AWS Compute Blog

Juan Villa – Partner Solutions Architect

When most people hear “Java” they think of Java the programming language. Java is a lot more than a programming language, it also implies a larger ecosystem including the Java Virtual Machine (JVM). Java, the programming language, is just one of the many languages that can be compiled to run on the JVM. Some of the most popular JVM languages, other than Java, are Clojure, Groovy, Scala, Kotlin, JRuby, and Jython (see this link for a list of more JVM languages).

Did you know that you can compile and subsequently run all these languages on AWS Lambda?

AWS Lambda supports the Java 8 runtime, but this does not mean you are limited to the Java language. The Java 8 runtime is capable of running JVM languages such as Kotlin and Groovy once they have been compiled and packaged as a “fat” JAR (a JAR file containing all necessary dependencies and classes bundled in).

In this blog post we’ll work through building AWS Lambda functions in both Kotlin and Groovy programming languages. To compile and package our projects we will use Gradle build tool.

(more…)

 
Stephen Liedig, Solutions Architect

One of the many challenges professional software architects and developers face is how to make cloud-native applications scalable, fault-tolerant, and highly available.

Fundamental to your project success is understanding the importance of making systems highly cohesive and loosely coupled. That means considering the multi-dimensional facets of system coupling to support the distributed nature of the applications that you are building for the cloud.

By that, I mean addressing not only the application-level coupling (managing incoming and outgoing dependencies), but also considering the impacts of of platform, spatial, and temporal coupling of your systems. Platform coupling relates to the interoperability (or lack thereof) of heterogeneous systems components. Spatial coupling deals with managing components at a network topology level or protocol level. Temporal, or runtime coupling, refers to the ability of a component within your system to do any kind of meaningful work while it is performing a synchronous, blocking operation.

The AWS messaging services, Amazon SQS and Amazon SNS, help you deal with these forms of coupling by providing mechanisms for:

• Reliable, durable, and fault-tolerant delivery of messages between application components
• Logical decomposition of systems and increased autonomy of components
• Creating unidirectional, non-blocking operations, temporarily decoupling system components at runtime
• Decreasing the dependencies that components have on each other through standard communication and network channels

Following on the recent topic, Building Scalable Applications and Microservices: Adding Messaging to Your Toolbox, in this post, I look at some of the ways you can introduce SQS and SNS into your architectures to decouple your components, and show how you can implement them using C#.

Walkthrough

To illustrate some of these concepts, consider a web application that processes customer orders. As good architects and developers, you have followed best practices and made your application scalable and highly available. Your solution included implementing load balancing, dynamic scaling across multiple Availability Zones, and persisting orders in a Multi-AZ Amazon RDS database instance, as in the following diagram.

In this example, the application is responsible for handling and persisting the order data, as well as dealing with increases in traffic for popular items.

One potential point of vulnerability in the order processing workflow is in saving the order in the database. The business expects that every order has been persisted into the database. However, any potential deadlock, race condition, or network issue could cause the persistence of the order to fail. Then, the order is lost with no recourse to restore the order.

With good logging capability, you may be able to identify when an error occurred and which customer order failed. This wouldn’t allow you to “restore” the transaction, and by that stage, your customer is no longer your customer.

As illustrated in the following diagram, introducing an SQS queue helps improve your ordering application. Using the queue isolates the processing logic into its own component and runs it in a separate process from the web application. This, in turn, allows the system to be more resilient to spikes in traffic, while allowing work to be performed only as fast as necessary in order to manage costs.

In addition, you now have a mechanism for persisting orders as messages (with the queue acting as a temporary database), and have moved the scope of your transaction with your database further down the stack. In the event of an application exception or transaction failure, this ensures that the order processing can be retired or redirected to the Amazon SQS Dead Letter Queue (DLQ), for re-processing at a later stage. (See the recent post, Using Amazon SQS Dead-Letter Queues to Control Message Failure, for more information on dead-letter queues.)

