Corus vs Chef

At a high level, the differences between Corus and Chef (or Ansible, or any solution of the sort) might not be self-evident. This page aims at explaining such differences.

Comparing Chef (and other such solutions) with Corus is like comparing potatoes with poutine.

Chef and the likes are distributed installers. Corus is a distributed application manager. You install Corus with Chef. You deploy and manage your apps with Corus.

Corus' philosophy is about:
  • Favoring rapid intervention in production when issues appear;
  • avoiding redeployments when application configuration changes are required across instances;
  • acting in a fine-grained manner on live processes;
  • controlling application lifecycle in a non-intrusive way, from deployment to undeployment through execution, and potentially rollback;
  • etc., you get the deal.

Corus manages runtime

Corus has built-in support not only for deployment (and undeployment), but also for all the stages in between: executing processes, restarting (i.e.: bouncing) processes, killing processes, etc. Corus also supports auto-restarting unresponsive processes.

The above functionality requires NO special scripting on the part of developers and system administrators.

Corus deploys fast

Corus' deployment protocol is designed to be fast and minimize rollout time: upon deployment, Corus nodes implement a replication chain whereby bits are streamed from one node to the next, in a multithreaded way. There is no significant difference between deploying on 2 nodes vs 50.

Corus has built-in support for controlled degradation when deploying

Say you have 10 instances of a given application per Corus cluster in production. Upon deploying a new version, you want to maintain capacity while deployment is on-going. One solution is to use the ripple command, in conjunction with http check. The following command will deploy over a cluster, 2 nodes at a time, making sure the application's HTTP endpoint is returning a HTTP 200 before moving on to deploying on the next 2 nodes:

ripple -c "deploy -r -cluster | http check -c /myapp -p myapp.http -m 6 -t 10 -cluster" -b 2

Corus has built-in support for rollbacks

Corus has support for archiving it's current state upon deployment, so that rollbacks do not require redeploying the bits of the previous version. Rollbacks can litterally take seconds: just invoke the rollback command in the CLI, or use the corresponding call from Corus' REST API.

As an example, the following command-line will perform rollback on a cluster, 2 nodes at a time, on all nodes whose current version is 2.0. In addition, the command-line forces a check on the nodes where rollback occured, insuring that the "previous" version is up and running prior to moving on to the next 2 nodes:

match command "ripple -c 'rollback -d myapp -v 2.0 -cluster | http check -c /myapp -p myapp.http -m 6 -t 10 -cluster' -b 2 -m 2" 2.0* ls -cluster

Corus allows for live configuration changes at scale

Well-designed server-side apps should have provisions for high-risk features and conditions, allowing for the enablement/disablement of certain functionality though predefined configuration flags. In addition, well-designed apps should allow for parametrization by using variables as placeholders for configuration values.

Corus is in complete harmony with the approach above: it allows storing so-called properties that are passed to JVMs in the form of System properties. Performing live config changes can be as simple as the following 2 commands (and spare a redeployment simply for the sake of changing a few configuration parameters):

conf add -p guiding.system.enabled=false -cluster
ripple -c "restart all -cluster | http check -c /myapp -p myapp.http -m 6 -t 10 -cluster" -b 2

Corus has built-in port management

Corus supports the notion of port ranges: it allocates ports to processes and deallocates them upon process termination. This is done dynamically, and to Corus there is no difference between running 1 or 5 instances of an application on the same host

Corus has a rich command-line interface allowing for granular live ops intervention

In situations where fine-grained control is required, Corus' command-line interface provides a rich set of commands allowing for rapidly visualizing and acting on application processes and application packages, across a cluster, from a simple terminal:

ps -cluster 
count ps -cluster
match act* ps -cluster
count match act* ps -cluster 
count hosts 
ripple -c "restart all -cluster | http check -c /myapp -p myapp.http -m 6 -t 10 -cluster" -b 2
conf add -p -cluster

Corus has an exhaustive REST API

Corus has a rich REST API, allowing for full automation and customization. All primitives are there that allow integrating Corus functionality into very specific deployment procedures.