Image and VM management for Jenkins, Spark and Mesos clusters in EC2
Switch branches/tags
Clone or download
hannes-ucsc Merge pull request #232 from BD2KGenomics/issues/11-expose-spark-work…

Have Spark worker advertise public IP for web app (connected to #11)
Latest commit 939e7ad Feb 6, 2018

CGCloud lets you automate the creation, management and provisioning of VMs and clusters of VMs in Amazon EC2. While allowing for easy programmatic customization of VMs in development, it also provides rock-solid reproducibility in production.


  • Works with base images of all actively supported releases of Ubuntu and Fedora, and some releases of CentOS

  • Lets you share VMs between multiple users, keeping the set of authorized SSH keys synchronized on all VMs in real-time as users/keypairs are added or removed from AWS.

  • Offers isolation between users, teams and deployments via namespaces

  • Lets you stand up a distributed, continuous integration infrastructure using one long-running Jenkins master and multiple on-demand Jenkins slaves

  • Lets you create an HDFS-backed Apache Spark cluster of any number of nodes in just three minutes, independently of the number of nodes, with our without attached EBS volumes

  • Lets you create a Mesos cluster of any number of Nodes

  • Supports running Spark, Mesos and Toil workers on the spot market

  • Is easily extensible via a simple plugin architecture

  • VMs created by CGCloud optionally report memory and disk utilization as custom CloudWatch metrics

So what does it not offer? What are its limitations? First and foremost, it is strictly tied to AWS and EC2. Other cloud providers are not supported and probably will not be in the near future. It does not have a GUI. It is written in Python and if you want to customize it, you will need to know Python. It makes extreme use of multiple inheritance. Some people frown at that since it will make it likely that your own customizations break between releases of CGCloud. While allowing CGCloud to be extremely DRY, multiple inheritance also increases the complexity and steepens the learning curve.

Where to go from here?

If you are a (potential) user of CGCloud, head on over to the CGCloud Core README and then move on to

If you are a developer, make sure you have pip and virtualenv, clone this repository and perform the following steps from the project root::

virtualenv venv
source venv/bin/activate
make develop sdist

That will set up the project in development mode inside a virtualenv and create source distributions (aka sdists) for those components that are be installed on remote boxes. In development mode, these components are not installed from PyPI but are instead directly uploaded to the box in sdist form and then installed from the sdist.

After pulling changes from the remote, you need to run make develop sdist again. This step is easy to forget because you often get by without it.

Specifically, make develop is necessary after any of the or files have changed, and make sdist is necessary after changes to the agent, spark-tools or mesos-tools subprojects. Otherwise, cgcloud create will install a stale version of these on the remote box.

To run the unittests, pip install pytest and then do make test.