Skip to content

Durable Actors 0.7.6

Choose a tag to compare

@oliviersm199 oliviersm199 released this 01 Oct 21:58
acb7239

Actor writes continue acknowledging after both Rapid replicas flush. Standard archival now runs in background batches, triggered by 16 MiB or 10 seconds by default. Configure the triggers through DURABLE_ACTORS_ARCHIVE_BATCH_BYTES / DURABLE_ACTORS_ARCHIVE_BATCH_INTERVAL_MS or Helm storage.rapid.archiveBatchBytes / storage.rapid.archiveBatchIntervalMs.

The runtime uses immutable archive batches, bounded buffering with Rapid catch-up reads, asynchronous stream rotation, and cleanup after complete archival. Recovery combines validated Standard batches with the Rapid tail. A concurrency fix lets a lagging archiver recognize work already completed by cleanup instead of retrying deleted Rapid objects.

This release also avoids remote reads for validation of the latest committed parent snapshot and includes the current production capacity defaults: two control-plane replicas and an enabled 32-slot GKE capacity buffer.