This is a continuation in the Klayers series, where I deep dive into the architecture of Klayers. At its core, Klayers is a collection of AWS Lambda Layers for Python3, with the idea that python packages in layers is more efficient than packaging them with application code.

Visit the GitHub repo here, where you'd find 50+ lambda layers for public consumption across most AWS regions (including HK and Oman). This post is how I automated the building of layers inside lambda functions -- but specifically on layers composed of Python Packages (e.g. requests, beautifulsoup4, etc)

Python Packages for Dummies

As a primer, let's take a look at python packages in general. Python utilizes the Python Package Index (or PyPI), this is similar to Maven for Java or NPM for Node. It's simply a package manager that helps with the installation of python packages for your application.

In order to help with this, there is a program called pip that helps with the installation of python packages. While pip isn't limited to packages from PyPI, you can use it to install packages from other sources as well -- it and PyPI are the dynamic duo of Python packages.

The problem is that while Python is a interpreted language, there are some components of it that are OS specific. When you pip install into Windows, you get a different package installation than when you pip install into Ubuntu or OSX. pip detects your OS and installs specific files for your specific purpose -- sometimes those files need to be compiled for your OS as well.

Which means, if you wanted to put a Python Package into a Lambda Layer, it would need the AWS Linux version of that Python Package (Ubuntu might be close enough, CentOS is even better), because Lambda functions run on AWS Linux. And because not many folks run Linux as their core distribution, the general recommendation for creating these lambda layers has always been to use Docker.

It's very easy to use a docker container based on lambci/lambda:build-python3.7, to build python packages for lambda. In fact I even have a script that does that here.

But to me, this seemed sub-optimal. After all, we preach the 'serverless first' mantra, yet when it comes to building lambda layers -- we default to a docker container on a serverful laptop .... there must be a serverless way.

I've been bitten by the serverless bug lately, and just completed my latest hobby project this week. It's a fully serverless pipeline that builds python packages as Lambda layers -- and it uses Lambda functions to do so.

As a quick side-note, before we dive into this intro, Lambda layers are simply zip files that get extracted into the /opt directory of your lambda function's container. This is a special directory because it's included in various paths including:

Just this week, my team was on the cusp of demo-ing a product they've been working on for the last 2 months, only for a build process to fail, just hours before the demo to some very high ranking people.

Troubleshooting the build took a while, but eventually we found the root cause, a missing package version! This probably wouldn't have been a big deal, had we not stumbled across it at midnight the day before an important demo -- but nevertheless failing builds are never a good thing!

I've been really digging into Lambda Layers lately, and once you begin using layers you'll wonder how you got by without them.

Layers allow you to package just about anything into lambda, but in a modular way. So elements of your code that don't change much, can be packaged into layers, while keeping your actual lambda deployment for just the code that's changing.

It's akin to docker cache, where you keep the un-changing elements higher up in your docker file, separate from the code that always changes. The difference though, is that docker cache speeds up builds, while layers speeds up lambda deployments.

But layers aren't magic, and they're still limited by the AWS size limit, hence your entire function (including all it's layers) need to be no larger than 250MB (unzipped).

Which is tough for something like spaCy -- because its default installation size on a AWS Linux is ~400MB (or 492MB based on my quick installation on lambci for python3.7). So, in order to get spaCy working on a lambda, certain tweaks are going to be necessary.

Some have tried working around this problem by installing spaCy onto the lambda container on cold-start -- i.e. pull the data into lambda only when you have access to the 512MB in /tmp. Cool solution, but it almost completely fills out /tmp, and makes a cold-start even slower.

A more optimal solution would be to reduce the size of the spaCy installation and have it fit into a layer! Fortunately I found a GitHub issue after some googling that enables us to do exactly this.

It involves removing unnecessary language files, which spaCy lazy load in, If you're only interested in one language, you can simply remove the unnnecessary language files in the site-packages/spacy/lang directory.

