This document outlines the process for migrating an existing Exposure Notifications server to the Google Exposure Notifications server. This process applies regardless of whether your Exposure Notifications server uses this codebase.
For purposes of this document, we’ll discuss migration from a state to a national key server.
export-importer: This server contains a job (./cmd/export-importer) that is capable
reading signed export files from other servers and importing those TEKs to the
local storage for re-exporting. We use this in the migration process.
mirror: This server contains a job (./cmd/mirror) that is capable
of mirriong export files from another server onto your CDN. The files to be mirrored
must be available without authentication, and you will not be able to
re-sign those files with a different signing key.
Full migration: The goal of such a migration is to have a state’s client application write to and read from the national key server and to decomission the state’s key server infrastructure.
Read-only migration: The goal of such a migration is to have a state’s client application reading from the national exports, but continuing to upload to the state server. This could be because of a custom verification protocol. In this case, decomissioning the state’s key sever infrastructure is a non goal.
The client / mobile application is live, and requires an update in order to switch servers. This update could be done via dynamic configuration or require a new mobile app version.
There can only be one active export signing key for the application.
The state key server has a text file index of available zip files that is sorted in ascending time order (newest files last). If a state server does not have this, it needs to be added before migration can occur.
The state server must generate export files in a second location (for consuption by the national sever). This is necessary because the state sever will need to continue to produce state-level exports for migration reasons, but will need to keep client-facing exports in their current location.
It is recommended to use a separate signing key for these exports than the one already used for exports destined for clients. This is only a recommendation and is not necessary since the current signing key will be rotated away from client usage.
Configure the export-importer job to import keys from the state server to the national server.
This requres that the public key portion of the export signing key for the state is shared with the national server.
These imported keys will start being shared as part of the national export.
These next actions need to happen as quickly as possible, and as close together as possible.
The state will rotate it’s signing key to be that of the national server by informing Apple and Google of the change. It’s best to request an expedited rotation if available.
Configure the state server to mirror the national exports on the currently known client CDN / download location.
This can be done via the mirror job (./cmd/mirror) located in this repository.
All client apps for the state will now be uploading keys to the state server and downloading keys from the national server. This gives all users full coverage.
This is a terminal state for the read-only migration option.
The state will now push a client application update that directs clients to upload to and download from the national server directly.
When this is done, the state server can be decommissioned and the export-importer job turned down.
If this is being down through an App Store / Play Store application update, this may take a while and it is up to the state to determine what an appropriate amount of time to run in this exprort-import-mirror state is.
Edit this page