This message was deleted.

for example, in this setup, I don't know how you can make druid split each task replica to a different zone (I think that's was the missing part for solving the issue if the entire zone goes down) if it will keep the task on the MM serving queries for the period of "Completion timeout", after this time elapses, if the metastore is still not up, the task will fail and the queries will lose this data until a new task is created to reprocess