This message was deleted.

true although the reverse is also not good— if we delete the segment from S3 first, then it's possible we'll leave an orphaned metadata record If it gets marked used again (someone trying to reload previously dropped data?) then servers will fail to fetch the segment

I recently ran into an issue where write privs were revoked to a specific S3 datasource folder, the kill task logged errors for the S3 delete attempt but still deleted the druid_segment records. A relatively simple improvement here might be to check for file existence after the delete attempt before deleting the metadata records ...? Thanks. John