Several bugs found via deterministic simulation testing (500 seeds, 30s each):
WAL format: Store full WALEntry on disk (including votes and checksum)
instead of a separate WALEntryDisk. This eliminates the need to
reconstruct transient vote bits after restart, and adds wal_update_entry()
for persisting in-place vote modifications.
Crash recovery: Restore 3-case startup logic (empty=normal, corrupt=recovery,
clean=replay). Fix last_normal_view initialization on clean restart. Call
adopt_view() after replay to reinitialize leader vote bits that may be stale
if a crash interrupted wal_replace before the atomic rename.
View change ordering: Move wal_replace before persist_state in
process_begin_view so the on-disk log is always at least as recent as the
persisted metadata. Reset votes on view_change_log entries before wal_replace
in complete_view_change_and_become_primary and complete_recovery.
View adoption: Add adopt_view() helper called from process_prepare and
process_prepare_ok when a NORMAL node learns of a higher view. Updates
last_normal_view and initializes leader vote bits.
Invariant checker: Track node liveness across ticks (prev_alive[]) so
min_commit monotonicity is correctly relaxed for nodes that just restarted.
https://claude.ai/code/session_01X2b5VRntA7LM32kdN9tPAM
On startup, the server now loads view_number, last_normal_view and
commit_index from a durable state file (vsr.state) and replays the
WAL up to the saved commit_index when the log is intact. If the log
is corrupt (checksum failure), the server truncates it and enters
recovery mode to fetch a valid log from peers.
ViewAndCommit follows the same persistence pattern as raft's
TermAndVote: fixed-size record with FNV-1a checksum, written with
fsync on every mutation.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Port the Write-Ahead Log implementation from raft/ into the VSR
server. Log entries are now written to disk with FNV-1a checksums
and fsynced before updating in-memory state. On startup, the WAL
file is loaded and validated, replacing the boot marker for crash
detection. View changes and recovery use atomic rename (tmp.log ->
vsr.log) to replace the WAL safely.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit completes the Write-Ahead Log (WAL) implementation for the
metadata server, including all missing functionality and error fixes:
File System Changes (src/file_system.c):
- Implemented file_set_offset() using lseek (Linux) and SetFilePointer (Windows)
- Implemented file_get_offset() to get current file position
WAL Header Changes (src/wal.h):
- Added file_tree pointer to WAL structure for snapshot serialization
- Added file_path to WAL structure for rotation operations
WAL Implementation (src/wal.c):
- Fixed handle assignment bug in wal_open (was using wal->handle before assignment)
- Fixed missing return statement in append_begin()
- Implemented serialize_callback for writing file tree snapshots to WAL
- Implemented deserialize_callback for reading file tree snapshots from WAL
- Implemented next_entry() to read WAL entries from file during recovery
- Implemented wal_append_write() to write file modifications to WAL
- Implemented swap_file() for complete WAL file rotation:
* Creates temporary file with new snapshot
* Writes WAL header and serialized file tree
* Atomically replaces old WAL file
* Resets entry counter for new rotation cycle
- Added WAL file initialization for newly created files
- Added helper functions: read_exact, read_u8, read_u16, read_u32, read_u64, write_u32
- Fixed typo in comment (Not -> Now)
The WAL now supports:
- Full crash recovery by replaying logged operations
- Automatic file rotation when entry limit is reached
- Atomic file replacement to ensure durability
- Proper file locking throughout rotation process
Claude
cozisandClaude Opus 4.6