Assume that you use an Always On availability group (AG) that contains heap tables. Starting in SQL Server 2016, parallel thread for redo operations is used in secondary replicas. In this case, heap tables redo operation may generate a runtime assert dump or the SQL Server may crash with an access violation error in some cases.
This issue occurs after you apply the cumulative update that is mentioned in the following article:
4018866 FIX: Parallel redo causes high memory usage in SQL Server 2016 when it's compared to SQL Server 2014 or earlier versions
To work around this issue, disable parallel redo by enabling Trace Flag 3459.
Microsoft has confirmed that this is a problem in the Microsoft products that are listed in the "Applies to" section.