After manually removing all non-English (en) language files, I managed to reduce the size of the spaCy package to 110MB, which fits very nicely into a lambda layer, in the end my lang directory only had the following files:

As a bonus, I also included the English en_core_web_sm-2.1.0 model, to make the lambda layer fully usable on its own .

Finally I published it as a publicly available layer, for anyone to consume. One of the amazing things about layers, is that once a layer is made, it can be shared across AWS for anyone to consume.

Recently I found myself working with an S3 bucket of 13,000 csv files that I needed to query. Initially, I was excited, because now had an excuse to play with AWS Athena or S3 Select -- two serverless tools I been meaning to dive into.

But that excitement -- was short-lived!

For some (as yet unexplained) reason, AWS Athena is not available in us-west-1. Which seemingly, is the only region in the us that Athena is not available on!

And.... guess where my bucket was? That's right, the one region without AWS Athena.

Now I thought, there'd a simple way to copy objects from one bucket to another -- after all, copy-and-paste is basic computer functionality, we have keyboard shortcuts to do this exact thing. But as it turns out, once you have thousands of objects in a bucket, it becomes a slow, painful and downright impossible task to get done sanely.

For one, S3 objects aren't indexed -- so AWS doesn't have a directory of all the objects in your bucket. You can do this from the console -- but it's a snap-shots of your current inventory rather than a real-time updated index, and it's very slow -- measured in days slow! An alternative is to use the list_bucket method.

But there's a problem with list_bucket as well, it's sequential (one at a time), and is limited 'just' 1000 items per request. A full listing of a million objects would require 1000 sequential api calls just to list out the keys in the your bucket. Fortunately, I had just 13,000 csv files, so this part for fast, but that's not the biggest problem!

Once you've listed out your bucket, you're then faced with the monumentally slow task of actually copying the files. The S3 API has no bulk-copy method, and while you can use the copy_object for a file or arbitrary size, but it only works on one file at a time.

Hence copying 1 million files, would require 1 million API calls -- which could be parallel, but would have been nicer to batch them up like the delete_keys method.

So to recap, copying 1 million objects, requires 1,001,000 API requests, which can be painfully slow, unless you've got some proper tooling.

AWS recommend using the S3DistCP, but I didn't want to spin up an EMR server 'just' to handle this relatively simple cut-n-paste problem -- instead I did the terribly impractical thing and built a serverless solution to copy files from one bucket to another -- which looks something like this:

The Serverless framework (SF) is a fantastic tool for testing and deploying lambda functions, but it's reliance on cloudformation makes it clumsy for infrastructure like DynamoDB, S3 or SQS queues.

For example, if your serverless.yml file had 5 lambdas, you'd be able to sls deploy all day long. But add just one S3 bucket, and you'd first have to sls remove before you could deploy again. This different behavior in the framework, once you introduce 'infra' is clumsy. Sometimes I use deploy to add functions without wanting to remove existing resources.

Terraform though, keeps the state of your infrastructure, and can apply only the changes. It also has powerful commands like taint, that can re-deploy a single piece of infrastructure, for instance to wipe clean a DynamoDB.

In this post, I'll show how I got Terraform and Serverless to work together in deploying an application, using both frameworks strengths to complement each other.

**From here on, I'll refer to tool Serverless Framework as SF to avoid confusing it with the actual term serverless

Terraform and Serverless sitting on a tree

First some principles:

• Use SF for Lambda & API Gateway
• Use Terraform for everything else.
• Use a tfvars file for Terraform variable
• Use JSON for the tfvars file
• Terraform deploys first followed by SF
• Terraform will not depend on any output from SF
• SF may depend on output from terraform
• Use SSM Parameter Store to capture Terraform outputs
• Import inputs into Serverless from SSM Parameter Store
• Use workspaces in Terraform to manage different environments.
• Use stages in Serverless to manage different environments.
• stage.name == workspace.name

In the end the deployment will look like this:

First a definition.

