Fallback Launcher Release Infrastructure

This document describes the fallback infrastructure that can be enabled to keep launcher updates functioning even if the primary release provider stops working.

Fallback Mechanism in the Launcher
Fallback Infrastructure Specification
- Requirements Inherited from Current Launcher Upgrade Scheme
Current Fallback Infrastructure Implementation
- Updating the Release List
- Marking the Release As Broken

Fallback Mechanism in the Launcher

The launcher has a built-in fallback mechanism for its self-upgrades. If a request to the default release provider (currently GitHub Releases) fails, the fallback mechanism is queried to see if it is enabled and available. If it is enabled, the original request is re-tried with it. So in case the primary release provider becomes permanently disabled, the fallback mechanism can be enabled allowing launcher updates to function without forcing the users to do the update manually.

Currently a fallback mechanism for engine releases is not provided and there are no plans to do so, since if the engine release provider becomes obsolete, a new launcher version can be released that will know about the new engine provider.

Fallback Infrastructure Specification

The launcher assumes that the fallback infrastructure is located at the URL https://launcherfallback.release.enso.org/ and is accessible over HTTPS. This URL should not change to ensure that old versions of the launcher can be upgraded.

Following files are required under that URL:

fallback-manifest.yaml - determines if the fallback mechanism is enabled. Should contain a single key enabled that should be set to true to enable the fallback.
release-list.json - contains a list of all available releases. It should contain a single field releases that is a list of entries. Each entry should have the following fields:
- tag - tag associated with the release
- assets - a list of filenames of assets available in the release, including the broken file if the release was marked as broken.

Moreover, it should provide a subdirectory called launcher. This subdirectory should contain directories for each release from the release list. Each directory is named after the release's tag. This directory should contain the assets listed in the release list. So assets are accessible under the URL https://launcherfallback.release.enso.org/launcher/<release-tag>/<asset-filename>.

So for example, if the release-list.json has the following contents:

{
  "releases": [
    {
      "tag": "0.1.0-test",
      "assets": [
        "broken",
        "launcher-manifest.yaml",
        "enso-launcher-0.1.0-test-linux-amd64.tar.gz"
      ]
    }
  ]
}

The launcher package should be accessible under the URL https://launcherfallback.release.enso.org/launcher/0.1.0-test/enso-launcher-0.1.0-test-linux-amd64.tar.gz.

The fallback infrastructure must be able to provide all released launcher versions.

Requirements Inherited from Current Launcher Upgrade Scheme

The fallback releases must also adhere to the release scheme as it was defined when the first launcher version was released. That brings the following requirements:

The version is a semantic versioning string.
The tag has format enso-<version>.
The release contains an asset called launcher-manifest.yaml which contains at least one key, minimum-version-for-upgrade which is a semantic versioning string.
The release contains at least the following packages:
- enso-launcher-<version>-linux-amd64.tar.gz
- enso-launcher-<version>-macos-amd64.tar.gz
- enso-launcher-<version>-windows-amd64.zip

Current Fallback Infrastructure Implementation

Currently the fallback mechanism is disabled, as it should only be enabled in case the primary release provider becomes permanently unavailable (or at least in case the unavailability is not temporary and is known to span a long enough time-frame that a decision is made that the fallback is necessary).

However, to be sure that we can easily enable it, the infrastructure is already in a working state. It is currently built on top of an AWS S3 bucket which provides the files explained in the specification above. Each release uploaded to GitHub Releases is automatically uploaded to this S3 bucket so it also acts as a backup.

If it is necessary, all that needs to be done to enable the fallback is to set the enabled property in the manifest to true.

Updating the Release List

When a new release is built on the CI, its packages are uploaded to the S3 bucket, but also the release-list.json has to be updated. This is handled by downloading the file, appending to it and re-uploading it. This mechanism is not safe from race conditions if multiple release jobs were running at the same time, so we require that release jobs are invoked serially (and since new releases are not added extremely frequently, this should not be an issue currently). If this ever becomes a problem, a more complex solution can be developed which synchronizes access to the release list.

Marking the Release As Broken

When a release is marked as broken on GitHub (by uploading a file called broken to the assets), a GitHub Action is triggered which uploads the same file to the release and modifies the release-list.json to include the broken asset.

The action uploads the added broken file to the release's files on S3 and updates the release-list.json to include the broken mark in the asset list. The update is done in the same way as described above, so broken marks should also be added one at a time and not at the same time as releases are made to avoid race conditions.

6.3 KiB Raw Blame History