Struct SnapshotSource 

pub struct SnapshotSource<L>
where
    L: SnapshotLoader,
{ /* private fields */ }

The whole-table snapshot cache described in the module doc comment, generic over a SnapshotLoader.

Implementations

impl<L> SnapshotSource<L>

where L: SnapshotLoader,

pub fn new(pool: Pool<Postgres>, ttl: Duration) -> SnapshotSource<L>

Builds a source over pool (expected to be a crate::db_api_key::dedicated_key_store_pool, not the caller’s lake pool directly), with no snapshot loaded yet – the first current() call performs the first query.

pub async fn current( &self, ) -> Result<Arc<<L as SnapshotLoader>::Snapshot>, Error>

Returns the current snapshot, refreshing it first if stale.

Both the cold-start path (no Snapshot yet) and the post-success path are throttled to at most one DB query per TTL window: cold-start via last_attempt_at (checked-and-set the same compare-exchange way db_api_key.rs::maybe_log_error rate-limits its own log line), the post-success path via Snapshot::fetched_at as before. A current() call that lands inside an already-throttled window returns the last snapshot if one exists; if not, it returns a freshly synthesized error describing the throttled cold-start state rather than storing and re-handing-out the prior attempt’s anyhow::Error. A concurrent caller that loses the last_attempt_at compare-exchange while the very first cold-start attempt is still in flight re-checks self.snapshot once more before giving up: if the winning caller has since populated it, the loser serves that snapshot instead of failing; only if it is still empty does it get the synthesized throttled-state error. It never blocks on another caller’s still-in-flight query, and it never skips the query silently with nothing to show for it.

Err only when there has never been one successful load – a fresh process whose first query hits a down DB has no “last good” to serve, so it fails closed like everything else on this seam, at a rate capped by last_attempt_at rather than once per request. Once any load has succeeded, a later refresh failure is logged + counted (via SnapshotLoader::count_refresh_error) and the last good snapshot keeps serving, unbounded – this store has no per-item TTL eviction to fall back on the way db_api_key.rs’s cache does, so an outage degrades to staleness for as long as it lasts, not just one TTL window.

Trait Implementations

impl<L> Debug for SnapshotSource<L>

where L: SnapshotLoader,

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

Formats the value using the given formatter.

impl GroupSnapshot for SnapshotSource<GroupsLoader>

fn current<'life0, 'async_trait>( &'life0 self, ) -> Pin<Box<dyn Future<Output = Result<Arc<GroupGraph>, Error>> + Send + 'async_trait>>

where 'life0: 'async_trait, SnapshotSource<GroupsLoader>: 'async_trait,

Returns the current GroupGraph snapshot, or Err (typically a crate::types::ProviderUnavailable) on a store outage with no prior successful load.