Skip to content

v1.2.2

Choose a tag to compare

@EugeneAndMonty EugeneAndMonty released this 01 Aug 22:16
· 24 commits to master since this release
f84c168

1.2.2

Adds a way to read the server's real semantic configuration, and a safe way to
change an enrolled keyspace's embedding model. Additive — upgrading from 1.2.1
requires no code changes.

Added

  • Engine.getSemanticStatus({ store, keyspace }) returns the server's actual
    semantic settings rather than what the caller assumed: the DB-wide switch and
    default model, plus each enrolled keyspace's model, dimensions, field,
    storage type, and whether a backfill is still pending.
  • Engine.reembedSemanticSearch({ model, field, store, keyspace }) atomically
    drops one keyspace's vectors, records the new configuration, and starts a
    complete backfill. It reports the previous model alongside the new one, so a
    caller can confirm what it replaced.

Changed

  • Documented that enableSemanticSearch leaves an already-enrolled keyspace
    alone. It was never a way to switch models; reembedSemanticSearch is.
    Behavior is unchanged — only the documentation was misleading.
  • Corrected the disableSemanticSearch docs: dropVectors is not "required
    before switching to a different embedding model". Use
    reembedSemanticSearch, which does not leave the keyspace unsearchable in
    between.