Problem

SQLite WAL mode conflicts with BTRFS COW causing lock timeouts and Dashboard/Kanban failures.

Root Cause

• BTRFS COW + SQLite WAL = concurrent write conflicts
• Default busy_timeout too short (1000ms)
• No retry logic for WAL initialization

Solution (tested in hermes_state.py)

• Increased busy_timeout from 1000ms to 30000ms (30 sec)
• Added retry logic: 3 attempts, 1s delay between tries
• Proper exception handling and fallback on failure

Test Results

• 400/400 concurrent operations completed successfully
• 0 errors under load
• Services stable after restart (hermes-gateway, hermes-dashboard)

Patch Location

See: https://github.com/savier89/hermes-btrfs-fix

Files Changed

• hermes_state.py: WAL initialization with busy_timeout + retry loop

Why This Works

BTRFS COW causes temporary file locking during writes. WAL mode needs longer timeouts and retry to handle these transient conflicts. 30s timeout + 3 retries gives enough time for BTRFS to complete COW operations without falling back to DELETE mode.