Posts Monitoring ec2 with clojure and Server-Stats
Post
Cancel

Monitoring ec2 with clojure and Server-Stats

Intro

Before going further, please take a look at Parallel SSH and system monitoring in Clojure.

Combining my need for monitoring an EC2 instance running nginx, bamboo and artifactory, with my will to code in clojure, I have decided to use server-stats to basic monitor my server via ssh. EC2 supports interactions using SSH without need its .pem file if you add your public key to it. Take a look here to see how.

These are dead simple commands and of course you should use more serious ones for critical services.

The server-stats config file is actually a clojure file named server-stats.cfg where you put some definitions for monitoring. The strength of server-stats is the running it against several server. But in this case, I have just one.

The config file

(import org.apache.commons.mail.SimpleEmail)


(defn send-mail [alert-msg server-name cmd-output]
    (doto (SimpleEmail.)
     ;; ... my email configs
      (.send)))

(add-alert-handler email [alert-msg server-name cmd-output]
  (send-mail alert-msg server-name cmd-output))

(set-ssh-username "ec2-user")

(add-server-group web-servers
  ["myserver.mydomain.com"])

(add-cmd disk
  {:doc "Get the disk usage using df"
   :cmd "df -h"
   :servers [web-servers]
   :alerts [{:column "Use%"
             :value-type percent
             :handlers [email]
             :msg "Disk space over 55% full"
             :trigger (> 55)}]})

(add-cmd is-nginx
  {:doc "Is nginx Running?"
   :servers [web-servers]
   :cmd "[[ -z `ps aux | grep nginx` ]] && echo 'false' || echo 'true'"
   :alerts [{:value-type bool
             :msg "Nginx is not running"
             :handlers [email]
             :trigger (= false)}]})

(add-cmd app-log
  {:doc "Get last 20 entries in todays app-servers log"
   :servers [web-servers]
   :cmd "tail -20 /home/ec2-user/artifactory-2.5.1.1/logs/access.log"})

(add-cmd http-errors
  {:doc "Show recent non-200s requests"
   :servers [web-servers]
   :cmd "tail -200 /var/log/nginx/error.log"})

(add-cmd is-artifactory
  {:doc "Is Artifactory Running?"
   :servers [web-servers]
   :cmd "[[ -z `ps aux | grep org.artifactory.standalone.main.Main` ]] && echo 'false' || echo 'true'"
   :alerts [{:value-type bool
             :msg "Artifactory is not running"
             :handlers [email]
             :trigger (= false)
             :mute-for 1860000}]})

(add-cmd is-bamboo
  {:doc "Is bamboo Running?"
   :servers [web-servers]
   :cmd "[[ -z `ps aux | grep com.attlassian.bamboo.server.Server` ]] && echo 'false' || echo 'true'"
   :alerts [{:value-type bool
             :msg "Bamboo is not running"
             :handlers [email]
             :trigger (= false)
             :mute-for 1860000}]})

Almost all the commands were borrowed from the sample cfg file. A small but important detail is the key :mute-for in alerts. As you can see, the above link to server-stats points to my fork of it.

The mute-for key adds the capability to prevent any scheduled ssh commands with short intervals to flood your alert communication channel (e-mail in this cfg). The alerts are sent only if the mute-for time has been elapsed.

This is done via empty files for controlling the alerts. server-stats checks the lastModified property of the file of a given alert, and activates the it only if the mute-for interval is over.

To make server-stats send alerts before the :mute-for time, you can just delete the file named .{Sanitized_alert_message}. The sanitized message is just the message alert message with no characters as underlines. I should consider a hash version of it to avoid big names for big messages, but it is ok for now.

Note that mute-for controls the alert activation per cmd message.

Use cron

Of course you can’t spend your time issuing ssh commands by hand. So, there are four cron entries for scheduled ssh interactions:

0 7 * * * SSH_AUTH_SOCK="$(find /tmp/keyring*/ -type s -user paulosuzart -group paulosuzart  -name 'ssh*' | head -n 1)" ~/workspace/p/Server-Stats/run.sh disk -log
*/10 * * * * SSH_AUTH_SOCK="$(find /tmp/keyring*/ -type s -user paulosuzart -group paulosuzart  -name 'ssh*' | head -n 1)" ~/workspace/p/Server-Stats/run.sh is-nginx -log
*/30 * * * * SSH_AUTH_SOCK="$(find /tmp/keyring*/ -type s -user paulosuzart -group paulosuzart  -name 'ssh*' | head -n 1)" ~/workspace/p/Server-Stats/run.sh is-artifactory -log
*/30 * * * * SSH_AUTH_SOCK="$(find /tmp/keyring*/ -type s -user paulosuzart -group paulosuzart  -name 'ssh*' | head -n 1)" ~/workspace/p/Server-Stats/run.sh is-bamboo -log

The run.sh is pretty simple. Just wraps java -jar to server-stats. Some times I want to log the ssh output, so I can pass the -log option and the output goes to log/cmd.

#!/bin/bash
# Wraps the start of server-stats. If -log is passed, will save
# the resulting execution to log/cmd
# paulosuzart@gmail.com

# java -jar should run from inside the server-stats dir
cd ~/workspace/p/Server-Stats

if [ -z $1 ]; then
  echo "usage: run cmd [-log]"
  exit 2
fi

COMM="java -jar server-stats-0.1-standalone.jar -a $1"
if [[ ! -z $2 && "-log"=$2 ]]; then
  echo "logging to log/$1.out"
  $COMM=$COMM:" > log/$1.out"
fi

echo $COMM
$COMM

Don’t ask me why, but ssh does not behave the way you except when called from a cron command. But I managed to find this article that helped a lot.

This crontable is configured on my laptop and any strange behavior is sent to my e-mail immediately. The other cmds (app-log, http-errors) are used directly with ./run.sh cmd to see the output.

Conclusion

Of course the mute-for feature might be merged to the original repo. So, I would recommend you to use the original version of server-stats.

If you need simple monitoring features or more elaborated ones, it is up to you. Just use server-stats as the base for it.

Update

This post was republished Dzone. See here.

This post is licensed under CC BY 4.0 by the author.