Vigilant provides application driven stats monitoring. When you integrate your application with the Daemon bindings/library every time your application starts a daemon is created and all other processes will attach to this daemon sending watch/log/alert messages which in turn are delivered to a datastore. Because your own applications know their pid (os.getpid()) you no longer need to manage your monitoring with runner scripts.
Dynamic cluster model visualized using vis.js.
Real-time graphs using the web-socket api using vis.js.
Detailed Setup tutorial can be found here https://github.com/redbrain/vigilant/wiki/QuickSetup
You will require Open/Oracle JDK >= 1.7, Python >= 3.4, Bower.
Mac OSX
# download orcale jdk.
$ brew install python3 npm
$ npm install -g bower
Ubuntu
$ sudo apt-get install default-jdk python3.4 nodejs
$ npm install -g bower
Currently daemon2 is a WIP and not ready but Daemon is proof of concept. It requires python >= 3.4
$ cd daemon
$ sudo pip3 install -r requirements.txt
$ ./daemon.py -c etc/vigilant/vigilant.json --start
Editing the vigilant.json declares where data is sent and the protocol. Currently only udp is supported by the datastore. And i aim to keep using UDP as the main protocol. And use ack's for alerts/triggers from code to ensure they are sent.
Using --stop or --status will stop the daemon or show status of what the agent is watching and sending the data to respectively.
Watch another process:
$ ps aux | grep -i spotify | grep -v grep
redbrain 54243 0.0 0.7 3465888 59412 ?? S 4:32pm 4:06.33 /Applications/Spotify.app/Contents/MacOS/Spotify
$ ./daemon.py -c etc/vigilant/vigilant.json --watch spotify:54243
If the process dies vigilant will stop watching the process automatically.
Once an agent is running the data needs to be recieved. The datastore will accept all the data and provide functionality over it. Written in Scala requires jdk >= 1.7.
$ cd datastore
$ bower install
$ cd etc/vigilant
$ export VIGILANT_HOME=`pwd`
$ cd -
$ ./sbt
> compile
> test
> container:start
Currently deploying the .war onto jetty or tomcat runner the websocket api doesnt work. Editing the vigilant configuration:
{
"transport": {
"type": "udp",
"host": "localhost",
"port": 8080
},
"triggers": {
"notification_threshold": 120 // How often should notifications be send if data continues to activate triggers. To stop notification spam.
},
"database": {
"jdbc": "jdbc:h2:./vigilant" // FIXME
},
"twillo": {
"account_sid": "", # twillo details to enable twillo notifications
"auth_token": "",
"from": ""
},
"email": {
"smtp_server": "localhost", # email details doesnt handle tls/ssl like gmail.com
"from": "someone@email.com"
}
}
View swagger api documentation: http://localhost:8080/api and use /api-doc as the location to the documenation.
The front-end webapp is a seperate project abstracting datastores.
$ sudo pip3 install -r requirements.txt
$ bower install
$ ./dashboard.py
Go to http://localhost:5000/#/dashboard?store=http://localhost:8080
The proof-of-concept daemon, this is being re-written in C/C++ to increase portability and simplicity of language bindings. Currently because this is written in Python3 Node bindings require python3 to be installed which isn't very elegant.
This is a WIP new Daemon written in C/C++ to increase the protability and simplicity of adding simple dependancies.
The scala data-store listens for the real-time data and in turn provides a rest-api for working with this. The api is fully documented at http://localhost:8080/api using swagger. The web-socket api isn't supported by swagger but it is there.
This is the current front-end it needs more work but its working quite well for now. It simply uses the data-store rest-api to access the monitoring data.