Background

The current multi-node setup in test/format mirrors the production environment with a leader and follower running in parallel. This allows us to test:

• Eviction during ingest
• Checkpoint processing
• Leader/follower correctness validation

Predictable replay ensures that both leader and follower apply the same logical operations. However, it does not currently support truncate operations, meaning they are untested in multi-node scenarios.

Problem Statement

• truncate operations are currently not supported under predictable replay
• Multi-node testing coverage for truncate is missing
• Any issues with truncate in a production-like setup may go undetected

Before implementing full support, we need a spike to explore feasibility and potential challenges.

Spike Goals

• Investigate how truncate operations can be captured and replayed in predictable replay
• Identify any limitations
• Estimate effort required to implement full support for truncate in multi-node tests