BigQuery provides insertIds and does some de-duplication to help with the failure scenarios when inserting data via APIs. According to documentation, BigQuery will remember an insert for up to a minute, so if an insert fails, it is possible to retry the insert via the API without worrying about possible (inserted) data duplication. This can be tricky to get right.
The problem is that on Google Cloud there is a ton of services that promise to insert data into BigQuery. For example, DataFlow / Apache Beam is a part of the recommended stack for getting data from many sources into BigQuery. There are also Dataprep, Stackdriver Logging, and others.
So is there a single consistent way to recover failed inserts in BigQuery when using an arbitrary 3rd party BigQuery client, i.e. not the BigQuery API?