You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
A transient failure of a collection-description read (a dropped connection, a
5xx) is no longer taken for "no encryption descriptor". WasRemoteStore.readCollectionEncryption answers undefined only for a
not-found response or a description with no encryption member; any other
failure is thrown, wrapped with the collection id. A login fails its connected
activation on it (falling back to local with the anonymous replica intact
for a retry), a hot restore opens that collection fail-closed and still lands
connected, the sync bootstrap skips the collection for the session with a
warning, and the unknown-epoch refresh warns and leaves the cipher as it was.
A transient failure of a collection's /meta read is no longer taken for "no
metadata". WasRemoteStore.readCollectionMeta answers undefined only for a
not-found response or a backend without metadata support (501); any other
failure is thrown, wrapped with the collection id, and the sync bootstrap's
blinded-index schema install warns and skips that collection.
Neither read retries on its own: the HTTP client underneath already retries
transient status codes and network errors, so a second retry layer only
multiplied the attempts and the delay before a failure surfaced.
A session that loses the attach-time storage-context claim (two providers
booting at once, or a keyed remount overlapping the old provider's teardown)
now closes the replica it opened before failing, instead of leaving an open
RxDB database behind.
StorageContext.detachStore now releases the active storage-context pointer
when that context holds it, so every teardown (destroy(), logout,
clear-data, the connected activation's fallback to local) releases it.
Previously nothing did, and after a provider unmount the facades kept
resolving the retired context; they now throw until the next boot claims a
live one.
A connected activation that fails and whose fallback to local also fails now
surfaces the activation's own error; the fallback's is warned about.