Monitoring Resource Usage of Jobs with cAdvisor

Introduction

Google's cAdvisor (https://github.com/google/cadvisor) provides information about the resources used by containers. A web UI is exposed at http://hostname:port/, and in addition data is exported to a central database. For sites running a batch system with cgroups enabled, cAdvisor can provide information about running jobs on worker nodes.

The main page of the UI shows an overview of CPU, memory, network and disk usage of the whole node.

You can then drill down and view information about individual jobs.

Clicking on one of the listed jobs gives information about the resource usage of that job.

The web UI is of limited use as it only shows data over the past minute. Using Grafana for visualizing data collected by cAdvisor and stored in InfluxDB is more interesting.

Installing InfluxDB

Download and install the rpm:

wget https://s3.amazonaws.com/influxdb/influxdb-latest-1.x86_64.rpm
rpm -ivh influxdb-latest-1.x86_64.rpm

then start the service

service influxdb start

In a browser go to http://hostname:8083/ and login using the default username (root) and password (root). To create a database for cAdvisor, specify a database name in the 'Database Details' part of 'Create a Database' and click 'Create Database'.

You should see something like the following:

Once the database has been created, click on the database name and create a user by specifying the username and password in the 'Create a New Database User' section.

More information available at http://influxdb.com

Installing Grafana

Download and install the rpm:

rpm -ivh https://grafanarel.s3.amazonaws.com/builds/grafana-2.0.2-1.x86_64.rpm

and start the service

service grafana-server start

Running cAdvisor

Example usage on a HTCondor worker node

/usr/local/bin/cadvisor -storage_driver=influxdb -storage_driver_host=hostname:8086 -storage_driver_db=database_name \ 
-storage_driver_password=password -storage_driver_user=user -storage_driver_secure=false -storage_driver_table=stats

where the InfluxDB hostname, database name, username and password should be changed as appropriate.