AWS Amazon Web Services training from QA

Converting CloudFormation JSON templates into YAML

People keep asking me why I'm so excited about YAML in CloudFormation. The simple answer is that it's easier for humans to read, and you can put comments in more easily.

Daniel Ives | 7 December 2016

AWS Certified

CloudFormation and YAML

Ever since YAML support in CloudFormation (hereafter referred to as ‘CFn’) templates was announced, I’ve developed a new hobby: converting templates from JSON to YAML.

Delegates keep asking me why I’m so excited about YAML in CFn and the simple answer is that it’s easier for humans to read, and you can put comments in more easily. But the more complex answer is that it’s also easier for humans to write and that user-data is so much clearer now.

However, I keep running into little road bumps in my conversions, so this post is partly for my benefit as a reminder of what I’ve discovered.

For this post, I’ll use the example of converting the WordPress single instance sample template. This one’s in eu-west-1, but there’s a copy in every region.

At the end, I was going to compare the number of lines of code (‘LOC’) between the original JSON and the new-and-improved YAML, but then I remembered that there’s a stupidly long set of mappings in this template, including a completely unused one for NAT instances but also the Linux AMI mappings (and AllowedValues) cover every single instance type. Are you seriously going to run WordPress on a cc2.8xl?!?!?! So I might cheat and compare LOC between everything except the Mappings section and the AllowedValues for InstanceType…

NOTE: I’ve got that re:Invent feeling - that as I’m writing this, someone from AWS is preparing to announce a tool that does awesome and intelligent conversions from JSON-Cfn to YAML-Cfn.

Step 1: Convert the JSON to YAML

Clearly, as YAML is a super-set of JSON, there are many ways to perform the initial conversion, from online tools to Python libraries, but I use Atom a lot these days and there’s a handy add-in for that called json-convertor (‘j-c’). You can install it by going to File | Settings | Install or at the command line with:

apm install json-converter

Step 2: Remove unnecessary quotes

For now, we’ll treat quotes around intrinsic functions as necessary for now, because we also want to clean those up in our subsequent passes.

Some quotes are clearly necessary. Working out which ones are which is not straightforward, but it’s all about which characters have special meaning in YAML. There’s two ways to do this; you can either remove them all using find and replace and then discover which ones were necessary by validating the template repeatedly:

aws cloudformation validate-template --template-body file://path/to/template.yaml

Or, you can base it on this (utterly incomplete) list of necessary ones:

  • ones around a RegEx containing square brackets [] (i.e. SSHLocation.AllowedPattern doesn’t need them, DBName.AllowedPattern does)
  • single asterisks * in IAM policies etc. (no example in the WP template)


Here’s a list of ones that are OK:

  • CFn types (i.e. AWS::EC2::KeyPair::KeyName) - j-c leaves quotes because they contain :s
  • dates (i.e. AWSTemplateFormatVersion)
  • whole numbers (i.e. DBName.MinLength)
  • booleans (i.e. DBUser.NoEcho)
  • urls (i.e.


And ones I suspect are required so haven’t bothered to test:

  • masks (i.e. mode in file in cfn-init) - I suspect leading zeros would be ignored if YAML thought it was a number not a string.


Step 3: Tidy up those Fn::FindInMaps for ImageIds

We have some nasty nested ones in the WordPress template, but WebServer | Properties | ImageId in JSON looks like this:

"ImageId" : { "Fn::FindInMap" : [ "AWSRegionArch2AMI", { "Ref" : "AWS::Region" },
                  { "Fn::FindInMap" : [ "AWSInstanceType2Arch", { "Ref" : "InstanceType" }, "Arch" ] } ] },

And we can make it look much nicer:

ImageId: !FindInMap
  - AWSRegionArch2AMI
  - !Ref AWS::Region
  - !FindInMap
      - AWSInstanceType2Arch
      - !Ref InstanceType
      - Arch

Or, possibly:

ImageId: !FindInMap
  - AWSRegionArch2AMI
  - !Ref AWS::Region
  - !FindInMap [AWSInstanceType2Arch, !Ref InstanceType, Arch]

Or, let’s really go for LOC reduction now! Note the quotes. We’re getting very nest-y now and I think CFn / YAML is starting to struggle:

ImageId: !FindInMap [AWSRegionArch2AMI, !Ref "AWS::Region", !FindInMap [AWSInstanceType2Arch, !Ref InstanceType, Arch]]

Step 4: Tidy up those function calls in user-datas and cfn-inits

First, a note about string literals in YAML.

The pipe character | tells YAML to preserve newlines, which means we can lose a lot of \ns.

The > character tells it to remove newlines, which means we can wrap stuff.

The - tells YAML to lose trailing newlines at the end of the string.

The + tells YAML to preserve trailing newlines. Very useful in some instances.

Generally speaking, we can eliminate all those Fn::s and replace them with !s, but there are restrictions in CFn around nesting short-form function calls, presumably to do with how the template is being pre-processed by CFn, so again it’s worth validating your templates after every tweak. Leave Fn::Base64s alone as a rule.

