News

Nicole Kobie

Appointment service and NHS 111 directory are the first major services to make the move

…

Read MoreNHS shifts two national services to the cloud

2019 saw a bumper year of data centre merger and acquisition deals with the total number passing 100 for the first time, according to Synergy Research. The record number has

Continue ReadingData centre M&A broke the 100 deal barrier in 2019 – driven by private equity

In this post, I’d like to show readers how to use Pulumi to create a VPC endpoint on AWS. Until recently, I’d heard of VPC endpoints but hadn’t really taken the time to fully understand what they were or how they might be used. That changed when I was presented with a requirement for the AWS EC2 APIs to be available within a VPC that did not have Internet access. As it turns out—and as many readers are probably already aware—this is one of the key use cases for a VPC endpoint (see the VPC endpoint docs). The sample code I’ll share below shows how to programmatically create a VPC endpoint for use in infrastructure-as-code use cases.

For those that aren’t familiar, Pulumi allows users to use one of a number of different general-purpose programming languages and apply them to infrastructure-as-code scenarios. In this example, I’ll be using TypeScript, but Pulumi also supports JavaScript and Python (and Go is in the works). (Side note: I intend to start working with the Go support in Pulumi when it becomes generally available as a means of helping accelerate my own Go learning.)

Here’s a snippet of TypeScript code that will create a VPC endpoint (I’ve randomized the resource IDs of the AWS resources below):

const ec2ApiEndpoint = new aws.ec2.VpcEndpoint("test-vpc-endpoint", { privateDnsEnabled: true, serviceName: "com.amazonaws.us-west-2.ec2", securityGroupIds: [ "sg-1db01e1f9be081d76" ], subnetIds: [ "subnet-0db757befe2c48624" ], vpcEndpointType: "Interface", vpcId: "vpc-0e2c348614e845763", tags: { Name: "test-vpc-endpoint" }
});

Let’s walk through this a bit (the full API reference is here):

• The privateDnsEnabled: true line enables us to use the “standard” EC2 API URL endpoint within the VPC and have it resolve to a private IP address out of the VPC/subnets. This relies on Route 53, and is necessary for use cases where you (the user) don’t have any control over how the API endpoint is being called from within the VPC (it may be a bit of commercial software you can’t modify). In this case, any software trying to reach “ec2.us-west-2.amazonaws.com” will resolve to a private IP address instead of the public IP address.
• The serviceName specifies which service is available via this endpoint (naturally, this is for EC2).
• The vpcEndpointType specifies “Interface”, which means that an ENI is created to allow traffic to reach the service represented by this endpoint. The other option is “Gateway”, which would require you to modify route tables, and is only supported for some services (like S3). This article provides an example of a gateway endpoint.

The other settings are all pretty self-explanatory. This being TypeScript, you could—of course—use references to variables or outputs from other portions of your Pulumi code instead of hard-coding values as I have in this example (and that would be highly recommended). This could be easily incorporated into a large portion of code that created a new VPC, new subnets, etc.

And that’s it! I know this is a fairly simple example, but I hadn’t found very many articles showing how this is done, and thought that sharing this may prove helpful to others. Feel free to contact me on Twitter if you have corrections (in the event I made a mistake), suggestions for improvement, or other feedback.

…

Read MoreCreating an AWS VPC Endpoint with Pulumi

I recently had a need to manually load some container images into a Linux system running containerd (instead of Docker) as the container runtime. I say “manually load some images” because this system was isolated from the Internet, and so simply running a container and having containerd automatically pull the image from an image registry wasn’t going to work. The process for working around the lack of Internet access isn’t difficult, but didn’t seem to be documented anywhere that I could readily find using a general web search. I thought publishing it here may help individuals seeking this information in the future.

For an administrator/operations-minded user, the primary means of interacting with containerd is via the ctr command-line tool. This tool uses a command syntax very similar to Docker, so users familiar with Docker should be able to be productive with ctr pretty easily.

In my specific example, I had a bastion host with Internet access, and a couple of hosts behind the bastion that did not have Internet access. It was the hosts behind the bastion that needed the container images preloaded. So, I used the ctr tool to fetch and prepare the images on the bastion, then transferred the images to the isolated systems and loaded them. Here’s the process I followed:

1. On the bastion host, first I downloaded (pulled) the image from a public registry using ctr image pull (the example I’ll use here is for the Calico node container image, used by the Calico CNI in Kubernetes clusters):

   ctr image pull docker.io/calico/node:v3.11.2
   

   (Note that sudo may be needed for all these ctr commands; that will depend on your system configuration.)

   I repeated this process for all the other images the isolated systems also needed.

2. Still on the bastion host, I exported the pulled images to standalone archives:

   ctr image export calico-node-v3.11.2.tar \
   docker.io/calico/node:v3.11.2
   

   The general format for this command looks like this:

   ctr image export <output-filename> <image-name>
   

   If you don’t know what the image name (according to containerd) is, use ctr image ls.

3. After transferring the standalone archives to the other systems (using whatever means you prefer; I used scp), then load (or import) the images into containerd with this command:

   ctr image import <filename-from-previous-step>
   

   Repeat as needed for additional images. It appears, by the way, that using wildcards in the ctr image import command won’t work; I had to manually specify each individual file for import.

