Explanation

Correct answer is Use a Batch job scope to bulk insert records into the database

* Batch Job is most efficient way to manage millions of records.

A few points to note here are as follows :

Reliability: If you want reliabilty while processing the records, i.e should the processing survive a runtime crash or other unhappy scenarios, and when restarted process all the remaining records, if yes then go for batch as it uses persistent queues.

Error Handling: In Parallel for each an error in a particular route will stop processing the remaining records in that route and in such case you'd need to handle it using on error continue, batch process does not stop during such error instead you can have a step for failures and have a dedicated handling in it.

Memory footprint: Since question said that there are millions of records to process, parallel for each will aggregate all the processed records at the end and can possibly cause Out Of Memory.

Batch job instead provides a BatchResult in the on complete phase where you can get the count of failures and success. For huge file processing if order is not a concern definitely go ahead with Batch Job