Struct WalPuller 

pub struct WalPuller<'a> { /* private fields */ }

Tails one tenant’s R2 frame stream once per box and fans each newly pulled frame to every attached WarmApplier. See the module doc for the full design, the v1 cold-start-only attach scope cut, and the restart-refusal posture.

Implementations

impl<'a> WalPuller<'a>

pub fn new( target: &'a BackupTarget, page_size: usize, cfg: WalPullerConfig, ) -> Self

pub fn page_size(&self) -> usize

pub fn attached_count(&self) -> usize

pub fn disk_used_bytes(&self) -> u64

pub fn has_pulled(&self) -> bool

true once pull_once() has succeeded at least once — the point past which attach() refuses (see module doc: v1 does not replay backlog to a late-joining applier).

pub fn attach( &mut self, name: impl Into<String>, seam: Box<dyn WarmApplier>, dest_size_bytes: u64, ) -> Result<()>

Attach a warm applier: begins its wal_insert session and trims its page cache to cfg.applier_cache_kb. From this call on it receives every frame this puller fans out.

dest_size_bytes is the caller’s own estimate of the applier’s on-disk footprint, counted against cfg.max_disk_bytes — see the module doc for why this crate takes a hint instead of stat-ing a file.

Errors if: an applier is already attached under name; the FD budget (max_appliers) or disk budget (max_disk_bytes) would be exceeded; or this puller has already completed a pull_once() (v1 scope cut — see module doc).

pub fn detach(&mut self, name: &str) -> Result<Box<dyn WarmApplier>>

Detach an applier without promoting it: closes its wal_insert session (force_commit=false, the same crash-consistent default restore_latest_stream uses) and returns ownership, freeing its share of the FD/disk budget.

pub fn promote(&mut self, name: &str) -> Result<Box<dyn WarmApplier>>

Promote a warm applier to serve live traffic. Mechanically identical to Self::detach today (close the session cleanly, hand back the seam — it is already caught up as of the last pull_once()); kept as a separate name so call sites read intent, matching §5’s “Promotable fast → SLA-tier RTO”.

pub async fn pull_once(&mut self) -> Result<PullReport>

Pull every frame newly written since the last call (or since this puller was created) and fan each one out to every attached applier. Each frame is downloaded from the object store once regardless of how many appliers are attached — R2 reads are O(boxes), not O(replicas).

Errors if: the chain’s page_size doesn’t match this puller’s; or the source’s checkpoint_seq has changed since the last successful pull (a WAL restart — see module doc, refused rather than patched over). A missing or wrong-length frame object errors the same way restore_latest_stream’s replay does.