In a concurrent system, you could encounter three different problems:
- Dirty reads: A dirty read happens when the database is allowing a transaction to see work-in-progress data from other not-yet-finished transactions. In other words, data that has not been consolidated is visible to other transactions. No production-ready database allows that, and PostgreSQL is no exception: you are assured your transaction will only perceive data that has been consolidated and, in order to be consolidated, the transactions that created such data must be complete.
- Nonrepeatable reads: An unrepeatable read happens when the same query, within the same transaction, executed multiple times, perceives a different set of data. This essentially means that the data has changed between two sequential executions of the same query in the same transaction. PostgreSQL does not allow this kind of problem by means of snapshots: every transaction can perceive the snapshot of the data available...
