Struct Registry 

pub struct Registry { /* private fields */ }

SQLite registry handle. SQLite-level WAL + busy_timeout handles concurrent reader/writer coordination across processes; we no longer hold a long-lived exclusive file lock for the lifetime of the handle. A short exclusive lock is taken only across the schema-migration window in open_at to serialise initial CREATE TABLE / CREATE INDEX against a concurrent first open.

Implementations

impl Registry

pub fn bidi_lock_acquire( &self, browser_name: &str, timeout: Duration, ) -> Result<BidiLockGuard>

Acquire the BiDi lock for browser_name. Blocks until granted or timeout elapses. On contention, evicts a stale row whose holder PID is no longer alive.

Polling rather than file-locking on the row: the registry’s process-level file lock already serializes connections, and the acquire path is single-row + atomic. We re-poll every [POLL_INTERVAL] until success or timeout.

pub fn bidi_lock_holder( &self, browser_name: &str, ) -> Result<Option<BidiLockRow>>

Read the current lock holder for diagnostics / tests. Returns None if no row exists.

impl Registry

pub fn scratch_get(&self, browser_name: &str) -> Result<Option<ScratchRow>>

Read the scratch row for browser_name. Returns None if no row.

pub fn scratch_upsert(&self, browser_name: &str, target_id: &str) -> Result<()>

Insert or replace the scratch row for browser_name. Used both when creating the first scratch and when recovering a wedged one.

pub fn scratch_touch(&self, browser_name: &str) -> Result<()>

Bump last_used_at for the scratch row (no-op if absent). Called after a successful op so the row reflects actual recency for diagnostics.

pub fn scratch_delete(&self, browser_name: &str) -> Result<()>

Remove the scratch row (used by recovery to mark the prior target dead before re-creating). Idempotent.

impl Registry

pub fn tab_get(&self, browser_name: &str, name: &str) -> Result<Option<TabRow>>

Read a single tab by composite key.

pub fn tabs_list_for(&self, browser_name: &str) -> Result<Vec<TabRow>>

All tabs for a browser, sorted by name (stable for tab list output).

pub fn tab_upsert( &self, browser_name: &str, name: &str, target_id: &str, last_url: &str, daemon_created: bool, ) -> Result<()>

Insert or replace a tab. Bumps last_used_at to “now”.

pub fn tab_touch(&self, browser_name: &str, name: &str) -> Result<()>

Bump last_used_at without rewriting other columns. No-op if absent.

pub fn tab_set_url( &self, browser_name: &str, name: &str, url: &str, ) -> Result<()>

Rewrite just last_url (after a successful Page.navigate).

pub fn tab_delete(&self, browser_name: &str, name: &str) -> Result<()>

Delete by composite key. Idempotent.

pub fn tabs_count_daemon_created(&self, browser_name: &str) -> Result<usize>

Count daemon_created = true rows for a browser. Used by the orchestration layer to decide when to recycle under budget pressure.

pub fn tabs_lru_daemon_created( &self, browser_name: &str, ) -> Result<Option<TabRow>>

LRU daemon-created row for a browser. Returns the row with the smallest last_used_at_epoch_s, or None if no eligible row.

impl Registry

pub fn open() -> Result<Self>

Open the registry at the OS-standard location.

pub fn open_at(path: &Path) -> Result<Self>

Open the registry at an explicit path.

pub fn open_in_memory() -> Result<Self>

Open an in-memory registry (tests only). No file lock taken.

pub fn insert(&self, row: &BrowserRow) -> Result<()>

Insert (or replace) a row.

pub fn delete(&self, name: &str) -> Result<()>

Delete a row by name. No error if it does not exist.

pub fn get_by_name(&self, name: &str) -> Result<Option<BrowserRow>>

Look up a row by name.

pub fn list_all(&self) -> Result<Vec<BrowserRow>>

All rows, no liveness check, ordered by started_at DESC.

pub fn list_alive(&self) -> Result<Vec<BrowserRow>>

All alive rows. Dead-PID rows are deleted as a side-effect; rows whose endpoint is temporarily unreachable are omitted but retained.

pub fn first_alive_by_kind(&self, kind: Kind) -> Result<Option<BrowserRow>>

First alive row of the given kind (most recent first). Dead-PID matches are pruned; endpoint-unreachable rows are retained.

pub fn most_recent_alive(&self) -> Result<Option<BrowserRow>>

Most recently started alive row across all kinds.