Scaling the order processing nodes

This change allows you now to scale the web application frontend independently from the processing nodes. The frontend application can continue to scale based on metrics such as CPU usage, or the number of requests hitting the load balancer. Processing nodes can scale based on the number of orders in the queue. Here is an example of scale-in and scale-out alarms that you would associate with the scaling policy.

Scale-out Alarm

aws cloudwatch put-metric-alarm --alarm-name AddCapacityToCustomerOrderQueue --metric-name ApproximateNumberOfMessagesVisible --namespace "AWS/SQS" 
--statistic Average --period 300 --threshold 3 --comparison-operator GreaterThanOrEqualToThreshold --dimensions Name=QueueName,Value=customer-orders
--evaluation-periods 2 --alarm-actions <arn of the scale-out autoscaling policy>

aws cloudwatch put-metric-alarm --alarm-name RemoveCapacityFromCustomerOrderQueue --metric-name ApproximateNumberOfMessagesVisible --namespace "AWS/SQS" 
 --statistic Average --period 300 --threshold 1 --comparison-operator LessThanOrEqualToThreshold --dimensions Name=QueueName,Value=customer-orders
 --evaluation-periods 2 --alarm-actions <arn of the scale-in autoscaling policy>

Scaling the order processing implementation

On top of scaling at an infrastructure level using Auto Scaling, make sure to take advantage of the processing power of your Amazon EC2 instances by using as many of the available threads as possible. There are several ways to implement this. In this post, we build a Windows service that uses the BackgroundWorker class to process the messages from the queue.

Here’s a closer look at the implementation. In the first section of the consuming application, use a loop to continually poll the queue for new messages, and construct a ReceiveMessageRequest variable.

public static void PollQueue()
{
    while (_running)
    {
        Task<ReceiveMessageResponse> receiveMessageResponse;

        // Pull messages off the queue
        using (var sqs = new AmazonSQSClient())
        {
            const int maxMessages = 10;  // 1-10

            //Receiving a message
            var receiveMessageRequest = new ReceiveMessageRequest
            {
                // Get URL from Configuration
                QueueUrl = _queueUrl, 
                // The maximum number of messages to return. 
                // Fewer messages might be returned. 
                MaxNumberOfMessages = maxMessages, 
                // A list of attributes that need to be returned with message.
                AttributeNames = new List<string> { "All" },
                // Enable long polling. 
                // Time to wait for message to arrive on queue.
                WaitTimeSeconds = 5 
            };

            receiveMessageResponse = sqs.ReceiveMessageAsync(receiveMessageRequest);
        }

The WaitTimeSeconds property of the ReceiveMessageRequest specifies the duration (in seconds) that the call waits for a message to arrive in the queue before returning a response to the calling application. There are a few benefits to using long polling:

• It reduces the number of empty responses by allowing SQS to wait until a message is available in the queue before sending a response.
• It eliminates false empty responses by querying all (rather than a limited number) of the servers.
• It returns messages as soon any message becomes available.

For more information, see Amazon SQS Long Polling.

After you have returned messages from the queue, you can start to process them by looping through each message in the response and invoking a new BackgroundWorker thread.

// Process messages
if (receiveMessageResponse.Result.Messages != null)
{
    foreach (var message in receiveMessageResponse.Result.Messages)
    {
        Console.WriteLine("Received SQS message, starting worker thread");

        // Create background worker to process message
        BackgroundWorker worker = new BackgroundWorker();
        worker.DoWork += (obj, e) => ProcessMessage(message);
        worker.RunWorkerAsync();
    }
}
else
{
    Console.WriteLine("No messages on queue");
}

The event handler, ProcessMessage, is where you implement business logic for processing orders. It is important to have a good understanding of how long a typical transaction takes so you can set a message VisibilityTimeout that is long enough to complete your operation. If order processing takes longer than the specified timeout period, the message becomes visible on the queue. Other nodes may pick it and process the same order twice, leading to unintended consequences.

