Multi-Region Raft Clusters

Design

Enable Suture Hub to run as a multi-region Raft cluster for global availability and low-latency reads.

Architecture

                  +----------+
                  |  Region  |
                  |  (Leader)|
                  +----+-----+
                       |
          +------------+------------+
          |            |            |
    +-----+---+  +----+----+  +-----+---+
    | Region 2 |  | Region 3 |  | Region 4 |
    | (Follower)|  | (Follower)|  | (Follower)|
    +----------+  +----------+  +----------+

Raft Protocol Enhancements

1. Multi-region transport: Replace loopback gRPC with TLS-secured

inter-region communication

1. Learner promotion: New regions start as learners, catch up via

snapshot transfer, then become voters

1. Leader election with region awareness: Prefer leader in region

with lowest latency to majority of clients

1. Read delegation: Followers can serve reads with configurable

consistency level (strong, bounded-staleness, eventual)

Configuration

[raft]
cluster_id = "suture-prod"
regions = ["iad", "lhr", "syd"]
region = "iad"  # this node's region
peers = [
    { id = 1, region = "iad", addr = "suture-iad.internal:7000" },
    { id = 2, region = "lhr", addr = "suture-lhr.internal:7000" },
    { id = 3, region = "syd", addr = "suture-syd.internal:7000" },
]
snapshot_interval = 10000
replication_factor = 2

Implementation Phases

Phase 1: Single-region HA (current)

• Multiple nodes in one region
• Leader election, log replication, snapshots
• Status: COMPLETE

Phase 2: Multi-region read replicas

• Add follower-only nodes in other regions
• Read delegation with bounded staleness
• Async snapshot transfer to new regions

Phase 3: Multi-region voting

• Full voting members in multiple regions
• TLS-secured inter-region gRPC
• Region-aware leader election