Server Failover: Design and Implement a Primary-Replica Failover Coordinator

Question

## Round 1 - Coding / System Design

## Problem

Implement a `FailoverCoordinator` that manages a set of servers. One server is the primary; the rest are replicas. If the primary becomes unavailable, the coordinator promotes the most up-to-date replica.

- `register_server(server_id, role, replication_lag)` — add a server with its current lag (in ms).
- `report_failure(server_id)` — mark server as failed.
- `get_primary() -> str` — return current primary id, or trigger failover if primary is down.
- `failover() -> str` — promote the replica with the lowest replication lag as new primary.

```python
class FailoverCoordinator:
    def register_server(self, sid: str, role: str, lag: int) -> None: ...
    def report_failure(self, sid: str) -> None: ...
    def get_primary(self) -> str: ...
    def failover(self) -> str: ...
```

```
register_server("P1", "primary", 0)
register_server("R1", "replica", 50)
register_server("R2", "replica", 20)
report_failure("P1")
get_primary() -> failover() -> "R2"  (lowest lag=20)
```

## Follow-ups

1. What if two replicas have identical lag? How do you break the tie deterministically?
2. After failover, old primary comes back online. How do you reintegrate it without split-brain?
3. How would you detect primary failure automatically (without `report_failure` being called)?
4. Describe the CAP tradeoff your coordinator makes during the failover window.

Server Failover: Design and Implement a Primary-Replica Failover Coordinator

Question Details

Full Details

Topics

More from Databricks