Struct WriteCoordinator 

pub struct WriteCoordinator { /* private fields */ }

The WriteCoordinator serializes the validation + commit_seq allocation + marker append section of the native commit protocol (§7.11.2).

The serialized section (steps 2-8) MUST NOT observe cancellation once commit_seq is allocated. The coordinator MUST NOT write page payloads in the serialized section — only marker + proof.

Implementations

impl WriteCoordinator

pub fn new( mode: OperatingMode, initial_seq: CommitSeq, group_commit_max: usize, ) -> Self

Create a new coordinator for the given mode.

initial_seq is the highest committed sequence from recovery. group_commit_max is the maximum batch size for group commit.

pub const fn mode(&self) -> OperatingMode

The current operating mode.

pub const fn commit_seq_tip(&self) -> CommitSeq

The current commit sequence tip (highest committed).

pub fn pending_count(&self) -> usize

Number of pending commits in the current group batch.

pub const fn current_epoch(&self) -> u64

The current epoch (incremented once per group commit flush).

pub fn initiate_shutdown(&mut self)

Initiate shutdown: new submissions will be rejected.

pub fn validate( &self, submission: &CommitSubmission, ) -> Result<(), CommitResult>

Step 1: Validate a submission (FCW + SSI re-validation).

Returns Ok(()) if validation passes, or Err(CommitResult) with the rejection reason.

The coordinator MUST NOT decode the entire capsule here — only check the write-set summary against the commit index.

pub fn submit( &mut self, submission: CommitSubmission, now_unix_ns: u64, ) -> Result<CommitSeq, CommitResult>

Steps 2-8: Process a validated submission through the serialized section.

This allocates a commit_seq, builds the proof and marker, and adds the commit to the group batch. Returns the allocated commit result.

The caller MUST call flush_batch to actually issue fsyncs and complete the commits.

Errors

Returns Err if validation fails (FCW or SSI conflict).

pub fn fsync1(&mut self) -> usize

Execute FSYNC_1 (pre-marker group commit point).

Makes all pending capsule symbols AND CommitProof objects durable BEFORE markers reference them. Without this barrier, NVMe write reordering can make a marker durable while referents are not.

Returns the number of commits covered by this fsync.

pub fn append_markers_and_fsync2(&mut self) -> Vec<CommitMarker>

Steps 5-6: Append markers and execute FSYNC_2 (post-marker).

For each pending commit with FSYNC_1 complete, appends a CommitMarker to the marker stream, then marks FSYNC_2 complete.

Returns the markers that were appended.

pub fn drain_committed(&mut self) -> Vec<CommitResult>

Steps 7-8: Drain committed entries and return results.

Step 7 (SHM publish) is the caller’s responsibility after receiving these results — the coordinator only manages the serialized section.

pub fn flush_batch(&mut self) -> Vec<CommitResult>

Flush the pending batch: fsync1, append markers, fsync2, drain.

Wraps the full group commit cycle in a parallel_wal_commit tracing span with epoch, group_size, and commit_seq range fields. Returns the committed results and records metrics.

pub fn submit_and_commit( &mut self, submission: CommitSubmission, now_unix_ns: u64, ) -> CommitResult

Convenience: submit, fsync1, append markers, fsync2, drain.

Processes a single submission through the entire protocol. In production, submissions are batched; this is for testing and single-commit workloads.

Trait Implementations

impl Debug for WriteCoordinator

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.