[FEATURE] Key prefix compression — block-level prefix encoding to reduce SSTable size

[ElioNeto]

Description

ApexStore stores full keys in every SSTable block. Competitive engines (RocksDB with prefix_extractor, LevelDB) use prefix compression within data blocks: consecutive keys sharing a common prefix store only the suffix, reducing I/O and storage.

Proposed Implementation

1. During SSTable block building, detect shared prefixes between consecutive keys
2. Store keys as (shared_prefix_len, suffix) instead of full key bytes
3. Add prefix_extractor configuration option
4. When enabled, bloom filters and index blocks can also use prefix-optimized lookups

Impact

• 30-50% SSTable size reduction for keys with common prefixes (e.g., time-series, user IDs)
• Faster scans (fewer bytes to read from disk)
• Better block cache utilization

Labels

[ElioNeto]

Fixed in commit 01211fe. Added prefix compression for SSTable blocks with shared-prefix key encoding.