Seems like we should update the MySQL metadata docs to also include modifying the transaction isolation level from the default REPEATABLE_READ to READ_COMMITTED. When measuring the task/action/runtime for several of our larger scale druid clusters on 27.0.0 (500+ peons, 100 Historicals), we noticed a significant drop from 15-20sec runtimes down to 1-2 seconds with chatAsync enabled and the modified READ_COMMITED transaction isolation level. We also noticed several stability improvements as well with scaling the number of tasks on our larger 2K+ peon cluster when scheduling a large amount of kafka indexing tasks (hitting upwards of 3K+ peons without any stability issues), compared to REPEATABLE_READ which was running into periodic deadlocks and persist errors with maintaining kafka offsets to the druid_dataSources table.