firefox-translations-models/remote_settings/README.md

7.4 KiB

Firefox Translations Models: remote_settings

A CLI tool to upload language models to Remote Settings.

When to use remote_settings

The remote_settings CLI tool should be used when language models are added or updated within the repository to publish them to Remote Settings.

Example: Publish a language pair to Remote Settings

1) Export your authentication token (see Authentication)

export REMOTE_SETTINGS_BEARER_TOKEN="Bearer ..."`

2) Ensure your files are present locally, and not stored remotely via git-lfs

git lfs fetch --all
git lfs pull {path_to_your_files}/*

[!NOTE] Requires installing git-lfs.

3) Unzip the model files you wish to publish

gzip -d {path_to_your_files}/*

[!NOTE] Requires installing gzip.

4) Inspect the metadata of your to-be-published records by using the --dry-run flag

First install the project dependencies.

poetry install --no-root

Then run the command using the following arguments:

poetry run python -m remote_settings create \
  --dry-run \
  --lang-pair {lang_pair} \
  --server {dev,stage,prod} \
  --version {version}

5) Publish the model records for a language pair by removing the --dry-run flag

poetry run python -m remote_settings create \
  --lang-pair {lang_pair} \
  --server {dev,stage,prod} \
  --version {version}

[!NOTE] At this time versions are manually entered by the CLI user.

For example, if you are publishing records for a version 1.0a2, and 1.0a1 already exists, you will need to input the version 1.0a2 manually.

In the future, we would like to support one or many of the following:

  1. Bumping versions relatively via the CLI, e.g. --version {bump-alpha,bump-beta,bump-minor,bump-major}

  2. Retrieving version data elsewhere, possibly from the model names, paths, or JSON data generated alongside the models.

Creating model records and attachments

The create subcommand is used to upload a new model to the Remote Settings.

Examples

When uploading a new record, it should be uploaded to Nightly only via the 1.0a1 version. To do this run the create subcommand to add it to the Remote Settings prod server. The record will still need to be approved via Remote Settings, but you can stage the changes for testing in Nightly using the Remote Settings Devtools.

poetry run python -m remote_settings create \
  --lang-pair esen \
  --server prod \
  --version 1.0a1

Create a record and attachment for esen/vocab.esen.spm on the dev server with a version of 1.0, but mock the server connection to inspect the record data and configuration without uploading.

poetry run python -m remote_settings create \
  --path models/prod/esen/vocab.esen.spm \
  --server dev \
  --version 1.0 \
  --mock-connection

Create a record and attachment for esen/vocab.esen.spm on the prod server with a version of 1.0a1, but only do a dry run to inspect the record data, configuration, and server authentication without uploading.

poetry run python -m remote_settings create \
  --path models/prod/esen/vocab.esen.spm \
  --server prod \
  --version 1.0a1 \
  --dry-run

Create a record and attachment for esen/vocab.esen.spm on the stage server with a version of 2.1b2.

poetry run python -m remote_settings create \
  --path models/prod/esen/vocab.esen.spm \
  --server stage \
  --version 2.1b2

Authentication

To authenticate your user with the server, you must export you token as an environment variable called REMOTE_SETTINGS_BEARER_TOKEN.

You can retrieve a bearer token from the Remote Settings admin dashboards.

On the top right corner, use the 📋 icon to copy the authentication string

Required arguments

There are three required arguments for the script to run:

  • --server - The server to which the record and attachment will be uploaded.
  • --version - The semantic version of the record and the attachment.

The final argument is either of the following:

  • --path - The path to a single file attachment to upload.
  • --lang-pair - The language pair for which to upload all file attachments.

Arg: --server

Determines which Remote Settings Server will receive the uploaded record.

Models can be uploaded to one of three servers using the --server flag:

  • --server dev
  • --server stage
  • --server prod

[!NOTE] Uploading to the prod or stage servers require VPN access, which can be acquired by following the steps here.

Arg: --version

Applies a semantic version to the record.

Record versions follow typical semantic versioning used throughout Firefox, and records will be made available only in certain channels based on their version.

  • Alpha-version records, e.g. --version 1.0a1 will be available only in local builds and nightly builds.
  • Beta-version records, e.g. --version 1.0b1 will be available in all builds except release builds.
  • Release-version records, e.g. --version 1.0 will be available in all builds.

Arg: --path

Uploads a single record and attachment located at the provided path.

Model attachment files are stored in the models directory at the root of the repository.

The remote_settings script derives metadata from the name of the file itself.

For example, the file named trgvocab.esen.spm will by of type trgvocab with a from-language of es and a to-language of en.

[!NOTE] Files are stored in compressed gzip archives. They must be decompressed before uploading.

Arg: --lang-pair

Uploads all file attachments in the directory associated with the given language pair.

This argument will take a language pair, e.g. "enes" and upload all files in the relevant path.

The path itself is based on the provided --version argument.

  • If the version is a release version, e.g. 1.0, the script will search for the language pair in the models/prod directory.
  • If the version is a pre-release version, e.g. 1.0a1, the script will search for the language pair in the models/dev directory.

[!NOTE] Files are stored in compressed gzip archives. They must be decompressed before uploading.

Args: --mock-connection and --dry-run

Before uploading a model, it can be useful to test your configuration before committing.

The remote_settings script offers two levels of preparation:

  • -m | --mock-connection
  • -d | --dry-run

The --mock-connection flag mocks the user authentication and connection to the server, allowing you to inspect the derived metadata of your record and attachment.

The --dry-run flag will attempt to connect and authenticate with the server as well as output the metadata, but it will not upload the record and attachment.

Using these flags is recommended to ensure that your setup is correct before you upload attachments to Remote Settings.

Testing

Tests can be run via the following command:

poetry run python -m pytest tests