We're happy you want to write a Python sample! Like a lot of Pythonistas, we're opinionated and fussy. This guide is a reference for the format and style expected of samples contributed to the python-docs-samples repo. The guidelines below are intended to ensure that all Python samples meet the following goals:

• Copy-paste-runnable. A developer should be able to copy and paste the code into their own environment and run it with as few modifications as possible.
• Teach through code. Each sample should demonstrate best practices for interacting with Google Cloud libraries, APIs, or services.
• Idiomatic. Each sample should follow widely accepted Python best practices as covered below.

We recommend referencing the following samples and sample tests:

• Storage client samples

See Folder Location. Samples live in this repository, python-docs-samples, or in a library repository.

Python libraries repositories live in https://github.com/googleapis/ in repositories named python-API. Each repository contains one library. For example, https://github.com/googleapis/python-bigquery contains the google-cloud-bigquery library.

This is a work in progress - in python-docs-samples, your PR will automatically be assigned to one of the owners in @GoogleCloudPlatform/python-samples-owners. You can assign a new person using the blunderbuss:assign label if your assignee is OOO or busy. You can (and probably should) also assign a teammate in addition to the auto-assigned owner to review your code for product-specific needs.

In library repositories with pre-existing samples GitHub will automatically assign a reviewer from python-samples-owners. If no reviewer is automatically assigned, contact @googleapis/python-samples-owners.

Please reach out to your assigned reviewer if it's been more than 2 days and you haven't gotten a response!

You should install the latest patch version of each minor version listed in Python Versions.

We recommend using the Python version management tool Pyenv if you are using MacOS or Linux.

Googlers: See the internal Python policies doc.

Using MacOS?: See Setting up a Mac development environment with pyenv and pyenv-virtualenv.

Afterwards, see Test Environment Setup.

This section covers guidelines for Python samples. Note that Testing Guidelines are covered separately below.

Samples that primarily show the use of one client library should be placed in the client library repository googleapis/python-{api}. Other samples should be placed in this repository python-docs-samples.

Library repositories: Each sample should be in a folder under the top-level samples folder samples in the client library repository. See the Text-to-Speech samples for an example.

python-docs-samples: Each sample should be in a folder under the top-level folder of python-docs-samples that corresponds to the Google Cloud service or API used by the sample. For example, a sample demonstrating how to work with Composer should be in a subfolder under the python-docs-samples/composer folder.

Conceptually related samples under a service or API should be grouped into a subfolder. For example, App Engine Standard samples are under the appengine/standard folder, and App Engine Flex samples are under the appengine/flexible folder.

If your sample is a set of discrete code snippets that each demonstrate a single operation, these should be grouped into a snippets folder. For example, see the snippets in the bigtable/snippets/writes folder.

If your sample is a quickstart — intended to demonstrate how to quickly get started with using a service or API — it should be in a quickstart folder.

Samples should support Python 3.6, 3.7, 3.8, and 3.9.

If the API or service your sample works with has specific Python version requirements different from those mentioned above, the sample should support those requirements.

Source code files should always begin with an Apache 2.0 license header. See the instructions in the repo license file on how to apply the Apache license to your work. For example, see the license header for the Datastore client quickstart sample.

If, and only if, your sample application is a command-line application, then include a shebang as the first line. Separate the shebang line from the rest of the application with a blank line. The shebang line for a Python application should always be:

#!/usr/bin/env python

Don't include shebang lines in web applications or test files.

All Python samples should follow the best practices defined in the PEP 8 style guide and the Google Python Style Guide. The automated linting process for Python samples uses flake8 to verify conformance to common Python coding standards, so the use of flake8 is recommended.

If you prefer to use pylint, note that Python samples for this repo are not required to conform to pylint’s default settings outside the scope of PEP 8, such as the “too many arguments” or “too many local variables” warnings.

The use of Black to standardize code formatting and simplify diffs is recommended, but optional.

The default noxfile has blacken session for convenience. Here are some examples.

