Interaction of ack timeout, retry middleware, and poison queue causes failures to write to poison queue.

[Zach-Johnson]

Steps to reproduce

• Postgres running with Watermill publisher/subscriber

Subscriber with a default config like this, ack deadline gets 30s by default:

	beginner := watermillSQL.BeginnerFromPgx(pool)

	subscriber, err := watermillSQL.NewSubscriber(
		beginner,
		watermillSQL.SubscriberConfig{
			SchemaAdapter:    watermillSQL.DefaultPostgreSQLSchema{},
			OffsetsAdapter:   watermillSQL.DefaultPostgreSQLOffsetsAdapter{},
			InitializeSchema: true,
		},
		NewWatermillLogger(ctx),
	)

Retry middleware, total time of retries exceeds 30s.

		middleware.Retry{
			MaxRetries:      5,
			InitialInterval: time.Second,
			MaxInterval:     time.Second * 30,
			Multiplier:      2,
			Logger:          events.NewWatermillLogger(ctx),
		}.Middleware,

Poison queue enabled.

Expected behavior

I would expect failed messages to be published to the poison queue.

Actual behavior

{"level":"error","time":"2025-11-17T10:39:08-07:00","message":"Handler returned error: context deadline exceeded\ncannot publish message to poison queue: could not insert message as row: context deadline exceeded (fields: map[handler_name:notification_handler message_uuid:4afd0039-e529-406b-8e22-3dfc96870e0e])"}

This happens because the default ack deadline causes immediate context expiration when attempting to write to the poison queue.

Possible solution

I would expect either:

• Watermill to fail on startup somewhere and flag this as a mis-configuration
• The context to reset on publishing to the poison queue - maybe with a configurable timeout separate for poison queue writes.