A lambda function is a service provided by aws that runs code for you without the introducing the complexity of provisioning servers of managing Operating Systems. It belongs in a category of architectures called serverless architectures.

There's a whole slew of folks trying to define with is serverless, but my favorite definition is this.

Serverless means No Server Ops

Joe Emison

They're the final frontier of compute, where the idea is that developers just write code, while allowing AWS (or Google/MSFT) to take care of everything else. This includes H/W management, OS Patching, even application level maintenance like Webserver upgrades are not your problem anymore with serverless.

Nothing runs on fairy-dust though, serverless still has servers -- but in this world those servers, their operating systems, and the underlying runtime (e.g. Python, Node, JVM) are fully managed services that you pay per use.

As a developer you write some code into a function. Upload that function to AWS -- and now you can invoke this function over and over again without worrying about servers, operating systems or run-time.

But how does AWS achieve this?

Before we can understand how to secure a serverless function, we need to at least have a fair understanding of how Serverless functions (like AWS Lambda) work.

So how does a lambda function work?

As a company, Grab has done enormously well for itself, and naturally will be the target of some hate.

But I think there's a deeper issue with Grab that needs addressing before it becomes an unsolvable problem.

Grab is a win-win

Let's start with what makes Grab so appealing.

Grab (at least in my mind) is the highest paying hourly wage job in the country. As long as you possess a car, and a valid driving license you can be a Grab driver, earning significantly more than any other hourly wage job.

According to this WOB article (which looks suspiciously like a paid ad), the average Grab driver earns RM5,000 per month, which is crazy money for a unskilled job -- and yes driving Grab is unskilled labour.

For unskilled work in Malaysia, earning RM5,000 per month is a god-send, after all even graduate employees don't earn that much. And like all hourly wage jobs, the more hours you put in, the more money they make -- 5,000 is just where it starts

So this seems like a win-win for everyone, drivers get to earn, and at the same time provide a service that is in high demand.

And in truth, Grab is a win-win -- at least for now.

Fast-forward

The problem is that when you fast-forward 10 years, or just 2 elections from now.

Most Grab drivers I've met aren't doing this part-time. They're driving as a full-time job, and they're putting in serious hours (10-12 a day) to make serious money. That means they've no time or to up-skill themselves, because every hour learning a new skill is an hour they could have been driving.

The cost of learning to them is a double-whammy, first they spend on acquiring the new skill (like everybody else), but also the lose income from their not driving. This for most, will be too high a price to pay.

You might argue that driving isn't un-skilled. But all it takes to be a Grab driver is a driving license and a car, skills don't factor into this. Grab doesn't care if you're a PhD, diploma holder or SPM drop-out, it'll pay the same.

Grab views all of it's drivers as a supplier of the one commodity it needs -- cars to move passengers. The only time Grab pays more to drivers is when they turn on the auto-accept feature, because that makes their algorithm more efficient. The more subservient you are to the algorithm, the better it will reward you -- that is a pretty nasty feeling.

So as more folks join the Grab band-wagon, we're sucking out skilled labour from the job-market. Leaving the entire country, as a whole, worse off in terms of competitiveness. But we're just getting started.

Android TV boxes, are computers that stream content from the internet onto your TV. The difference between them and your smart-phone is that it has a HDMI connector to your TV, and it usually comes pre-loaded with software to illegally stream content.

While the boxes themselves, are general purpose computers running Android (the most popular OS today), the real focus of any regulation should be on the software on the device and the internet-based streaming services that support them.

Which seems to be the case...

Today, TheStar reports that the MCMC will begin blocking these unauthorized streaming services, rendering the boxes that connect to them useless.

But, if the MCMC uses it's usual method of DNS filtering to implement the block, it'll be trivial for most folks to circumvent the issue, the boxes run Android after all. The government will very quickly find itself in a cat and mouse situation in trying to block them.

I'm still deep in my lambda love affair. Here's my talk at the AWS meetup in SG on Lambda layers :