We had issues with dropping data from druid and revisited our deleting strategy via this commit: https://gerrit.wikimedia.org/r/#/c/analytics/refinery/+/502858/3/python/refinery/druid.py
that disables a datasource and later sends a delete task. Still last month when the system timer kicked in that did the deletion we trigger an outage, documented here:
https://wikitech.wikimedia.org/wiki/Incident_documentation/20190616-AQS
There has to be a bigger issue with deletion of data from deep storage, we need to research that.