Isolation level comparison: which level disallows dirty reads and nonrepeatable reads, but still allows phantom reads?

Introduction / Context:
Understanding which anomalies are blocked at each isolation level is fundamental for designing correct transactional behavior under concurrency.

Given Data / Assumptions:

• Dirty read must be disallowed.
• Nonrepeatable read must be disallowed.
• Phantom reads may still occur.

Concept / Approach:
Repeatable read ensures that if a transaction reads a row twice, the values do not change between reads (no dirty or nonrepeatable reads). It does not, however, lock ranges in a way that always prevents new qualifying rows from appearing, so phantom reads can still occur (engine specifics aside).

Step-by-Step Solution:

Verification / Alternative check:
Refer to ANSI SQL isolation definitions; behavior matches the description, noting DBMS nuances (e.g., Next-Key locks may reduce phantoms).

Why Other Options Are Wrong:
Read committed and read uncommitted are too permissive; serializable is stricter than required.

Common Pitfalls:
Thinking repeatable read also blocks phantoms universally; that typically requires serializable or range locking.

Final Answer:
Repeatable read