HA Backend Data Model
Refer to the HA Design document for language and current design.
This issue discusses the data model needed to satisfy the HA requirements. Specifically, these requirements are:
- Assign backend nodes to a project
- Identify which backend nodes are degraded for a given project
- Map backend nodes to a storage location
Relational Tables
The existing SQL data model has a projects table which contains a relevant repository_storage field. Currently, repository_storage points to the Gitaly storage location for a project.
The future design intends to map the repository_storage location to a number of backend nodes. This way, a request routed to a specific storage location will transparently be routed to a number of backend storage locations.
The current design is to introduce two new tables:
- projectStorages - junction table that ties together a project and the backend storage nodes. Allows us to track when a backend node becomes degraded (inconsistent) for a given project. This can occur when a write operation fails on a project on a given backend node.
- storages - the actual backend node storage locations and the associated service instance currently operating on it.
╔══════════════════════════════════════════════════════╗
║ Gitaly HA - Data Model ║
║ ║
║ ║
║ ║
║ ┌───────────────────────────────┐ ║
║ │projects │ ║
║ │ - repository_storage (string) │ (already exists) ║
║ │ │ ║
║ └───────────────────────────────┘ ║
║ │ ║
║ │ ║
║ ○ ║
║ ╱│╲ ║
║ ┌───────────────────────────────┐ ║
║ │projectStorages │ ║
║ │ - consistent (bool) │ ║
║ │ │ ║
║ └───────────────────────────────┘ ║
║ ╲│╱ ║
║ ○ ║
║ │ ║
║ │ ║
║ ┌───────────────────────────────┐ ║
║ │storages │ ║
║ │ - PK: backend_storage (string)│ ║
║ │ - Instance UUID (string) │ ║
║ └───────────────────────────────┘ ║
║ ║
╚══════════════════════════════════════════════════════╝ Edited by Paul Okstad (ex-GitLab)