AWSLambda

Note, the master branch of this package has recently been updated for Julia 0.6.2. The update includes breaking API changes. This README.md file describes the new interface. The updated version does not yet have a release tag pending futhrur testing. If you would like to help with testing, please follow the instructions below.

AWS Lambda Interface for Julia.

If you are new to Lambda, please read What is AWS Lambda and try the Create a Simple Lambda Function (Node.js) exercise.

With this package you can:

• Invoke a Lambda function (defined in Node.js, Python, etc) from Julia.
• Evaluate a Julia expression using Lambda.
• Create a local Julia function who's body is evaluated on in the cloud.
• Deploy a Julia Lambda function that can be called using the AWS SDK for JavaScript, Python etc.

Getting Started

AWS Credentials

Your AWS credentials should be configured in the ~/.aws/credentials file or in the environment variables AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY. See the AWSCore configuration documentation or the AWS CLI User Guide for mode detail.

You can verify that your credentials are working by calling the IAM GetUser API:

julia> using AWSCore
julia> AWSCore.set_debug_level(1)
julia> AWSCore.Services.iam("GetUser")["User"]
Loading "octech" AWSCredentials from /Users/sam/.aws/credentials... (AKIAXXXXXXXXXXXXXXXX, XXX...)
Dict{String,Any} with 7 entries:
  "Arn"              => "arn:aws:iam::XXXXXXXXXXXX:user/sam"
  ...

Invoke a Lambda function from Julia

This example assumes that you created a Node.js Lambda function MyFunction in the Create a Simple Lambda Function exercise.

julia> using AWSLambda
julia> AWSLambda.invoke_lambda("MyFunction")
"Hello from Lambda"

Using the AWS Lambda Management Console, modify the Node.js source code of MyFunction to return the values of event arguments foo and bar:

exports.handler = (event, context, callback) => {
    let x = 'foo=' + event['foo'] + ', ' +
            'bar=' + event['bar']
    callback(null, x)
};

Now the function can be called with keyword arguments foo= and bar=:

julia> AWSLambda.invoke_lambda("MyFunction", foo=7, bar="xyz")
"foo=7, bar=xyz"

Deploy jl_lambda_eval to Lambda

The jl_lambda_eval Lambda function takes a Julia expression as input and returns the result of evaluating that expression. This function must be deployed to your AWS account before AWSLambda.@lambda_eval, AWSLambda.@lambda or AWSLambda.@deploy can be used:

julia> using AWSLambda
julia> AWSLambda.deploy_jl_lambda_eval()

After deployment jl_lambda_eval should be visible in the AWS Lambda Console (be sure to set the console to the correct AWS region).

To learn more about the internals of jl_lambda_eval see the Dockerfile and the make.jl script.

Run a Julia expression in the cloud

The @lambda_eval macro passess a Julia expression to the jl_lambda_eval Lambda and returns the result.

>julia AWSLambda.@lambda_eval run(`uname -snr`)
Linux ip-10-13-21-185 4.9.85-38.58.amzn1.x86_64

>julia AWSLambda.@lambda_eval ENV["LAMBDA_TASK_ROOT"]
"/var/task"

julia> AWSLambda.@lambda_eval @time binomial(big(10^6), big(10^5))
  0.559756 seconds (7.31 k allocations: 94.242 KiB)

73331919...

A more complex expression example:

julia> x = AWSLambda.@lambda_eval begin
           l = open(readlines, "/proc/cpuinfo")
           l = filter(i->ismatch(r"^model name", i), l)
           strip(split(l[1], ":")[2])
       end

"Intel(R) Xeon(R) CPU E5-2680 v2 @ 2.80GHz"

An expression with an embedded module:

julia> r = AWSLambda.@lambda_eval begin

           module Foo

               export foo

               using HTTP
               using JSON

               const url = "http://httpbin.org/ip"

               function foo()
                   JSON.parse(String(HTTP.get(url).body))
               end
           end

           using .Foo

           foo()
       end

Dict{String,Any} with 1 entry:
  "origin" => "13.55.241.245"

Run a Julia function in the cloud

The @lambda function ... macro creates a local Julia function that passes its arguments and its function body expression to jl_lambda_eval and returns the result. Functions defined this way are not deployed as new Lambda functions (they are executed dynamically by the jl_lambda_eval Lambda) so they are only callable from the Julia program where they are defined.

