Monitoring a Server with the Telegram API

Nov 03, 2015 by Cesare Rocchi

Some servers behind Podrover are just workers and do not run a web application, so I can’t use Pingdom and the like to monitor them.

I could use some library like Monit but honestly I didn’t want to install one more thing on the server and I like to exploit push notifications when possible.

I am already monitoring some behavior of my DB servers via Slack. When some key activity is performed a message is sent to a private Slack group, and I get a notification on my iPhone. I didn’t want to clutter that monitoring channel with more notifications about workers, plus I wanted to learn something new so I gave Telegram a shot. It turned out to be pretty easy.

I created a bot, then I created a new group (conversation) to which I added the bot and myself. A script on my servers tells the bot to post a message in that group. That’s it. To post to a group you need its id, which is a little tricky to find it. You need to fetch updates via the getUpdates API, like this.

curl -s \
  -X POST \
  https://api.telegram.org/bot<YOUR_BOT_TOKEN>/getUpdates

This returns something like

{
  "result": [
    {
	  ...
      "chat": {
        "username": "funkyboy",
        "first_name": "Cesare",
        "id": 123 // <- this ID
      },
	  ...
   ]
}

That id field within the chat object is the one you need. Now you can post messages to that specific group:

curl -s \
  -X POST \
  https://api.telegram.org/bot<YOUR_BOT_TOKEN>/sendMessage \
  -d text="This is a message from the bot" \
  -d chat_id=123

You can schedule such a script on the server via Cron. On my servers I use a Rake task like this

task :dude_i_am_still_alive do
  require 'socket'
  require 'httparty'
  host = Socket.gethostname
  body = {:text => "#{host} is alive", :chat_id => 123}
  response = HTTParty.post('https://api.telegram.org/bot<YOUR_BOT_TOKEN>/sendMessage', :body => body)
end

scheduled via whenever.

It’s still not ideal to me, because essentially I should worry when I am NOT getting a notification but for now it will do. After all a dead machine can not tell you that it’s dead :)

If you are using a different tool to solve this problem, let me know on Twitter.