[mushi]

I think maybe the doubt is based on case like https://aphyr.com/posts/282-jepsen-postgres which all systems are subject to.

[continuational]

That's not a case of inconsistency. There are just three outcomes for a transaction: Committed, Aborted and Unknown.

The client will never experience Committed unless it really is committed, and thus never report that it was successful to the user.

[throwawayish]

Uh, no. That article is a bit of a long winded way to say that 2PC with timeouts is 2PC with timeouts.

[aphyr]

It might be better to think of it as a limitation of two-generals, rather than 2PC in particular.