When the in-memory MemTable exceeds its size threshold, it must be persisted to disk as an SSTable (Sorted String Table). The SSTable is a fundamental building block of LSM trees, used in RocksDB, Cassandra, LevelDB, and HBase.

SSTable properties:

1. Sorted: keys are stored in lexicographic order, enabling efficient range scans and merge operations
2. Immutable: once written, an SSTable is never modified (append-only philosophy)
3. Bloom filter: a probabilistic data structure attached to each SSTable that answers "is this key possibly in this file?" with no false negatives
4. Sparse index: samples every Nth key to enable fast binary search within the file

The flush process:

1. Freeze the current MemTable (stop writes to it, create a new MemTable for incoming writes)
2. Sort the frozen MemTable entries by key
3. Write them sequentially to a new SSTable file
4. Build the Bloom filter and sparse index
5. Write the footer (Bloom filter + index) and close the file

Request:  {"type": "sstable_flush", "msg_id": 1, "memtable_size_bytes": 4194304}
Response: {"type": "sstable_flush_ok", "in_reply_to": 1, "sstable_file": "L0_001.sst", "entries": 50000, "bloom_filter_bits": 480000, "size_bytes": 4000000}

Request:  {"type": "sstable_lookup", "msg_id": 2, "sstable": "L0_001.sst", "key": "user:42"}
Response: {"type": "sstable_lookup_ok", "in_reply_to": 2, "found": true, "value": "Alice", "bloom_checked": true, "disk_reads": 2}