My interpretation is... Atlassian Cloud is hosted in shards where a shard is allocated a single AWS RDS database with (say) 100 customers.
Their script was run improperly and dropped tables from 400 customers across 400 different shards (across different databases)
Their backup strategy was just to restore RDS backups (recover whole database with many customers data together)
But they can't simply restore from RDS backups due to the co-mingling of customers. As otherwise unaffected customers would now lose data.
I suspect they're manually restoring RDS backups to new DB instances and having to scrape out the data for a single customer - a process they'd not built any tooling for as it was a fault case they thought they didn't need. Possibly as running the Jira/confluence backups are very heavy and create a lot of load/disk that they can't easily accomodation in a multi tenant hosting architecture.