Orchestrator Goes Open-Source

By Ganna Shargorodska

EPAM Cloud Orchestrator was created in 2012 as a cloud solution for the EPAM Systems community. It appeared when the issue of redesigning the growing infrastructure of more than 500 projects became critical. After a thorough analysis of the existing solutions, EPAM experts had to admit that none of them fit the company needs.

However, that was not a problem but a challenge. If no perfect solution existed, EPAM created its own from scratch. It took into account the specific requirements of EPAM development teams, including the most popular platform services used in project development.


In four years, EPAM Cloud Orchestrator evolved into a true hybrid cloud with over 5000 virtual machines in hundreds of projects. In addition to its basic function - hosting virtual resources - it supports a number of related components helping the users to be in full control of their virtual infrastructures, such as audit, monitoring, billing, reporting.

Now we are bringing the Orchestrator to a new level - we are introducing a new framework allowing the users, on one hand, to select the Orchestrator components they need and, on the other hand, to contribute to the development by adding their code. Yes, indeed, Orchestrator goes open-source. And not only that, it goes beyond EPAM.

The new Orchestrator, to be released under the name Maestro 3, will utilize Amazon resources in all its components. Here we are taking advantage of our long-term integration with Amazon during which we have had a great chance to try and test what Amazon has to offer and to find ways to adapt Amazon services to the Orchestrator requirements. At the same time, the infrastructure control will be performed by the native Orchestrator tools, as usual.

All Maestro 3 components are designed as standalone units which can be used separately, also for purposes outside the EPAM Cloud. Authentication is done via the Active Directory. The user can access AWS services directly using their domain credentials. In this case, the user is authenticated via the SAML protocol and allowed access within the pre-configured IAM role. This type of authentication requires no interaction with Maestro 3 whatsoever. Access to Maestro 3 is based on the user's domain credentials, as well, however, in this case the OAuth 2.0 protocol is used. Maestro 3 includes a mechanism allowing authentication for any enterprise with an Active Directory.

Deployment Framework

Remember, we said "open-source"? Open-source means that the product code is publicly shared and that anyone can contribute to it. The same approach will be applied in Maestro 3, and for that we have developed a special framework.


The new framework called the Deployment Framework will be fully based on Amazon using the native AWS tools and services.

In implementing its Deployment Framework, EPAM Cloud uses AWS services:

- Amazon Lambda - a serverless AWS service where custom code can be uploaded and executed using the Amazon infrastructure. Code is uploaded creating so-called Lambda functions which are triggered by certain events. Lambda functions are associated with AWS resources, for example, an Amazon DynamoDB table or an AWS S3 bucket, and changes to the associated resource trigger the Lambda function. The required computing resources are allocated and provided automatically.

- Amazon DynamoDB - a NoSQL database service where you can create tables to serve as the source of triggers for Lambda functions. With DynamoDB, you can create tables of any size and then easily scale them up or down without affecting the performance, as the capacities are provided automatically by internal AWS tools.

- Amazon S3 - a native Amazon cloud storage. As any storage, S3 buckets are used to store uploaded objects and, also being an AWS resource, can trigger Lambda functions by changes to their content.

- Amazon API Gateway - a native Amazon service where each user can easily and quickly create and manage own RESTful APIs. API Gateway can also trigger a Lambda function referring to the corresponding back-end service, such as Amazon DynamoDB, Amazon S3, etc.

- Amazon CloudWatch - an Amazon monitoring service collecting metrics from AWS resources and logging changes occurring to them. CloudWatch allows automatic responses to changes to AWS resources, including scheduled ones.

The Deployment Framework will be able to detect new code uploaded by the contributors in the form of Lambda functions and create the corresponding AWS resources. For example, if a Lambda function is associated with a DynamoDB table, the Deployment Framework will create such table which will then trigger the Lambda function.