If you have pyenv configured:

nox -s blacken

If you only have docker:

cd proj_directory
../scripts/run_tests_local.sh . blacken

In addition to the syntax guidelines covered in PEP 8, samples should strive to follow the Pythonic philosophy outlined in the PEP 20 - Zen of Python as well as the readability tenets presented in Donald Knuth's Literate Programming. Notably, your sample program should be self-contained, readable from top to bottom, and fairly self-documenting. Prefer descriptive names, and use comments and docstrings only as needed to further clarify the code’s intent. Always introduce functions and variables before they are used. Prefer less indirection. Prefer imperative programming as it is easier to understand.

Very few samples will require authoring classes. Prefer functions whenever possible. See this video for some insight into why classes aren't as necessary as you might think in Python. Classes also introduce cognitive load. If you do write a class in a sample, be prepared to justify its existence during code review.

Always prefer descriptive function names, even if they are long. For example upload_file, upload_encrypted_file, and list_resource_records. Similarly, prefer long and descriptive parameter names. For example source_file_name, dns_zone_name, and base64_encryption_key.

Here's an example of a top-level function in a command-line application:

def list_blobs(bucket_name):
    """Lists all the blobs in the bucket."""
    storage_client = storage.Client()
    bucket = storage_client.get_bucket(bucket_name)

    blobs = bucket.list_blobs()

    for blob in blobs:
        print(blob.name)

Notice the simple docstring and descriptive argument name (bucket_name implying a string instead of just bucket which could imply a class instance).

This particular function is intended to be the "top of the stack" - the function executed when the command-line sample is run by the user. As such, notice that it prints the blobs instead of returning. In general, top of the stack functions in command-line applications should print, but use your best judgment.

Here's an example of a more complicated top-level function in a command-line application:

def download_encrypted_blob(
        bucket_name, source_blob_name, destination_file_name,
        base64_encryption_key):
    """Downloads a previously-encrypted blob from Google Cloud Storage.

    The encryption key provided must be the same key provided when uploading
    the blob.
    """
    storage_client = storage.Client()
    bucket = storage_client.get_bucket(bucket_name)
    blob = bucket.blob(source_blob_name)

    # Encryption key must be an AES256 key represented as a bytestring with
    # 32 bytes. Since it's passed in as a base64 encoded string, it needs
    # to be decoded.
    encryption_key = base64.b64decode(base64_encryption_key)

    blob.download_to_filename(
        destination_file_name, encryption_key=encryption_key)

    print(f'Blob {source_blob_name} downloaded to {destination_file_name}.'

Note the verbose parameter names and the extended description that helps the user form context. If there were more parameters or if the parameters had complex context, then it might make sense to expand the docstring to include an Args section such as:

Args:
    bucket_name: The name of the cloud storage bucket.
    source_blob_name: The name of the blob in the bucket to download.
    destination_file_name: The blob will be downloaded to this path.
    base64_encryption_key: A base64-encoded RSA256 encryption key. Must be the
        same key used to encrypt the file.

Generally, however, it's rarely necessary to exhaustively document the parameters this way. Lean towards unsurprising arguments with descriptive names, as having to resort to this kind of docstring might be extremely accurate but it comes at the cost of high redundancy, signal-to-noise ratio, and increased cognitive load.

Argument types should be documented using Python type annotations as introduced in PEP 484. For example:

def hello_world(name: str) -> None:
    print(f"Hello {name}!")

def adder(a: int, b: int) -> int:
    return a+b

Type hinting is enforced using flake8-annotations, which is enabled by setting the enforce_type_hints variable to True in the appropriate noxfile_config.py. Type hinting is expected in all new samples, and will gradually be added to all compatible existing samples.

If there is an Args section within the function's docstring, consider documenting the argument types there as well. For example:

Args:
    credentials (google.oauth2.credentials.Credentials): Credentials
      authorized for the current user.

When documenting primitive types, be sure to note if they have a particular set of constraints. For example, A base64-encoded string or Must be between 0 and 10.

Always create timezone aware datetime objects. For libraries that use protobuf, omitting the timezone may lead to unexpected behavior when the datetime is converted to a protobuf timestamp.

import datetime

now = datetime.datetime.now(tz=datetime.timezone.utc)

For more information see the Python datetime documentation.

Each sample should have a README.md file that provides instructions for how to install, configure, and run the sample. Setup steps that cover creating Google Cloud projects and resources should link to appropriate pages in the Google Cloud Documentation, to avoid duplication and simplify maintenance.