   If you need these images to be available to Kubernetes, you must be sure to add the -n=k8s.io flag to the ctr image import command, like this:

   ctr -n=k8s.io images import <filename-from-previous-step>
   

4. Verify that the image(s) are present and recognized by containerd using ctr image ls.

   If you specified the k8s.io namespace when importing the images in the previous step—so as to make the images available to Kubernetes—then you can verify that CRI (Container Runtime Interface, the means by which Kubernetes talks to containerd) sees these images by running crictl images (again, sudo may be required, based on your configuration).

That should do it for you!

How I Tested

To test this procedure, I used multiple Ubuntu 18.04 instances on AWS. One instance had Internet access (the bastion host); the others did not. The instances had been built from AMIs prepared using Packer and Ansible with playbooks from the Kubernetes “image-builder” repository (specifically, the playbooks used to build the Kubernetes Cluster API images), so containerd and all the other CLI tools mentioned in this article were already installed and available for use.

Additional Resources

I did find a few sites with some containerd information while searching, and wanted to include them here for credit and additional context:

Getting started with Containerd

crictl User Guide

I hope the information in this post is helpful. If you have questions, comments, suggestions, or corrections, you are welcome to contact me on Twitter, or drop me an e-mail (my e-mail address isn’t too hard to find).

…

Read MoreManually Loading Container Images with containerD

In July of 2018 I talked about Polyglot, a very simple project I’d launched whose only purpose was simply to bolster my software development skills. Work on Polyglot has been sporadic at best, coming in fits and spurts, and thus far focused on building a model for the APIs that would be found in the project. Since I am not a software engineer by training (I have no formal training in software development), all of this is new to me, and I’ve found myself encountering lots of questions about API design along the way. In the interest of helping others who may be in a similar situation, I thought I’d share a bit here.

I initially approached the API in terms of how I would encode (serialize?) data on the wire using JSON (I’d decided on using a RESTful API with JSON over HTTP). Starting with how I anticipated storing the data in the back-end database, I created a representation of how a customer’s information would be encoded (serialized) in JSON:

{ "customers": [ { "customerID": "5678", "streetAddress": "123 Main Street", "unitNumber": "Suite 123", "city": "Anywhere", "state": "CO", "postalCode": "80108", "telephone": "3035551212", "primaryContactFirstName": "Scott", "primaryContactLastName": "Lowe" } ]
}

It’s pretty obvious, though, that this representation is closely tied to how I anticipate I’ll store customer data in the back-end database, right down to the field names. One of the books I’ve been reading is Building Microservices, an O’Reilly book by Sam Newman (see here; I’ll post a review of the book once I complete it). One of the things Newman cautions readers against in his book is exposing implementation details internal to a service to other services—did this qualify? Newman also discussed hypermedia as the engine of application state (HATEOAS, what an acronym!), but I’ll be honest in admitting that it didn’t click with me at first.

While still working through Newman’s book, I added REST API Design Rulebook by Mark Masse to the mix (see my review of the book). Masse talked extensively about HATEOAS, but—as you can see in my review of the book—it still didn’t click for me.

Next up, I added RESTful Web APIs by Leonard Richardson, Sam Ruby, and Mike Amundsen to my reading list (here’s the O’Reilly page for the book; I’m still reading the book so no review yet). Whether by repetition or by seeing the information presented in a different context, the discussion of HATEOAS and hypermedia by Richardson et al. finally made sense. It also unlocked some of the ideas presented by Masse and Newman earlier, so that I was able to really understand that I’d made several critical mistakes in my approach thus far:

• I chose a media type (JSON) before fully understanding the particulars of my application without realizing the constraints that media type creates.
• The client representation was too closely tied to the back-end/server-side implementation, making things far too brittle (what would happen if I needed to change the database schema?).
• I invented new field names instead of re-using properties and fields that are already well-known and well-accepted (like re-using information from an appropriate microformat).
• The client representation doesn’t provide any information on state transitions (How is a customer record updated? How is a customer record deleted?) or relationships between resources (What’s the relationship between an order record and a customer record? Or vice versa?).

It’s clear that something more than just encoding data in JSON is required.

The question, for me, is how much more? And what to include, exactly? Richardson et al. seem to advocate the Collection+JSON format, but it’s only a personal standard (see his book for definitions of various standards) and not necessarily widely accepted/used. However, even within Collection+JSON, the specific links provided as hypermedia are left as “implementation details.” So what sort of information makes sense?

Given that Polyglot is intended to be a simple order entry/lookup system, a few potential ideas come to mind:

• All discussions of HATEOAS and hypermedia agree that a “self” link is needed that provides the URL to the resource (customer record or order record) being returned.
• It seems reasonable to return “related” information. So, if viewing an order record, provide a link to the associated customer record (for example), the previous order, the next order, etc.

Clearly I have lots of work to do!

Now comes the task of shaping the limited work I’ve done so far into something more usable. I also need to finish reading RESTful Web APIs; although the content was difficult for me to grok at first, recently I’ve found that I’m beginning to more fully understand the concepts the authors are presenting. I still have lots of questions—how do things like OpenAPI fit in here?—so if you have answers, feel free to hit me on Twitter. This is definitely a learning exercise for me, so I welcome any other resources I should be using to help guide my way.

…

Read MoreThinking and Learning About API Design