Message lost on stopMessage processing during rolling restart

We have 2 pods and 2 actor cluster roles T and A. Each pod has the 2 cluster roles. Actor T1 sends a message to actors A1 and A2 while a rolling restart is happening. A1 gets the message. But A2 receives a stop message prior to receiving the T1 message. In the stop message handling it persists an event before it stops the actor by doing Effect.persist.thenStop. The message from T1 never gets to A2 even when it restarts in a new pod. Shouldn’t the message be stored in the Shard so it will be resent to the actor when it wakes up? Or is it lost because the Shard is being destroyed for a new pod? Shouldn’t the Shard hand off any queued message to the new pod?