Every sample should include a requirements.txt file that lists all of its dependencies, to enable others to re-create the environment that was used to create and test the sample. All dependencies should be pinned to a specific version, as in this example:

Flask==1.1.1
PyMySQL==0.9.3
SQLAlchemy==1.3.12

If a sample has testing requirements that differ from its runtime requirements (such as dependencies on pytest or other testing libraries), the testing requirements may be listed in a separate requirements-test.txt file instead of the main requirements.txt file.

Pip has VCS support. Use the branch name or commit hash instead of the package name.

pip install:

pip install git+https://github.com/googleapis/python-firestore.git@ee518b741eb5d7167393c23baa1e29ace861b253

requirements.txt:

Flask==1.1.1
PyMySQL==0.9.3
git+https://github.com/googleapis/python-firestore.git@ee518b741eb5d7167393c23baa1e29ace861b253

Sample code may be integrated into Google Cloud Documentation through the use of region tags, which are comments added to the source code to identify code blocks that correspond to specific topics covered in the documentation. For example, see this sample — the region tags are the comments that begin with [START or [END.

The use of region tags is beyond the scope of this document, but if you’re using region tags they should start after the source code header (license/copyright information), but before imports and global configuration such as initializing constants.

Sample code should use standard Python exception handling techniques as covered in the Google Python Style Guide.

Samples should include tests to verify that the sample runs correctly and generates the intended output. Follow these guidelines while writing your tests:

• Use pytest-style tests and plain asserts. Don't use unittest-style tests or assertX methods.
• Whenever possible, tests should allow for future changes or additions to APIs that are unrelated to the code being tested. For example, if a test is intended to verify a JSON payload returned from an endpoint, it should only check for the existence of the expected keys and values, and the test should continue to work correctly if the order of keys changes or new keys are added to the response in a future version of the API. In some cases, it may make sense for tests to simply verify that an API call was successful rather than checking the response payload.
• Samples that use App Engine Standard should use the App Engine testbed for system testing, as shown in this example.
• All tests should be independent of one another and order-independent.
• We use parallel processing for tests, so tests should be capable of running in parallel with one another.
• Use pytest's fixture for resource setup and teardown, instead of having them in the test itself.
• Avoid infinite loops.
• Retry RPCs

Tests for samples should follow the “Arrange, Act, Assert” structure:

• Arrange — create and configure the components required for the test. Avoid nesting; prioritize readability and simplicity over efficiency. For Python tests, typical "arrange" steps include imports, copying environment variables to local variables, and so on.
• Act — execute the code to be tested, such as sending a request to an API and receiving a response.
• Assert — verify that the test results match what is expected, using an assert statement.

Whenever possible, tests should run against the live production version of cloud APIs and resources. This will assure that any breaking changes in those resources are identified by the tests.

External resources that must exist prior to the test (for example, a Cloud SQL instance) should be identified and passed in through an environment variable. If specific data needs to exist within such infrastructure resources, however, the test should create this data as part of its Arrange steps and then clean up when the test is completed.

Creating mocks for external resources is strongly discouraged. Tests should verify the validity of the sample against the APIs, and not against a mock that embodies assumptions about the behavior of the APIs.

When tests need temporary resources (such as a temp file or folder), they should create reasonable names for these resources with a UUID attached to assure uniqueness. Use the Python uuid package from the standard library to generate UUIDs for resource names. For example:

glossary_id = f'test-glossary-{uuid.uuid4()}'

or:

# If full uuid4 is too long, use its hex representation.
encrypted_disk_name = f'test-disk-{uuid.uuid4().hex}'

# If the hex representation is also too long, slice it.
encrypted_disk_name = f'test-disk-{uuid.uuid4().hex[:5]}'

All temporary resources should be explicitly deleted when testing is complete. Use pytest's fixture for cleaning up these resouces instead of doing it in test itself.

If the sample prints output to the console, the test should capture stdout to a file and verify that the captured output contains the key information that is expected. Strive to verify the content of the output rather than the syntax. For example, the test might verify that a string is included in the output, without taking a dependency on where that string occurs in the output.

Never put potential infinite loops in the test code path. A typical example is about gRPC's LongRunningOperations. Make sure you pass the timeout parameter to the result() call.

Good:

# will raise google.api_core.GoogleAPICallError after 60 seconds
operation.result(60)

Bad:

operation.result()  # this could wait forever.

