Skip to content

Wolverine 6.58.0

Choose a tag to compare

@jeremydmiller jeremydmiller released this 29 Oct 22:57
f0e529d

New

  • GH-5279 (#5280, #5282): New opt-in Durability.UseTransactionScopedDurabilityPassLock for PostgreSQL. The serverless durability pass takes a transaction-scoped advisory lock (pg_try_advisory_xact_lock) instead of a session lock. This makes the pass lock safe behind a transaction-mode pooler such as PgBouncer or Supabase's pooler, where a session lock can leak onto a pooled connection or be held by two passes at once. #5282 hardens it:

    • the lock connection stays out of an ambient TransactionScope
    • an idle_in_transaction_session_timeout can no longer drop the lock in the middle of a pass
    • a lock transaction that ends before the pass does is reported, and its pooled connection is discarded
    • a session unlock that fails is logged as a warning

    Wolverine also logs a one-time warning when the store doesn't support the setting, or when extension steps such as Marten's projection catch-up run under the open lock transaction. That open transaction can stall Marten's gap detection, which is tracked in JasperFx/marten#5791. See the serverless guide.

Fixes

  • GH-5276 (#5277, thanks @3nth): The HTTP transport's batch endpoint now answers 503 when a durable local queue cannot write the batch to its inbox. Before, it answered 200 while the messages were dropped, so a durable outbox deleted messages that were never stored on either side. The sender now keeps the batch and retries it.
  • GH-5278 (#5281): A durable TCP listener now answers ProcessingFailure instead of Received when its inbox could not store a batch. The sender keeps the batch and retries it, and a late duplicate is deduplicated by id. gRPC listeners can't be durable, so they aren't affected. A guard test now pins that.
  • #5275 (thanks @lahma): With Durability.EnableInboxPartitioning, PostgreSQL's durable inbox now refuses a redelivery whose identity was already handled. Before, the kept Handled row sat in a different partition and didn't collide, so the redelivery was handled a second time. This completes the stable-id deduplication from 6.57.0 for partitioned inboxes.