But for the simplest examples, we can replace Fn::Joins and replace them with !Subs, which makes most files and outputs so much neater. Also, using !Subs, we can almost entirely eliminate Fn::GetAtts.

Simple Example 1: Losing joins and refs

WebServer | Metadata | AWS::CloudFormation::Init | install_cfn | files | cfn-hup.conf | content

In the original JSON (clearly the original template author has tried to make it as legible as possible with the formatting):

"content": { "Fn::Join": [ "", [
  "stack=", { "Ref": "AWS::StackId" }, "\n",
  "region=", { "Ref": "AWS::Region" }, "\n"

J-c is seeing those newlines and trying to preserve that formatting, which looks a bit messy, but we can lose the join completely, so:

    - ''
    - - |
      - stack=
      - Ref: 'AWS::StackId'
      - |+

      - region=
      - Ref: 'AWS::Region'
      - |+

becomes the much neater:

content: !Sub |

Note the | to preserve the newlines.

Simple Example 2: Losing Fn::GetAtts

Outputs | WebsiteURL | Value

The original JSON:

"Value" : { "Fn::Join" : ["", ["http://", { "Fn::GetAtt" : [ "WebServer", "PublicDnsName" ]}, "/wordpress" ]]},

isn’t visually enhanced by j-c. Replace the Fn::GetAtt and the Fn::Join so that:

    - ''
    - - 'http://'
      - 'Fn::GetAtt':
          - WebServer
          - PublicDnsName
      - /wordpress

becomes the much, much neater:

Value: !Sub

Trickier Example 3: Interesting formatting

WebServer | Metadata | AWS::CloudFormation::Init | install_cfn | files | /etc/cfn/hooks.d/cfn-auto-reloader.conf | content

Two tidy-up options here, depending on whether you care about line wrapping. Here’s the original JSON version:

"content": { "Fn::Join": [ "", [
  "action=/opt/aws/bin/cfn-init -v ",
          "         --stack ", { "Ref" : "AWS::StackName" },
          "         --resource WebServer ",
          "         --configsets wordpress_install ",
          "         --region ", { "Ref" : "AWS::Region" }, "\n"

Here’s the j-c converted version (j-c is trying to preserve the formatting as much as possible with the pipes and the angle brackets):

    - ''
    - - |
      - |
      - >
      - 'action=/opt/aws/bin/cfn-init -v '
      - '         --stack '
      - Ref: 'AWS::StackName'
      - '         --resource WebServer '
      - '         --configsets wordpress_install '
      - '         --region '
      - Ref: 'AWS::Region'
      - |+

and here’s your conversion if you’re not worried about line wrapping (which will ironically be wrapped when you see this post):

content: !Sub |+
  action=/opt/aws/bin/cfn-init -v --stack ${AWS::StackName} --resource WebServer --configsets wordpress_install --region ${AWS::Region}

and if you do care, and don’t mind putting \s in commands, it gets even better!

  !Sub |+
    action=/opt/aws/bin/cfn-init -v \
             --stack ${AWS::StackName} \
             --resource WebServer \
             --configsets wordpress_install \
             --region ${AWS::Region}

Trickier Example 4: escaping single-quote literals

WebServer | Metadata | AWS::CloudFormation::Init | install_wordpress | files | /tmp/setup.mysql

Decipherable JSON:

"content" : { "Fn::Join" : ["", [
  "CREATE DATABASE ", { "Ref" : "DBName" }, ";\n",
  "CREATE USER '", { "Ref" : "DBUser" }, "'@'localhost' IDENTIFIED BY '", { "Ref" : "DBPassword" }, "';\n",
  "GRANT ALL ON ", { "Ref" : "DBName" }, ".* TO '", { "Ref" : "DBUser" }, "'@'localhost';\n",

So those single-quotes cause problems for us and we’ll need to escape them. Or will we? I need to test my template out I guess…The following validates OK, but I have a feeling it won’t build because of those @s. OMG, yes it does! Perhaps I should re-classify this as not particularly tricky!

content: !Sub |
  CREATE USER '${DBUser}'@'localhost' IDENTIFIED BY '${DBPassword}';
  GRANT ALL ON ${DBName}.* TO '${DBUser}'@'localhost';

Trickier Example 5: Multiple lines of “interesting” formatting

WebServer | Properties | UserData


"Fn::Base64" : { "Fn::Join" : ["", [
               "#!/bin/bash -xe\n",
               "yum update -y aws-cfn-bootstrap\n",

               "/opt/aws/bin/cfn-init -v ",
               "         --stack ", { "Ref" : "AWS::StackName" },
               "         --resource WebServer ",
               "         --configsets wordpress_install ",
               "         --region ", { "Ref" : "AWS::Region" }, "\n",

               "/opt/aws/bin/cfn-signal -e $? ",
               "         --stack ", { "Ref" : "AWS::StackName" },
               "         --resource WebServer ",
               "         --region ", { "Ref" : "AWS::Region" }, "\n"

Again, depends on your formatting preferences. If you’re avoiding long lines, you have a couple of options; one involves more !Joining than the other. I’ve preferred the version that uses less, on the basis that if you’re basically copying-and-pasting this stuff and you care about text wrapping, you’ll already be using \s. So here’s that user data:

Fn::Base64: !Sub |+
  #!/bin/bash -xe
  yum update -y aws-cfn-bootstrap
  /opt/aws/bin/cfn-init -v \
      --stack ${AWS::StackName} \
      --resource WebServer \
      --configsets wordpress_install \
      --region ${AWS::Region}
  /opt/aws/bin/cfn-signal -e $? \
      --stack ${AWS::StackName} \
      --resource WebServer \
      --region ${AWS::Region}

Without the \s I’d need more joins because the leading -s would cause problems. My final minified version will have a “don’t care about wrapping” version.

