fujin/MusicFS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
53 Commits 1 Branch 0 Tags
0ff2a17ab780bfbdacc0d3cbd58bca4ff9eb1d66
Commit Graph

4 Commits

Author SHA1 Message Date
Alexander 0ff2a17ab7 Implement Phase C: Production Hardening 
Implements phase-c-hardening.md to fix 6 RED resilience tests:

- D1/D2: Health check timeout (1.5s) + parallel execution via join_all
- C6: Recursive CAS calculate_size() to scan shard subdirectories
- C7: FUSE read timeout (30s) returns EIO instead of hanging
- 6.4: Auto-re-fetch corrupt/missing chunks from origin
- 6.6: Passthrough mode - continue even when CAS write fails
- C9: PID file with flock prevents concurrent mounts
- 5.3: fd exhaustion handling test

All 27 resilience tests now pass. Full test suite green.

Files changed:
- musicfs-origins/src/health.rs: timeout + join_all
- musicfs-origins/Cargo.toml: add futures dependency
- musicfs-cas/src/store.rs: recursive calculate_size
- musicfs-cas/src/reader.rs: auto-re-fetch on IntegrityError/NotFound
- musicfs-cas/src/fetcher.rs: passthrough fallback
- musicfs-fuse/src/filesystem.rs: 30s read timeout
- musicfs-cli/src/main.rs: PID file with flock
- musicfs-test-utils/tests/resilience.rs: updated tests
 2026-05-13 15:55:22 +02:00
Alexander 5da96ffab2 Implement Phase B: Crash Recovery 
Add startup integrity checks, corruption recovery, CAS size limits,
graceful shutdown orchestration, and a task supervisor — turning 5
previously-RED resilience tests GREEN and adding 5 new tests.

- CAS: pre-check size limit in put(), add StoreFull error variant
- CAS: sled corruption recovery in open() (retry then recreate)
- SQLite: open_with_integrity_check() via PRAGMA integrity_check(1)
- tantivy: open_with_recovery() deletes and rebuilds corrupt index
- CLI: CancellationToken-based ordered shutdown sequence
- Core: TaskSupervisor with spawn_supervised/spawn_critical + backoff
- Tests: replace 4 todo!() stubs, add 5 new shutdown/supervisor tests
 2026-05-13 15:33:23 +02:00
Alexander 6285eeb6c0 Implement Phase A: Stop Dying resilience fixes 
Implements all 6 critical resilience fixes from phase-a-stop-dying.md:

- Issue 2.9: Migrate std::sync::RwLock → parking_lot::RwLock (7 files)
  Prevents lock poisoning cascade on writer panic

- Issue 2.2: Add install_panic_hook() to log panics via tracing
  Ensures panics are captured in logs/journald before process death

- Issue 3.7: Add ExecStopPost to systemd service
  Cleans up stale FUSE mounts on service stop

- Issue 2.7: Add check_stale_mount() detection on startup
  Auto-cleans leftover mounts from previous crashes

- Issue 2.10: Integrate sd_notify for systemd lifecycle
  Sends READY=1 after mount, STOPPING on shutdown

- Issue 2.1: Add signal handling with spawn_mount
  Catches SIGTERM/SIGINT for clean shutdown instead of instant death

All 7 Phase A tests pass:
- test_poisoned_tree_lock_returns_eio_not_panic
- test_parking_lot_rwlock_survives_panic
- test_panic_hook_logs_to_tracing
- test_systemd_service_has_execstoppost
- test_stale_mount_check_function_exists
- test_sd_notify_ready_sent
- test_sigterm_triggers_shutdown
 2026-05-13 14:48:32 +02:00
Alexander e3eeba4650 Add musicfs-test-utils crate with RED resilience tests 
Phase 1 of resilience testing design doc implementation:
- New musicfs-test-utils crate with FaultyOrigin, FaultyCasStore, fixtures
- Failpoints instrumented in musicfs-cas/store.rs
- 16 resilience tests (13 RED for missing features, 3 GREEN for existing)
- 3 Docker/Toxiproxy network tests (RED until docker-compose up)
- docker-compose.yml for Toxiproxy + MinIO + SFTP test infrastructure

Tests properly fail-first (TDD): check_all() sequential, no health timeout,
missing corruption detection, no passthrough mode, etc.
 2026-05-13 13:49:25 +02:00