Serverless Research Kit

Serverless Research Kit (SRK) exists to accelerate research progress in serverless computing. It aims to make it easy to innovate on services typically offered by cloud providers by including, for example, readily hackable versions of cloud functions or cloud object storage. SRK also plans to include common benchmarks and operational tools, so that launching and evaluating a multi-tenant serverless service is quick and easy.

Installation

SRK requires a number of runtime files in order to operate. These are all contained in the runtime/ directory. You may optionally use the 'install.sh' script to place these files in a global location (defaults to ~/.srk). In either case, you should set the SRKHOME environment variable to the absolute path of whatever directory you'd like to use for these files (you can set it to PATH/TO/SRKREPO/runtime to always use the most up to date files when developing srk). In this case we'll set it to the repo's version of runtime:

$ ./install.sh
Please specify an install location (or press enter to default to ~/.srk)
./runtime
SRK installed to /PATH/TO/REPO/runtime
Please add /PATH/TO/REPO/runtime/config.yaml and configure for your needs
You should add "export SRKHOME=/PATH/TO/REPO/runtime" to your .bashrc or equivalent
$ export SRKHOME=$(pwd)/runtime

You may also want to add your srk repo to PATH or copy the built binary to a location on your path.

Configuration

Wherever you installed SRK, you must now configure it to use different providers. You can start by copying $SRKHOME/example-config.yaml to $SRKHOME/config.yaml. You can then edit that file to configure your services:

• OpenLambda needs to know where you OL command and working directories are.
• AWS needs to know your role and vpc/security group configurations

You should also set the 'default-provider' to either 'aws' or 'local' (local uses openlambda).

Build

To build this project, just run:

go build

Examples

Echo (Hello World)

The echo function is defined at examples/echo. It serves as a "Hello World" for FaaS.

Source Files

• echo.py: This is our actual function logic, it can be anything you want as long as all of it's dependencies are in the echo/ folder that we pass to srk function create -s
• f.py: This is the open-lambda glue code. OL requires that functions be named f() and live in f.py. The signature of f() must be preserved. The body of f() does any OL-specific actions and calls echo() with the correct arguments.
• aws.py: This is the aws Lambda glue code. AWS lambda is more flexible with function naming, but has an incompatible signature requirement (it includes a 'context' field that OL doesn't). SRK requires that you include an aws.py with a function f(event, context).

Installation

To install to the configured service, run:

./srk function create --source examples/echo

Invocation

This example simply echos back any arguments you pass it. To do a simple invocation, we will use the 'one-shot' benchmark that simply sends a single request and prints a single response:

./srk bench \
    --bench one-shot \
    --function-args '{"hello" : "world"}' \
    --function-name echo

You should see {"hello" : "world"} printed back to you as the function response.

Concurrency Sweep Benchmark

NOTE This section does not currently work (but will soon)

A more advanced benchmark attempts to measure the function invocation scaling of different providers.

./srk function create \
  --source examples/cfbench/sleep_workload.py \
  --include cfbench

Install it as an AWS Lambda function:

aws lambda create-function \
  --function-name sleepworkload \
  --runtime python3.7 \
  --handler sleep_workload.lambda_handler \
  --timeout 10 \
  --role {{YOUR_ROLE_ARN_HERE}} \
  --zip-file fileb://$(pwd)/build/functions/sleep_workload.zip \
  --memory-size 128 \
  --vpc-config SubnetIds={{YOUR_SUBNET_IDS_HERE}},SecurityGroupIds={{YOUR_SECURITY_GROUP_IDS_HERE}}

In the command above you will need to make some substitutions. {{YOUR_ROLE_ARN_HERE}} should look something like arn:aws:iam::123459789012:role/service-role/my-service-role-ae04d032, {{YOUR_SUBNET_IDS_HERE}} should look something like subnet-dd045605a058b8946,subnet-e56ceb1a1832684a4, and {{YOUR_SECURITY_GROUP_IDS_HERE}} should look something like sg-4cf9dbb40b73ca192,sg-016fb0eb84b2f3fee. You may also need to set the region, e.g., with a command like export AWS_DEFAULT_REGION=us-west-2.

Now you can run a command like this to test the cloud function:

./srk bench \
  --mode concurrency-scan \
  --function-name sleepworkload \
  --function-args '{"sleep_time_ms":5000}' \
  --params '{"begin_concurrency":1,"delta_concurrency":1,"num_steps":5,"step_duration":5}'

You can also view the example test function.