Step 5: Tidy up all the remaining Ref:s

This could arguably be done at step 2.5 or later, but I’ve kept it until last on the grounds that if I’d said to do this earlier, and you were following along blindly and find-and-replacing, it might have made a mess of our user-datas and cfn-inits.

So !Ref, as opposed to Ref:, doesn’t need to be a separate key, meaning we can save lots of lines. What I mean is WebServerSecurityGroup | Properties | SecurityGroupIngress | CidrIp can go from this:

  Ref: SSHLocation

to this:

CidrIp: !Ref SSHLocation

There aren’t many in the WordPress template, and make sure you don’t accidentally remove a newline / hyphen from one that’s part of an array (such as WebServer | Properties | SecurityGroups in this case), but that could save a lot of LOC in some templates.

And the final line count







1+129 = 130




59+261+1+6+4 = 331




59+261 = 320


1Mappings, AllowedValues and my comments in the non-minified version are ignored

Finally, a couple of other tricks, including FindInMap in Sub

Something I’ve been trying to get to work for a while, and seen floating around the interwebs, is including a Fn::FindInMap in a Fn::Sub (or indeed a !FindInMap in a !Sub) in a CloudFormation template. I wanted to use it to create a shell script which invokes run-instances as part of a cfn-init section, but it could also prove useful in a user-data section. It’s not relevant to the WordPress template, but it’s been bugging me for a while. So I solved it today, which is what prompted this blog in the first place.

I banged my head against several brick walls while working on this, but I believe I have a general format now.

Replacing Fn::GetAtts with placeholders is one thing, but replacing them with what the AWS documentation confusingly refers to as “mappings” [not to be confused with ‘Mappings’] is another level. I’ll present it without the rabbit-holes I fell into.

So according to the afore-mentioned documentation, one can substitute a

${pseudo-parameter}, a ${resource.attribute}, a ${mapping}, a ${!literal-string-do-not-substitute-this}

in a !Sub

Sadly, there’s no documentation about how to do that with a FindInMap, hence this post. I couldn’t work out how to do it inline, despite trying multiple combinations of colons and quotes, so I decided that I’d need to use a mapping. Now, such an item will have one of two forms; either a one line sub or a multiline sub. Chances are you’re looking for a mutliline solution.

NOTE: I haven’t actually tried this with user-data yet, only on Outputs, but the principal should be globally applicable. Just don’t blame me when the wheels fall off.


The !Sub function wants two arguments (an array) passed to it, but Fn::FindInMap has some odd behaviour alluded to above. Arg0 contains the mapping, arg1 the map key and its value. This is where things normally go wrong.

  Value: !Sub
    - I've been subbed ${Subbed}
    - Subbed:
        Fn::FindInMap: [RegionOS2AMI, Ref: "AWS::Region", Windows]


Buoyed by my success with a single line, I went over various iterations trying to get it to work with multiline strings, failing spectacularly until I realised that the validation errors I was getting were actually telling me what the problem was. Again it’s down to the order in which things are evaluated, so here’s a successful multiliner:

  Value: !Sub
    - |
      This is line one
      And line ${Subbed2} two
    - Subbed2:
          - RegionOS2AMI
          - !Ref AWS::Region
          - Windows

NOTE: the pipe’s location at the start of arg0, rather than where I had it originally which was after the !Sub. Note also the multiline Fn::FindInMap; the single line version from the previous example should work equally well.

But do you really need it?

In this case, no I don’t. AMI Mappings are so last decade. The modern approach would be to use a CustomResource to look up the AMI. Here’s a walkthrough of creating such a thing. And then my !Sub would be as simple as:

aws ec2 run-instances --image-id ${AMILookup.Id}

I hope you find this useful.


Daniel Ives

Daniel Ives

Principal Technologist - AWS

Daniel joined QA in 2006, having previously worked as a developer trainer on the Microsoft stack. He is an Authorized Amazon Instructor and holds all seven of the current AWS certifications. As a Principal Technologist, Daniel focuses on creating and delivering courses about cloud services, service-oriented architectures, data engineering and enterprise application integration. Areas of expertise: Amazon Web Services, C#, .NET and agile development. His areas of interest include all of the above, plus Google Cloud Platform, Microsoft Azure, Python, sailing, skiing and cycling, although not necessarily in that order or at the same time.
Talk to our learning experts

Talk to our team of learning experts

Every business has different learning needs. QA has over 30 years of experience in combining the highest quality training with the most comprehensive range of learning services, ensuring the very best fit for your organisation.

Get in touch with our learning experts to talk about how we can help.