By default, the Deployment Framework accepts code written in Java or Python, however, any other programming language may also be acceptable, if the contributor keeps to a certain set of rules, such as structure and file naming conventions, so that a .json configuration file recognizable by any language can be generated.

Code contributors can choose to integrate their code with the existing Maestro 3 functionality only for their purposes or to make it generally available. The Deployment Framework processes uploaded code according to the rules, and if contributed code does not match the rules, it is ignored.

Front-End


The front-end is written in Angular 2.0, an open-source JavaScript-based framework for web application development. Angular 2.0 was chosen as a flexible framework supporting component approach - if one of the components fails, this does not affect the rest of the system.

As the previous versions of EPAM Orchestrator were written in Angular 1.0, it was only logical to continue with Angular 2.0, as such approach could speed up the application development by reusing the already developed elements with certain adaptations. In addition, Angular 2.0 allows creating dynamic items, for example, by supporting the drag-and-drop functionality which will be featured in Maestro 3 UI.

The mobile Maestro 3 application written in Angular 2.0, can be used on Android, iOS and Windows without any special adaptations. What is also worth mentioning, is that about 70% of single-page applications currently on the market are written in Angular 1.0. Angular 2.0 supports backward compatibility, therefore, Maestro 3 components can be easily integrated with other web applications.

Finally, let's look briefly "under the hood" of the Maestro 3 user interface. In Maestro 3, the front-end part involves Amazon CloudFront, a content delivery network which receives the hyperlink from the browser and delivers the HTML stored in the special S3 storage. This HTML is a single-page application written in Angular 2.0 which represents the Orchestrator UI. When another request is made from the UI, only the affected segment of the page is updated, which increases the response and efficiency.

Billing

Maestro 3 supports AWS Consolidated Billing, in which you get a single bill for all your accounts. The billing data is obtained from AWS directly in the form of .csv files. The billing data is updated on the hourly basis, therefore, you can see the current cost of your infrastructure at any time.

Similarly to the EPAM Cloud, Maestro 3 receives bills from Amazon and sends them to the customers "as is", without adding or removing anything. However, there is an option of making adjustments to the bills, for example, extending credits to particular accounts for certain amounts.

The billing engine is based on the same concepts and mechanisms as were applied in EPAM Orchestrator. Virtual resources are billed according to the timelines - periods of time during which the resource is in a certain state. Timeline beginnings and endings are marked with audit events which in Maestro 3 are supplied by CADF (Cloud Auditing Data Federation). A virtual machine start generates an audit event as well as its stop generates another. The period between these events is a timeline which is billed in the same way for its entire duration. The same principle is applied to all other virtual resources, such as checkpoints or volumes.

Billing is designed as a serverless component, with different tasks performed by dedicated Lambdas.

Auto-Configuration

The Auto-Configuration functionality which is a great advantage of EPAM Orchestrator has also been recreated in Maestro 3. As with other components, auto-configuration was designed on the basis of AWS services to ensure smooth integration. However, the implementation has proved to be challenging and tricky.

In EPAM Orchestrator, auto-configuration is based on the Chef service which, obviously, means that it requires a Chef server. This did not fit into the serverless architecture concept applied in Maestro 3. Therefore, the Chef server functions (cookbook storage and search) were divided between Amazon S3 and DynamoDB.

The framework for cookbook development is based on the Dynamic Discovering principle allowing to discover existing services and use their components rather than creating them from scratch. Cookbooks can be created using the Chef Solo client and need no special adaptation. The completed cookbooks are then uploaded to an S3 bucket from where they can be retrieved.


As the database, the auto-configuration module uses DynamoDB, a native Amazon non-relational database which has shown optimal performance, quick response and convenient search possibilities.

For automated infrastructure creation Maestro 3 uses AWS CloudFormation service.

When?

OK, now the big question. When are we going to see it and try it? Maestro 3 is planned for release in the end of December, 2016. All components are to be ready and tested by then, and Maestro 3 will see the light as the first open-source product ever developed by EPAM Cloud. Stay tuned!