Skip to content

custom_runtime.md

Latest commit

 

History

History
61 lines (42 loc) · 2.35 KB

custom_runtime.md

File metadata and controls

61 lines (42 loc) · 2.35 KB

Two approaches are available to prepare a custom container image containing your function code, as explained below.

Custom image (the easy way)

The easiest way to build a custom function image is by leveraging the Serverledge base runtime image, i.e., grussorusso/serverledge-base. This image contains a simple implementation of the Executor server. When the function is invoked, the Executor runs a user-specified command as a new process and sets a few environment variables that may be used by the called process:

  • PARAMS_FILE: path of a file containing JSON-marshaled function parameters
  • RESULT_FILE: name of the file where the function must write its JSON-encoded result
  • CONTEXT: (optional) a JSON-encoded representation of the execution context

You can write a Dockerfile as follows to build your own runtime image, e.g.:

FROM grussorusso/serverledge-base as BASE

# Extend any image you want, e.g.;
FROM tensorflow/tensorflow:latest

# Required: install the executor as /executor
COPY --from=BASE /executor /
CMD /executor

# Required: this is the command representing your function
ENV CUSTOM_CMD "python /function.py"

# Install your code and any dependency, e.g.:
RUN pip3 install pillow
COPY function.py /
# ...

A complete example is provided for C++.

Custom image (the better way)

For higher efficiency, instead of using the default Executor implementation, you can define your image from scratch rewriting/extending the Dockerfile and code used to build Serverledge runtime images. For instance, if you want to create a custom image for a Python-based function, you may write your own Dockerfile based on the content of <ServerledgeRepo>/images/pythonXXX/.

By doing so, you get rid of some process creation overheads, as your function is directly called upon arrival of invocation requests.

Using a custom image

The new function can be created using the CLI interface, setting custom as the desired runtime and specifying the custom_image parameter.

bin/serverledge-cli create -function myfunc -memory 256 -runtime custom -custom_image MY_IMAGE_TAG

Example

The examples/jsonschema directory of the repository provides example files on how to build a custom image for a Python function requiring additional libraries.