 1 = run or running

You can add more states as per your convinience, stale process can also be a state.

I have used processname to kill or start or check processes, you can easily modify it to read specific PID files.

check_status.py

import sys
import redis
import subprocess

import sys
import boto.ses


def send_mail(recipients, message_subject, message_body):
    """
    uses AWS SES to send mail.
    """
    SENDER_MAIL = 'xxx@yyy.com'
    AWS_KEY = 'xxxxxxxxxxxxxxxxxxx'
    AWS_SECRET = 'xxxxxxxxxxxxxxxxxxx'
    AWS_REGION = 'xx-xxxx-x'

    mail_conn = boto.ses.connect_to_region(AWS_REGION, 
                                           aws_access_key_id=AWS_KEY, 
                                           aws_secret_access_key=AWS_SECRET
                                           )

    mail_conn.send_email(SENDER_MAIL, message_subject, message_body, recipient, format='html')
    return True

class Shell(object):
    '''
    Convinient Wrapper over Subprocess.
    '''
    def __init__(self, command, raise_on_error=True):
        self.command = command
        self.output = None
        self.error = None
        self.return_code

    def run(self):
        try:
            process = subprocess.Popen(self.command, shell=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
            self.return_code = process.wait()
            self.output, self.error = process.communicate()
            if self.return_code and self.raise_on_error:
                print self.error
                raise Exception("Error while executing %s::%s"%(self.command, self.error))    
        except subprocess.CalledProcessError:
            print self.error
            raise Exception("Error while executing %s::%s"%(self.command, self.error))


redis_client = redis.Redis('xxxredis_hostxxx')

def get_state(process_name, state_type): #state_type will be expected or actual.
    state = redis.get('{process_name}_{state_type}_state'.format(process_name=process_name, state_type=state_type)) #value could be 0 or 1
    return state

def set_state(process_name, state_type, state): #state_type will be expected or actual.
    state = redis.set('{process_name}_{state_type}_state'.format(process_name=process_name, state_type=state_type), state)
    return state

def get_stale_state(process_name):
    state = redis.get('{process_name}_stale_state'.format(process_name=process_name)) #value could be 0 or 1
    return state

def check_running_status(process_name):
    command = "ps -ef|grep {process_name}|wc -l".format(process_name=process_name)
    shell = Shell(command = command)
    shell.run()
    if shell.output=='0':
        return False
    return True

def start_process(start_command): #pass start_command with a '&' so the process starts in the background.
    shell = Shell(command = command)
    shell.run()

def stop_process(process_name):
    command = "ps -ef| grep {process_name}| awk '{print $2}'".format(process_name=process_name)
    shell = Shell(command = command, raise_on_error=False)
    shell.run()
    if not shell.output:
        return
    process_ids = shell.output.strip().split()
    for process_id in process_ids:
        command = 'kill {process_id}'.format(process_id=process_id)
        shell = Shell(command=command, raise_on_error=False)
        shel.run()


def check_process(process_name, start_command):
    expected_state = get_state(process_name, 'expected')
    if expected_state == 0: #stop
        stop_process(process_name)
        set_state(process_name, 'actual', 0)

    else if expected_state == -1: #restart
        stop_process(process_name)
        set_state(process_name, 'actual', 0)
        start_process(start_command)
        set_state(process_name, 'actual', 1)
        set_state(process_name, 'expected', 1) #set expected back to 1 so we dont keep on restarting.

    elif expected_state == 1:
        running = check_running_status(process_name)
        if not running:
            set_state(process_name, 'actual', 0)
            send_mail(reciepients=["abc@admin.com", "xyz@admin.com"], message_subject="Alert", message_body="Your process is Down. Trying to restart")
            start_process(start_command)
            running = check_running_status(process_name)
            if running:
                send_mail(reciepients=["abc@admin.com", "xyz@admin.com"], message_subject="Alert", message_body="Your process is was restarted.")
                set_state(process_name, 'actual', 1)
            else:
                send_mail(reciepients=["abc@admin.com", "xyz@admin.com"], message_subject="Alert", message_body="Your process is could not be restarted.")


if __name__ == '__main__':
    args = sys.argv[1:]
    process_name = args[0]
    start_command = args[1]
    check_process(process_name, start_command)