raft-snapshot-repair

Answer

The implementation provides a complete Raft snapshot installation handler in raft_snapshot.go (457 lines) with 17 tests in raft_snapshot_test.go (945 lines). The core design addresses all edge cases:

Snapshot Assembly — Each leader's snapshot transfer is tracked independently via a snapshotAssembly struct keyed by leaderID. Chunks are stored in a map[uint64][]byte keyed by chunk index, enabling out-of-order delivery and natural deduplication (duplicate chunks overwrite with the same data, which is idempotent).

Atomic Log Truncation & State Machine Update — When all chunks are received, the handler: 1. Reassembles snapshot data by concatenating chunks in order 2. Decodes the state machine from the snapshot 3. Atomically (under a single mutex lock) truncates the log, installs the snapshot metadata, replaces the state machine, and advances commit/last-applied indices

Log Conflict Resolution — The log is truncated at the snapshot boundary. Entries strictly after lastIncludedIndex are preserved only if the entry at lastIncludedIndex matches the snapshot's term (indicating consistency). Conflicting entries are discarded, letting the leader re-replicate them.

Edge case handling: - Stale snapshots (older than current snapshot) → return success (already caught up) - Stale leader terms → reject with current term - Zero chunk count → reject immediately - Interrupted transfers → assembly state persists; retransmission from leader resets the assembly (different LastIncludedIndex triggers fresh state) - Multiple concurrent leaders → each leader's transfer is tracked separately by leaderID

// Core handler signature — the heart of the solution
func (rn *RaftNode) HandleInstallSnapshot(req *InstallSnapshotRequest) *InstallSnapshotResponse {
    rn.mu.Lock()
    defer rn.mu.Unlock()
    // 1. Term check
    // 2. Stale snapshot check  
    // 3. Zero chunk count rejection
    // 4. Assembly management (get or create per leader)
    // 5. Chunk storage (idempotent)
    // 6. Wait for all chunks
    // 7. Reassemble and decode
    // 8. Atomic: truncate log + install snapshot + replace state machine
}