Handling Duplicate Messages

In order to manage duplicate messages, seek to make your processing application idempotent. In mathematics, idempotent describes a function that produces the same result if it is applied to itself:

f(x) = f(f(x))

No matter how many times you process the same message, the end result is the same (definition from Enterprise Integration Patterns: Designing, Building, and Deploying Messaging Solutions, Hohpe and Wolf, 2004).

There are several strategies you could apply to achieve this:

• Create messages that have inherent idempotent characteristics. That is, they are non-transactional in nature and are unique at a specified point in time. Rather than saying “place new order for Customer A,” which adds a duplicate order to the customer, use “place order <orderid> on <timestamp> for Customer A,” which creates a single order no matter how often it is persisted.
• Deliver your messages via an Amazon SQS FIFO queue, which provides the benefits of message sequencing, but also mechanisms for content-based deduplication. You can deduplicate using the MessageDeduplicationId property on the SendMessage request or by enabling content-based deduplication on the queue, which generates a hash for MessageDeduplicationId, based on the content of the message, not the attributes.

var sendMessageRequest = new SendMessageRequest
{
    QueueUrl = _queueUrl,
    MessageBody = JsonConvert.SerializeObject(order),
    MessageGroupId = Guid.NewGuid().ToString("N"),
    MessageDeduplicationId = Guid.NewGuid().ToString("N")
};

• If using SQS FIFO queues is not an option, keep a message log of all messages attributes processed for a specified period of time, as an alternative to message deduplication on the receiving end. Verifying the existence of the message in the log before processing the message adds additional computational overhead to your processing. This can be minimized through low latency persistence solutions such as Amazon DynamoDB. Bear in mind that this solution is dependent on the successful, distributed transaction of the message and the message log.

Handling exceptions

Because of the distributed nature of SQS queues, it does not automatically delete the message. Therefore, you must explicitly delete the message from the queue after processing it, using the message ReceiptHandle property (see the following code example).

However, if at any stage you have an exception, avoid handling it as you normally would. The intention is to make sure that the message ends back on the queue, so that you can gracefully deal with intermittent failures. Instead, log the exception to capture diagnostic information, and swallow it.

By not explicitly deleting the message from the queue, you can take advantage of the VisibilityTimeout behavior described earlier. Gracefully handle the message processing failure and make the unprocessed message available to other nodes to process.

In the event that subsequent retries fail, SQS automatically moves the message to the configured DLQ after the configured number of receives has been reached. You can further investigate why the order process failed. Most importantly, the order has not been lost, and your customer is still your customer.

private static void ProcessMessage(Message message)
{
    using (var sqs = new AmazonSQSClient())
    {
        try
        {
            Console.WriteLine("Processing message id: {0}", message.MessageId);

            // Implement messaging processing here
            // Ensure no downstream resource contention (parallel processing)
            // <your order processing logic in here…>
            Console.WriteLine("{0} Thread {1}: {2}", DateTime.Now.ToString("s"), Thread.CurrentThread.ManagedThreadId, message.MessageId);
            
            // Delete the message off the queue. 
            // Receipt handle is the identifier you must provide 
            // when deleting the message.
            var deleteRequest = new DeleteMessageRequest(_queueName, message.ReceiptHandle);
            sqs.DeleteMessageAsync(deleteRequest);
            Console.WriteLine("Processed message id: {0}", message.MessageId);

        }
        catch (Exception ex)
        {
            // Do nothing.
            // Swallow exception, message will return to the queue when 
            // visibility timeout has been exceeded.
            Console.WriteLine("Could not process message due to error. Exception: {0}", ex.Message);
        }
    }
}

Using SQS to adapt to changing business requirements

One of the benefits of introducing a message queue is that you can accommodate new business requirements without dramatically affecting your application.

If, for example, the business decided that all orders placed over $5000 are to be handled as a priority, you could introduce a new “priority order” queue. The way the orders are processed does not change. The only significant change to the processing application is to ensure that messages from the “priority order” queue are processed before the “standard order” queue.

