Skip to main content
Version: 2.22

Google Cloud Functions

Create a Cloud Function with Python.


Pants can create a Google Cloud Function-compatible zip file or directory from your Python code, allowing you to develop your functions in your repository.

FYI: how Pants does this

Under-the-hood, Pants uses the PEX project, to select the appropriate third-party requirements and first-party sources and lay them out in a zip file or directory, in the format recommended by Google Cloud Functions.

Step 1: Activate the Python Google Cloud Function backend

Add this to your pants.toml:

pants.toml
[GLOBAL]
backend_packages.add = [
"pants.backend.google_cloud_function.python",
"pants.backend.python",
]

This adds the new python_google_cloud_function target, which you can confirm by running pants help python_google_cloud_function

Step 2: Define a python_google_cloud_function target

First, add your Cloud function in a Python file like you would normally do with Google Cloud Functions, such as creating a function def my_handler_name(event, context) for event-based functions.

Then, in your BUILD file, make sure that you have a python_source or python_sources target with the handler file included in the sources field. You can use pants tailor :: to automate this.

Add a python_google_cloud_function target and define the runtime, handler, and type fields. The type should be either "event" or "http". The runtime should be one of the values from https://cloud.google.com/functions/docs/concepts/python-runtime. The handler has the form handler_file.py:handler_func, which Pants will convert into a well-formed entry point. Alternatively, you can set handler to the format path.to.module:handler_func.

For example:

# The default `sources` field will include our handler file.
python_sources(name="lib")

python_google_cloud_function(
name="cloud_function",
runtime="python38",
# Pants will convert this to `project.google_cloud_function_example:example_handler`.
handler="google_cloud_function_example.py:example_handler",
type="event",
)

Pants will use dependency inference based on the handler field, which you can confirm by running pants dependencies path/to:cloud_function. You can also manually add to the dependencies field.

You can optionally set the output_path field to change the generated zip file's path.

Using layout

Use layout to determine whether to build a .zip file or a directory

Use resource instead of file

file / files targets will not be included in the built Cloud Function because filesystem APIs like open() would not load them as expected. Instead, use the resource / resources target. See Assets and archives for further explanation.

Step 3: Run package

Now run pants package on your python_google_cloud_function target to create a zipped file.

For example:

$ pants package project/:cloud_function
Wrote dist/project/cloud_function.zip
Handler: handler
Running from macOS and failing to build?

Cloud Functions must run on Linux, so Pants tells PEX and Pip to build for Linux when resolving your third party dependencies. This means that you can only use pre-built wheels (bdists). If your project requires any source distributions (sdists) that must be built locally, PEX and pip will fail to run.

If this happens, you must either change your dependencies to only use dependencies with pre-built wheels or find a Linux environment to run pants package.

"Encountering collisions" errors and failing to build?

If a build fails with an error like Encountered collisions populating ... from PEX at faas_repository.pex:, listing one or more files with different sha1 hashes, this likely means your dependencies package files in unexpected locations, outside their "scoped" directory (for instance, a package example-pkg typically only includes files within example_pkg/ and example_pkg-*.dist-info/ directories). When multiple dependencies do this, those files can have exactly matching file paths but different contents, and so it is impossible to create a GCF artifact: which of the files should be installed and which should be ignored? Resolving this requires human intervention to understand whether any of those files are important, and hence PEX emits an error rather than making an (arbitrary) choice that may result in confusing and/or broken behaviour at runtime.

Most commonly this seems to happen with metadata like a README or LICENSE file, or test files (in a tests/ subdirectory), which are likely not important at runtime. In these cases, the collision can be worked around by adding a pex3_venv_create_extra_args=["--collisions-ok"] field to the python_google_cloud_function target.

A better solution is to work with the dependencies to stop them from packaging files outside their scoped directories.

Step 4: Upload to Google Cloud

You can use any of the various Google Cloud methods to upload your zip file or directory, such as the Google Cloud console or the Google Cloud CLI.

You must specify the --entry-point as handler. This is a re-export of the function referred to by the handler field of the target.

For example, if using layout="flat":

gcloud functions deploy --source=dist/project/cloud_function --entry-point=handler --trigger-topic=<TOPIC> --runtime=python38 <FUNCTION_NAME>

Advanced: Using PEX directly

In the rare case where you need access to PEX features, such as dynamic selection of dependencies, a PEX file created by pex_binary can be used as a Google Cloud Function package directly. A PEX file is a carefully constructed zip file, and can be understood natively by Google Cloud Functions. Note: using pex_binary results in larger packages and slower cold starts and is likely to be less convenient than using python_google_cloud_function.

The handler of a pex_binary is not re-exported at the fixed main.handler path, and the Google Cloud Function handler must be configured as the __pex__ pseudo-package followed by the handler's normal module path (for instance, if the handler is in some/module/path.py within a source root, then use __pex__.some.module.path). This may require being configured via GOOGLE_FUNCTION_SOURCE. The __pex__ pseudo-package ensures dependencies are initialized before running any of your code.

For example:

python_sources()

pex_binary(
name="gcf",
entry_point="gcf_example.py",
# specify an appropriate platform(s) for the targeted GCF runtime (complete_platforms works too)
platforms=["linux_x86_64-cp39-cp39"],
)

Then, use pants package project:gcf, and upload the resulting project/gcf.pex to Google Cloud Functions. You will need to specify the handler as example_handler and set GOOGLE_FUNCTION_SOURCE=__pex__.gcf_example (assuming project is a source root).

Migrating from Pants 2.16 and earlier

Pants implemented a new way to package Google Cloud Functions in 2.17, which became the only option in 2.19, resulting in smaller packages and faster cold starts. This involves some changes:

  • In Pants 2.16 and earlier, Pants used the Lambdex project. First, Pants would convert your code into a Pex file and then use Lambdex to adapt this to be better understood by GCF by adding a shim handler. This shim handler first triggers the Pex initialization to choose and unzip dependencies, during initialization.
  • In Pants 2.17, the use of Lambdex was deprecated, in favour of choosing the appropriate dependencies ahead of time, as described above, without needing to do this on each cold start. This results in a zip file laid out in the format recommended by GCF, and includes a re-export of the handler.
  • In Pants 2.18, the new behaviour is now the default behaviour.
  • In Pants 2.19 and later, the old Lambdex behaviour has been entirely removed.

If your code can be packaged without warnings using Pants 2.18, no change is required when upgrading to Pants 2.19 (except removing the [lambdex] section in pants.toml if that still remains). If not, follow its instructions to upgrade to Pants 2.18 fully first, and upgrade to Pants 2.19 after that.

If you encounter a bug with the new behaviour, please let us know. If you require advanced PEX features, switch to using pex_binary directly.