fix(daemon): prevent zombie state after database file replacement
Adds checkFreshness() to health check paths (GetMetadata, GetConfig, GetAllConfig) and refactors reconnect() to validate new connection before closing old.
PR-URL: https://github.com/steveyegge/beads/pull/1213
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Fixes stack overflow and database initialization failures when running bd on WSL2 with Docker Desktop bind mounts.
## Problem
The bd CLI crashed with stack overflow when running on WSL2 with repositories on Docker Desktop bind mounts (/mnt/wsl/docker-desktop-bind-mounts/...). SQLite WAL mode fails with 'locking protocol' error on these network filesystems.
## Solution
- Expand WAL mode detection to identify Docker bind mounts at /mnt/wsl/* (in addition to Windows paths at /mnt/[a-zA-Z]/)
- Fall back to DELETE journal mode on these problematic paths
- Add comprehensive unit tests for path detection
Fixes GH #1224, relates to GH #920
Co-authored-by: maphew <matt.wilkie@gmail.com>
Auto-detect WSL2 environment with Windows filesystem paths (/mnt/c/, etc.)
and fall back to DELETE journal mode instead of WAL. SQLite WAL mode
does not work reliably across the WSL2/Windows 9P filesystem boundary
due to shared-memory limitations.
Detection checks:
1. Path matches /mnt/[a-z]/ pattern (Windows drive mount)
2. /proc/version contains microsoft or wsl
Fixes#920
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Executed-By: beads/crew/dave
Rig: beads
Role: crew
Add SQLite read-only mode for commands that only query data (list, ready,
show, stats, blocked, count, search, graph, duplicates, comments, export).
Changes:
- Add NewReadOnly() and NewReadOnlyWithTimeout() to sqlite package
- Opens with mode=ro to prevent any file writes
- Skips WAL pragma, schema init, and migrations
- Skips WAL checkpoint on Close()
- Update main.go to detect read-only commands and use appropriate opener
- Skip auto-migrate, FlushManager, and auto-import for read-only commands
- Add tests verifying file mtime is unchanged after read operations
This fixes the issue where file watchers (like beads-ui) would go into
infinite loops because bd list/show/ready modified the database file.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Strip (bd-xxx), (gt-xxx) suffixes from code comments and changelog
entries. The descriptions remain meaningful without the ephemeral
issue IDs.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add CheckOrphanedIssues to detect issues referenced in commits but still open
- Pattern matches (prefix-xxx) in git log against open issues in database
- Reports warning with issue IDs and commit hashes
- Add 8 comprehensive tests for the new check
Also:
- Add tests for mol spawn --attach functionality (bd-f7p1)
- Document commit message convention in AGENT_INSTRUCTIONS.md
- Fix CheckpointWAL to use wrapDBError for consistency
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Change reconnectMu from sync.Mutex to sync.RWMutex so read operations
can hold RLock during database access. This prevents reconnect() from
closing the connection while queries are in progress.
- GetIssue and SearchIssues now hold RLock during database operations
- Close() acquires write lock to coordinate with reconnect
- Add TestConcurrentReadsWithReconnect to verify the fix
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
When git merge replaces the .beads/beads.db file, the daemon's
SQLite connection becomes stale (still reading deleted inode).
This adds FreshnessChecker that detects file replacement via
inode/mtime comparison and triggers automatic reconnection.
Implementation:
- freshness.go: monitors db file for replacement
- store.go: adds EnableFreshnessChecking() and reconnect()
- queries.go: calls checkFreshness() on GetIssue/SearchIssues
- daemon.go: enables freshness checking at startup
- freshness_test.go: comprehensive tests including merge scenario
Code quality (per review):
- Extract configureConnectionPool() helper to reduce duplication
- Handle Close() error in reconnect() (log but continue)
- Use t.Cleanup() pattern in tests per project conventions
- Rename setupFreshnessTest() per naming conventions
Overhead: ~2.6μs per read op (~0.8% of total query time)
Signed-off-by: Alessandro De Blasis <alex@deblasis.net>
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Implements single-shot mode improvements for Windows and Docker scenarios:
- Add --lock-timeout global flag (default 30s, 0 = fail immediately)
- Add config file support: lock-timeout: 100ms
- Parameterize SQLite busy_timeout via NewWithTimeout() function
- In --sandbox mode: default lock-timeout to 100ms
- In --sandbox mode: skip FlushManager creation (no background goroutines)
This addresses bd.exe hanging on Windows and locking conflicts when
using beads across host + Docker containers.
Closes: bd-59er, bd-r4od, bd-dh8a
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
SQLite WAL mode writes go to the -wal file, not the main database.
Without an explicit checkpoint before Close(), writes can be stranded
and lost between CLI invocations.
This was causing `bd migrate` to report success but not actually
persist the version update to the database.
Fixes#434🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
Standardized error handling across the SQLite storage layer by
consistently using wrapDBError() helper functions that were already
defined in errors.go.
Changes:
- config.go: Applied wrapDBError to all config/metadata functions
- queries.go: Fixed bare 'return err' in CreateIssue, UpdateIssue, DeleteIssues
- store.go: Changed %v to %w for proper error chain preservation
- errors_test.go: Added comprehensive test coverage for error wrapping
All error paths now:
- Wrap errors with operation context using %w
- Convert sql.ErrNoRows to ErrNotFound consistently
- Preserve error chains for unwrapping and type checking
This improves debugging by maintaining operation context throughout
the error chain and enables type-safe error checking with sentinel
errors.
All tests passing ✓
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