The following diagram shows how this logic could be isolated in an “order dispatcher,” whose only purpose is to route order messages to the appropriate queue based on whether the order exceeds $5000. Nothing on the web application or the processing nodes changes other than the target queue to which the order is sent. The rates at which orders are processed can be achieved by modifying the poll rates and scalability settings that I have already discussed.

Extending the design pattern with Amazon SNS

Amazon SNS supports reliable publish-subscribe (pub-sub) scenarios and push notifications to known endpoints across a wide variety of protocols. It eliminates the need to periodically check or poll for new information and updates. SNS supports:

• Reliable storage of messages for immediate or delayed processing
• Publish / subscribe – direct, broadcast, targeted “push” messaging
• Multiple subscriber protocols
• Amazon SQS, HTTP, HTTPS, email, SMS, mobile push, AWS Lambda

With these capabilities, you can provide parallel asynchronous processing of orders in the system and extend it to support any number of different business use cases without affecting the production environment. This is commonly referred to as a “fanout” scenario.

Rather than your web application pushing orders to a queue for processing, send a notification via SNS. The SNS messages are sent to a topic and then replicated and pushed to multiple SQS queues and Lambda functions for processing.

As the diagram above shows, you have the development team consuming “live” data as they work on the next version of the processing application, or potentially using the messages to troubleshoot issues in production.

Marketing is consuming all order information, via a Lambda function that has subscribed to the SNS topic, inserting the records into an Amazon Redshift warehouse for analysis.

All of this, of course, is happening without affecting your order processing application.

Summary

While I haven’t dived deep into the specifics of each service, I have discussed how these services can be applied at an architectural level to build loosely coupled systems that facilitate multiple business use cases. I’ve also shown you how to use infrastructure and application-level scaling techniques, so you can get the most out of your EC2 instances.

One of the many benefits of using these managed services is how quickly and easily you can implement powerful messaging capabilities in your systems, and lower the capital and operational costs of managing your own messaging middleware.

Using Amazon SQS and Amazon SNS together can provide you with a powerful mechanism for decoupling application components. This should be part of design considerations as you architect for the cloud.

For more information, see the Amazon SQS Developer Guide and Amazon SNS Developer Guide. You’ll find tutorials on all the concepts covered in this post, and more. To can get started using the AWS console or SDK of your choice visit:

• Getting Started with Amazon SQS
• Getting Started with Amazon SNS

Happy messaging!

Ed Lima, Solutions Architect

Our identities are what define us as human beings. Philosophical discussions aside, it also applies to our day-to-day lives. For instance, I need my work badge to get access to my office building or my passport to travel overseas. My identity in this case is attached to my work badge or passport. As part of the system that checks my access, these documents or objects help define whether I have access to get into the office building or travel internationally.

This exact same concept can also be applied to cloud applications and APIs. To provide secure access to your application users, you define who can access the application resources and what kind of access can be granted. (more…)

Michael G. Khmelnitsky, Senior Programmer Writer

Sometimes, messages can’t be processed because of a variety of possible issues, such as erroneous conditions within the producer or consumer application. For example, if a user places an order within a certain number of minutes of creating an account, the producer might pass a message with an empty string instead of a customer identifier. Occasionally, producers and consumers might fail to interpret aspects of the protocol that they use to communicate, causing message corruption or loss. Also, the consumer’s hardware errors might corrupt message payload. For these reasons, messages that can’t be processed in a timely manner are delivered to a dead-letter queue. (more…)

This post was graciously contributed by:

Chad Schmutzer Solutions Architect	Shawn O’Connor Solutions Architect

Today we are excited to announce that Amazon EC2 Container Service (Amazon ECS) now supports the ability to launch your ECS cluster on Amazon EC2 Spot Instances directly from the ECS console.