We recommend the timeout parameter to be around the number that gives you more than 90% success rate. Don't put too long a timeout.

Now this test is inevitably flaky, so consider marking the test as flaky as follows:

@pytest.mark.flaky(max_runs=3, min_passes=1)
def my_flaky_test():
    # test that involves LRO poling with the timeout

This combination will give you very high success rate with fixed test execution time (0.999 success rate and 180 seconds operation wait time in the worst case in this example).

All the RPCs are inevitably flaky. It can fail for many reasons. The google-cloud Python client retries requests automatically for most cases.

The old api-client doesn't retry automatically, so consider using backoff for retrying. Here is a simple example:

import backoff
from googleapiclient.errors import HttpError

@pytest.fixture(scope='module')
def test_resource():
    @backoff.on_exception(backoff.expo, HttpError, max_time=60)
    def create_resource():
        try:
            return client.projects().imaginaryResource().create(
                name=resource_id, body=body).execute()
        except HttpError as e:
            if '409' in str(e):
                # Ignore this case and get the existing one.
                return client.projects().imaginaryResource().get(
                    name=resource_id).execute()
            else:
                raise

    resource = create_resource()

    yield resource

    # cleanup
    ...

When writing a test for a list method, consider filtering the possible results. Listing all resources in the test project may take a considerable amount of time. The exact way to do this depends on the API.

Some list methods take a filter/filter_ parameter:

from datetime import datetime

from google.cloud import logging_v2

client = logging_v2.LoggingServiceV2Client()
resource_names = [f"projects/{project}"]
   # We add timestamp for making the query faster.
    now = datetime.datetime.now(datetime.timezone.utc)
    filter_date = now - datetime.timedelta(minutes=1)
    filters = (
        f"timestamp>=\"{filter_date.isoformat('T')}\" "
        "resource.type=cloud_run_revision "
        "AND severity=NOTICE "
)

entries = client.list_log_entries(resource_names, filter_=filters)

Others allow you to limit the result set with additional arguments to the request:

from google.cloud import asset_v1p5beta1

# TODO project_id = 'Your Google Cloud Project ID'
# TODO asset_types = 'Your asset type list, e.g.,
# ["storage.googleapis.com/Bucket","bigquery.googleapis.com/Table"]'
# TODO page_size = 'Num of assets in one page, which must be between 1 and
# 1000 (both inclusively)'

project_resource = "projects/{}".format(project_id)
content_type = asset_v1p5beta1.ContentType.RESOURCE
client = asset_v1p5beta1.AssetServiceClient()

# Call ListAssets v1p5beta1 to list assets.
response = client.list_assets(
    request={
        "parent": project_resource,
        "read_time": None,
        "asset_types": asset_types,
        "content_type": content_type,
        "page_size": page_size,
    }
)

Because all tests are system tests that use live resources, running tests requires a Google Cloud project with billing enabled, as covered under Creating and Managing Projects.

Once you have your project created and configured, you'll need to set environment variables to identify the project and resources to be used by tests. See testing/test-env.tmpl.sh for a list of all environment variables that must be set manually. Not every test needs all of these variables. All required environment variables are listed in testing/test-env.tmpl.sh. If you need to add a new secret, follow instructions in Secrets.

We suggest that you copy this file as follows:

$ cp testing/test-env.tmpl.sh testing/test-env.sh
$ editor testing/test-env.sh  # change the value of `GCLOUD_PROJECT`.

You can easily source this file for exporting the environment variables.

This repository supports two ways to run tests locally.

1. nox

   This is the recommended way. Setup takes little more efforts than the second one, but the test execution will be faster.

2. Docker

   This is another way of running the tests. Setup is easier because you only need to instal Docker. The test execution will be bit slower than the first one. This option is also useful if you need to simulate the CI system.

Automated testing for samples is managed by nox. Nox allows us to run a variety of tests, including the flake8 linter, Python 2.7, Python 3.x, and App Engine tests, as well as automated README generation.

Sample tests are run through pytest. Do not use unittest.

Note:

Library repositories: If you are working on an existing project (meaning that a samples directory already exists), a noxfile.py will already exist within that samples directory.