Multiple invocations of these functions can be run in parallel using Julia tasks. e.g. using asyncmap with ntasks=500 will run 500 Labda invocations in parallel. By default AWS Account concurrency limit is 1000 but this can be increased if needed.

julia> AWSLambda.@lambda function foo(x)
           x = x * 2
           system = chomp(readstring(`uname`))
           return x, system
       end

julia> foo(7)

(14, "Linux")

A function to count primes in the cloud:

julia> AWSLambda.@lambda function count_primes(low::Int, high::Int)

           function is_prime(n)
               if n ≤ 1
                   return false
               elseif n ≤ 3
                   return true
               elseif n % 2 == 0 || n % 3 == 0
                   return false
               end
               i = 5
               while i * i ≤ n
                   if n % i == 0 || n % (i + 2) == 0
                       return false
                   end
                   i += 6
               end
               return true
           end

           c = count(is_prime, low:high)
           println("$c primes between $low and $high.")
           return c
       end
count_primes (generic function with 1 method)

julia> count_primes(10, 100)
21 primes between 10 and 100.

21

Using asyncmap to count primes in parallel:

julia> sum(asyncmap(x->count_primes(x.start, x.stop),
                   [1:1000000, 1000001:2000000, 2000001:3000000]))
78498 primes between 1 and 1000000.

70435 primes between 1000001 and 2000000.

67883 primes between 2000001 and 3000000.

216816

The Base.asyncmap function uses three concurrent tasks to call three instances of count_primes in parallel.

The @lambda function ... macro accepts an optional using ... argument that allows local modules to be used.

e.g. Count primes faster using the Primes.jl package:

julia> AWSLambda.@lambda using Primes function count_primes_fast(low::Int, high::Int)

           c = length(Primes.primes(low, high))
           println("$c primes between $low and $high.")
           return c
       end

julia> count_primes_fast(10, 10^9)
50847530 primes between 10 and 1000000000.

50847530

The @lambda using ... function ... macro bundles up the source files for specified local modules and passess them to jl_lambda_eval along with the function body and arguments each time the function is called.

Deploy a Julia Lambda function

The examples above all rely on the jl_lambda_eval Lambda function to execute Julia code on demand. The code is uploaded, compiled and executed at call time.

Deploying a new named Lambda function will enable your Julia code to be called using the AWS SDK for JavaScript, Python etc. The deployment process also precompiles the Julia code to help speed up execution time.

Deploy a Lambda function to count prime numbers:

julia> AWSLambda.@deploy using Primes function count_primes_fast(low::Int, high::Int)

           c = length(Primes.primes(low, high))
           println("$c primes between $low and $high.")
           return c
       end

The @deploy macro creates a new AWS Lambda named count_primes_fast. It wraps the body of the function with serialisation/deserialistion code and packages it into a .ZIP file along with the source code for the required modules. The .ZIP file is then deployed to AWS Lambda. After deployment the new function should be visible in the AWS Lambda Console.

Use the console to configure a test event as follows, and then use the Test button to invoke the function.

{
  "low": 10,
  "high": 100000000
}

Or invoke the deployed Lambda function from the AWS CLI:

bash$ aws lambda invoke --function-name count_primes_fast \
                    --payload "{\"low\": 10, \"high\": 100}" \
                      output.txt
{
    "ExecutedVersion": "$LATEST",
    "StatusCode": 200
}
$ cat output.txt
"21"

Or from Julia:

julia> AWSLambda.invoke_lambda("count_primes_fast", low = 10, high = 100)

The examples above all pass the low and high arguments to the Lambda function using JSON. The invoke_jl_lambda function uses Julia's built-in serialization mechanism to pass arguments and return values as native Julia objects:

julia> AWSLambda.invoke_jl_lambda("count_pri