[evanelias]

> the PostgreSQL MVCC approach is well worth never seeing ORA-01555 like error message.

But that's a false comparison; it's not a binary decision. As InnoDB demonstrates, you can take the undo log approach without having ORA-01555 style errors.

The article greatly exaggerates the operational pain from InnoDB purge. In practice, with an OLTP workload, substantial InnoDB purge lag problems are much rarer than Postgres XID wraparound problems. Generally in InnoDB they'll only come up if you take a logical dump of an overly-large unsharded data set (solvable by using a read replica instead), or perform slow OLAP-style queries (ditto), or do a massive data migration in a single transaction (solve by breaking it up).

The rollback speed criticism is much more valid, but the impact of that depends on how often you generally need to issue rollbacks of large writes in production. But for most workloads this is rare in my experience: the large production writes should be rare in the first place, and needing to roll them back is rarer still.