pomazanbohdan/memory-mcp-1file

🔴 Critical Fix: Docker Crash at Indexing Start

Root Cause: DEFINE INDEX IF NOT EXISTS in SurrealDB does not update an existing HNSW index when the dimension changes. Docker volumes retained old DB with stale dimension, causing a "Max Key length exceeded" crash on first embedding insert.

Fixes

🔧 CRASH FIX: HNSW Vector Index Migration

• schema.surql: IF NOT EXISTS → OVERWRITE for all 4 vector indices
• Indices are always redefined with the current model dimension on startup

🐛 BUG-14: Dynamic Schema Dimension (Closes #4)

• schema.surql uses {dim} placeholder instead of hardcoded 768
• SurrealStorage::new() accepts model_dim and templates the schema at runtime
• check_dimension() now auto-rebuilds indices on mismatch and marks old embeddings as stale, instead of calling process::exit(1)

🐛 BUG-11: Real Storage Flush + Dead Code Removal (Closes #3)

• storage.shutdown() replaced noop RETURN true with real WAL flush via SELECT count() queries
• Removed dead ShutdownCoordinator code (never instantiated in main.rs)

🐛 BUG-5: Idle Timeout (Closes #2)

• Server now auto-exits after idle_timeout minutes of no MCP activity
• Default: 30 min, --idle-timeout 0 to disable
• Implemented as an additional branch in the tokio::select! reconnect loop

⚠️ Upgrade Notes

# Update to v0.2.11 (auto-migrate)
docker pull ... && docker compose up -d

# If issues persist — delete the volume
docker volume rm memory-mcp-data

Changes: 8 files, +77/-101 lines

• src/storage/schema.surql — OVERWRITE + {dim} placeholder
• src/storage/surrealdb.rs — new(model_dim), auto-rebuild, real flush
• src/main.rs — idle timeout, dimension pass-through
• src/lifecycle/shutdown.rs — removed dead ShutdownCoordinator
• src/lifecycle/mod.rs — removed dead export
• Cargo.toml / npm/package.json / Cargo.lock — version bump