For new samples, create a new noxfile.py and paste the contents of noxfile-template.py. Note - there may be a noxfile.py in the repo already in the root directory, but this is used for testing the libraries, not the samples, so you will still need to make a samples noxfile.

python-docs-samples: As a temporary workaround, each project currently uses first noxfile-template.py found in a parent folder above the current sample. In order to simulate this locally, you need to copy + rename the parent noxfile-template.py as noxfile.py in the folder of the project (containing the requirements.txt for the file).

cd python-docs-samples
cp noxfile-template.py PATH/TO/YOUR/PROJECT/noxfile.py
cd PATH/TO/YOUR/PROJECT/

To use nox, install it globally with pip:

$ pip install nox

To run style checks on your samples:

nox -s lint

To run tests with a python version, use the correct py-3.* sessions:

nox -s py-3.6

To run a specific file:

nox -s py-3.7 -- snippets_test.py

To run a specific test from a specific following:

nox -s py-3.7 -- snippets_test.py:test_list_blobs

The noxfile_config.py allows for customization of some options:

• Ignore specific Python versions.
• Enforce type hints.
• Specify a different Google Cloud Project.
• Add additional environment variables. Also see Environment Variables.
• Override the version of pip used by nox

Options are documented inside the noxfile_config.py.

Note: This is currently only available for samples in python-docs-samples.

If you have Docker installed and runnable by the local user, you can use scripts/run_tests_local.sh helper script to run the tests. For example, let's say you want to modify the code in cdn directory, then you can do:

$ cd cdn
$ ../scripts/run_tests_local.sh .
# This will run the default sessions; lint, py-3.6, and py-3.7
$ ../scripts/run_tests_local.sh . lint
# Running only lint

If your test needs a service account, you have to create a service account and download the JSON key to testing/service-account.json.

On MacOS systems, you also need to install coreutils to use scripts/run_tests_local.sh. Here is how to install it with brew:

$ brew install coreutils

This section explains how to set environment variables that are needed by tests.

If a noxfile_config.py does not exist, copy noxfile_config.py into the directory.

Add the new environment variables to the envs dictionary.

TEST_CONFIG_OVERRIDE = {
    # You can opt out from the test for specific Python versions.
    "ignored_versions": ["2.7"],
    # Old samples are opted out of enforcing Python type hints
    # All new samples should feature them
    "enforce_type_hints": True,
    # An envvar key for determining the project id to use. Change it
    # to 'BUILD_SPECIFIC_GCLOUD_PROJECT' if you want to opt in using a
    # build specific Cloud project. You can also use your own string
    # to use your own Cloud project.
    "gcloud_project_env": "GOOGLE_CLOUD_PROJECT",
    # 'gcloud_project_env': 'BUILD_SPECIFIC_GCLOUD_PROJECT',
    # A dictionary you want to inject into your test. Don't put any
    # secrets here. These values will override predefined values.
    "envs": {"DJANGO_SETTINGS_MODULE": "mysite.settings"},
}

For setting up a local test environment, see Test Environment Setup.

Secrets (e.g., project names, API keys, passwords) are kept in Cloud Secret Manager. See python-docs-samples-test-env. If you are unable to access the link, reach out to your assigned pull request reviewer or someone in @GoogleCloudPlatform/python-samples-owners for assistance.

1. Add the new environment variable to testing/test-env.tmpl.sh in your pull request.
2. Run scripts/decrypt-secrets.sh to fetch the secrets. A new file testing/test-env.sh will appear.
3. Add the new environment variable to testing/test-env.sh.
4. Run scripts/encrypt-secrets.sh to upload the secrets to secret manager.

Certain samples require integration with Google Cloud Storage (GCS), most commonly for APIs that read files from GCS. To run the tests for these samples, configure your GCS bucket name via the CLOUD_STORAGE_BUCKET environment variable.

The resources required by tests can usually be found in the ./resources folder inside the samples/snippets directory in client libraries, as in this example. You can upload those resources to your own GCS bucket to run the tests with gsutil. For example:

gsutil cp ./samples/snippets/resources/* gs://{$CLOUD_STORAGE_BUCKET}/

Yes, you can use pdb or any Python debugger. For pdb, use import pdb; pdb.set_trace() (<3.7) or breakpoint (3.7+). See https://docs.python.org/3/library/pdb.html.

These IDEs just inject the breakpoint above into the code, so it should work.