As with any other financial company, at Marqeta, we have a good number of batch jobs, which we are migrating over to AWS Batch. However, even in managed mode, AWS Batch needs us to define Compute Environments, which are clusters of EC2 instances running ECS (and Docker) agents.

AWS Fargate was announced very recently at re:Invent 2017. Fargate adds a layer of abstraction on top of the Compute Environment, or the ECS Cluster. We no longer have to worry about the AMI, EC2 types, task placement, etc. In this post I cover the POC done to use Fargate over AWS Batch for batch processing, but this can also be used as a tutorial for running any type of tasks using Fargate. We create an ECS Task definition, a Fargate Cluster, and a Lambda to run the task using CloudWatch Event trigger.

Slides from my talk at Scale By the Bay 2017.

In this session I will talk about Immutable Deployments - which have become almost essential in the world of Microservices. As the frequency of deployments across multiple services increases with increasing granularity, it is critical to have repeatable, predictable and immutable deployments serving our customers. In practice, this is achieved via several DevOps tools. We will use Hashicorp Packer and Jenkins to build a simple, immutable AWS deployment of a hello-world microservice. Familiarity with AWS is recommended, but not required for this talk.

At Marqeta, we strive to continually evolve our platform to make it scalable and highly performant. We rely heavily on MySQL, and have many MySQL instances hosted across data centers, as well as on EC2s for various purposes. While refactoring some of our APIs, we thought of giving Amazon Aurora a try. Having heard about Aurora’s performance and high availability, this was definitely a great opportunity. Setting up a single node cluster (one db.t2.small) via the Console was the first step. After a few clicks, we had our first Aurora Cluster running happily. Next step was to fire up our regression tests while pointing to a schema in Aurora. Our database fixtures worked like a charm, and we were surprised to see all of our (thousand+) tests pass - while we knew it was a MySQL drop in replacement, we still expected some drama. Great first impression!

One of the very common architectural patterns is fan-out where an event is sent to multiple subscribers by a broker. An event can be like placing an order, which can then be handled by inventory service, record-keeping, as well as shipping service concurrently. These events can also be very frequent - like clickstreams, or search strings on a website. AWS allows for fan-out architecture with SNS topics, and SQS queues. SQS queue(s) can subscribe to SNS topics and receive any message sent to the SNS topic(s). It is documented that FIFO queues cannot be SNS subscribers here.

However, as I recently found out, SSE-enabled SQS queues cannot subscribe to SNS either. While AWS stops us from subscribing a FIFO queue to SNS, SSE-enabled queues are allowed to subscribe, but they never get any events. The purpose of this post is to document this previously undocumented behavior.

Its a challenge to log messages with a Lambda, given that there is no server to run the agents or forwarders (splunk, filebeat, etc.) on. Here is a quick and easy tutorial to set up ELK logging by writing directly to logstash via the TCP appender and logback. This is for a Java/Maven based Lambda.