Cells ADR 012: Cell Unique Identifier

Context

We will have multiple Cells, each of which has a subset of customer data in them. This means we need a unique identifier for each Cell to Route, Claim, and create sequences.

Cell provisioning tooling already has Tenant ID which is used as a unique identifier for all the infrastructure that is provisioned.

Decision

Introduce the concept of Cell ID, where it will be an incrementing ID starting from 1. This Cell ID will be used across multiple parts of the system:

• Topology Service: Used to classify resources, claim resources by a specific Cell ID, and create Database Sequence.
• Tenant Model: Used to configure the Rails Monolith with the correct Cell ID and enable Cell-specific configuration inside of Instrumentor.
• HTTP Router: Used as identifiable information to route the request to the correct Cell.
• Rails Monolith: When configured it will start claiming globally unique resources to Topology Service and create routable information like tokens

The Cell ID will have the following rules:

• Cell ID 1 will be reserved for the existing legacy Cell, and any new Cell provisioned will increment by 1, meaning the first we will provision will have a Cell ID of 2.
• Cell ID can only be used once, and new provisioned Cell should have a new ID.

Consequences

The Cell ID will need to be propagated through all the services that care about Cells. For this to scale we need a single service that will be the single source of truth for the Cell ID and where it points to.

Topology Service will be the owner and single source of truth of Cell ID, every other system will use Topology Service to see how many Cells we have and the IDs of those Cells.

Alternatives

1. Use Tenant ID as an identifier: This would couple the Rails Monolith, Topology Service, and HTTP Router with the provisioned infrastructure. Making it hard to move the immutable infrastructure to a new tenant model, in case we want to provision a brand new tenant with the same Cell ID.
2. Cell Name: Give a Cell a specific name, this name still needs to be unique and probably derived from an ID anyway. If we were to use a cell name, that would be purely to make it human-readable, but no system would use it as an identifier, since it would be hard to keep the ID and Name in sync. The name of the Cell should follow convention over configuration and follow the format of cell-#{ID}.