Introduce swap journaling crash recovery system with tests.

- Added detailed journaling system (`SwapManager`) for crash recovery, including edit recording and replay.
- Integrated recovery prompts for handling swap files during file open flows.
- Implemented swap file cleanup, checkpointing, and compaction mechanisms.
- Added extensive unit tests for swap-related behaviors such as recovery prompts, file pruning, and corruption handling.
- Updated CMake to include new test files.
This commit is contained in:
2026-02-13 08:44:35 -08:00
parent 895e4ccb1e
commit 2a6ff2a862
19 changed files with 1697 additions and 144 deletions

54
Swap.h
View File

@@ -32,6 +32,18 @@ struct SwapConfig {
// Grouping and durability knobs (stage 1 defaults)
unsigned flush_interval_ms{200}; // group small writes
unsigned fsync_interval_ms{1000}; // at most once per second
// Checkpoint/compaction knobs (stage 2 defaults)
// A checkpoint is a full snapshot of the buffer content written as a CHKPT record.
// Compaction rewrites the swap file to contain just the latest checkpoint.
std::size_t checkpoint_bytes{1024 * 1024}; // request checkpoint after this many queued edit-bytes
unsigned checkpoint_interval_ms{60000}; // request checkpoint at least this often while editing
std::size_t compact_bytes{8 * 1024 * 1024}; // compact on checkpoint once journal grows beyond this
// Cleanup / retention (best-effort)
bool prune_on_startup{true};
unsigned prune_max_age_days{30};
std::size_t prune_max_files{2048};
};
// SwapManager manages sidecar swap files and a single background writer thread.
@@ -45,13 +57,36 @@ public:
void Attach(Buffer *buf);
// Detach and close journal.
void Detach(Buffer *buf);
// If remove_file is true, the swap file is deleted after closing.
// Intended for clean shutdown/close flows.
void Detach(Buffer *buf, bool remove_file = false);
// Reset (truncate-by-delete) the journal for a buffer after a clean save.
// Best-effort: closes the current fd, deletes the swap file, and resumes recording.
void ResetJournal(Buffer &buf);
// Best-effort pruning of old swap files under the swap directory.
// Never touches non-`.swp` files.
void PruneSwapDir();
// Block until all currently queued records have been written.
// If buf is non-null, flushes all records (stage 1) but is primarily intended
// for tests and shutdown.
void Flush(Buffer *buf = nullptr);
// Request a full-content checkpoint record for one buffer (or all buffers if buf is null).
// This is best-effort and asynchronous; call Flush() if you need it written before continuing.
void Checkpoint(Buffer *buf = nullptr);
void SetConfig(const SwapConfig &cfg)
{
std::lock_guard<std::mutex> lg(mtx_);
cfg_ = cfg;
cv_.notify_one();
}
// Obtain a per-buffer recorder adapter that emits records for that buffer.
// The returned pointer is owned by the SwapManager and remains valid until
// Detach(buf) or SwapManager destruction.
@@ -67,6 +102,10 @@ public:
// treat this as a recovery failure and surface `err`.
static bool ReplayFile(Buffer &buf, const std::string &swap_path, std::string &err);
// Compute the swap path for a file-backed buffer by filename.
// Returns empty string if filename is empty.
static std::string ComputeSwapPathForFilename(const std::string &filename);
// Test-only hook to keep swap path logic centralized.
// (Avoid duplicating naming rules in unit tests.)
#ifdef KTE_TESTS
@@ -114,6 +153,10 @@ private:
void RecordJoin(Buffer &buf, int row);
void RecordCheckpoint(Buffer &buf, bool urgent_flush);
void maybe_request_checkpoint(Buffer &buf, std::size_t approx_edit_bytes);
struct JournalCtx {
std::string path;
int fd{-1};
@@ -121,6 +164,9 @@ private:
bool suspended{false};
std::uint64_t last_flush_ns{0};
std::uint64_t last_fsync_ns{0};
std::uint64_t last_chkpt_ns{0};
std::uint64_t edit_bytes_since_chkpt{0};
std::uint64_t approx_size_bytes{0};
};
struct Pending {
@@ -134,16 +180,22 @@ private:
// Helpers
static std::string ComputeSidecarPath(const Buffer &buf);
static std::string ComputeSidecarPathForFilename(const std::string &filename);
static std::uint64_t now_ns();
static bool ensure_parent_dir(const std::string &path);
static std::string SwapDirRoot();
static bool write_header(int fd);
static bool open_ctx(JournalCtx &ctx, const std::string &path);
static void close_ctx(JournalCtx &ctx);
static bool compact_to_checkpoint(JournalCtx &ctx, const std::vector<std::uint8_t> &chkpt_record);
static std::uint32_t crc32(const std::uint8_t *data, std::size_t len, std::uint32_t seed = 0);
static void put_le32(std::vector<std::uint8_t> &out, std::uint32_t v);