Spot Instances allow you to bid on spare Amazon EC2 compute capacity. Spot Instances typically cost 50-90% less than On-Demand Instances. Powering your ECS cluster with Spot Instances lets you reduce the cost of running your existing containerized workloads, or increase your compute capacity by two to ten times while keeping the same budget. Or you could do a combination of both!

(more…)

Aaron Friedman is a Healthcare and Life Sciences Partner Solutions Architect at AWS

Angel Pizarro is a Scientific Computing Technical Business Development Manager at AWS

This post is the fourth in a series on how to build a genomics workflow on AWS. In Part 1, we introduced a general architecture, shown below, and highlighted the three common layers in a batch workflow:

• Job
• Batch
• Workflow

In Part 2, you built a Docker container for each job that needed to run as part of your workflow, and stored them in Amazon ECR.

In Part 3, you tackled the batch layer and built a scalable, elastic, and easily maintainable batch engine using AWS Batch. This solution took care of dynamically scaling your compute resources in response to the number of runnable jobs in your job queue length as well as managed job placement. (more…)

Aaron Friedman is a Healthcare and Life Sciences Partner Solutions Architect at AWS

Angel Pizarro is a Scientific Computing Technical Business Development Manager at AWS

This post is the third in a series on how to build a genomics workflow on AWS. In Part 1, we introduced a general architecture, shown below, and highlighted the three common layers in a batch workflow:

• Job
• Batch
• Workflow

In Part 2, you built a Docker container for each job that needed to run as part of your workflow, and stored them in Amazon ECR. (more…)

Jakub Wojciak, Senior Software Development Engineer

Throughout our careers, we developers keep adding new tools to our development toolboxes. These range from the programming languages we learn, use, and become experts in, to architectural components such as HTTP servers, load balancers, and databases (both relational and NoSQL).

I’d like to kick off a series of posts to introduce you to the architectural components of messaging solutions. Expand your toolbox with this indispensable tool for building modern, scalable services and applications. In the coming months, I will update this post with links that dive deeper into each topic and illustrate messaging use cases using Amazon Simple Queue Service (SQS) and Amazon Simple Notification Service (SNS). (more…)

Aaron Friedman is a Healthcare and Life Sciences Partner Solutions Architect at AWS

Angel Pizarro is a Scientific Computing Technical Business Development Manager at AWS

This post is the second in a series on how to build a genomics workflow on AWS. In Part 1, we introduced a general architecture, shown below, and highlighted the three common layers in a batch workflow:

• Job
• Batch
• Workflow

In Part 2, you tackle the job layer and package a set of bioinformatics applications into Docker containers and store them in Amazon ECR. We illustrate some common patterns and best practices for these containers, such as how you can effectively use Amazon S3 to exchange data between jobs. (more…)

Aaron Friedman is a Healthcare and Life Sciences Partner Solutions Architect at AWS

Angel Pizarro is a Scientific Computing Technical Business Development Manager at AWS

Deriving insights from data is foundational to nearly every organization, and many customers process high volumes of data every day. One common requirement of customers in life sciences is the need to analyze these data in a high-throughput fashion without sacrificing time-to-insight.

Such analyses, which tend to be composed of a series of massively parallel processes (MPP) are well suited to the AWS Cloud. Many AWS customers and partners today, such as Illumina, DNAnexus, Seven Bridges Genomics, AstraZeneca, UCSC Genomics Institute, and Human Longevity, Inc., have built scalable and elastic genomics processing solutions on AWS.

One such use case is genomic sequencing. Modern DNA sequencers, such as Illumina’s NovaSeq, can produce multiple terabytes of raw data each day. The data must then be processed into meaningful information that clinicians and research can act on in a timely fashion. This processing of genomic data is commonly referred to as “secondary analysis”.

Most common secondary analysis workflows take raw reads generated from sequencers and then process them in a multi-step workflow to identify the variation in a biological sample compared to a standard genome reference. The individual steps are normally similar to the following: (more…)