[BUG] Snapshot restore may lose data when all data was flushed to SSTables

[ElioNeto]

Description

After per-CF WALs (#157), create_snapshot() writes an empty wal.log when no WAL data exists (all data already flushed). On restore, the engine finds no WAL records and the data appears lost.

Root cause

create_snapshot() saves WAL data before flushing, but if the WAL is already empty (all data flushed), it writes an empty wal.log as a sentinel. restore_snapshot() copies it back, and the engine starts with no data.

The SSTables are also copied, but VersionSet::get() doesn't read on-disk SSTables (see #180), making the restored data invisible.

Severity

Medium — blocks snapshot-based recovery for production use.

Depends on

• [BUG] Point reads always miss for data in on-disk SSTables — VersionSet::get() never opens SstableReader #180 (cold reads from SSTables) — without it, restored SSTables are invisible anyway

[ElioNeto]

Fixed in commit b6ecb48. create_snapshot() now flushes memtables and writes snapshot.manifest with CF-to-SSTable mapping. restore_snapshot() reads manifest and registers SSTables in VersionSet.