The failure store is already available as an experimental feature in the 8.15 snapshot build of Elasticsearch.

While it's not ready yet to be enabled as a default feature for integrations, it can be used as an additiona layer of security, making sure no documents are dropped during system tests which would indicate an issue with the integration.

I propose the following steps:

• In stack version >=8.15, always enable the failure store on all data streams
• Similar to Extend system test to validate absence of _ignored #1738, validate that a datastream under test didn't drop any documents in the failure store (fail the test in case it happens)
• Add a configuration option to the test config that allows skipping this check
• Document what the failure store means and how to skip the check

What do you think @ruflin @jsoriano ? It seems like this can be done even before the failure store is ready to be used on customers clusters.