Select-then-Update vs Single Update: What Matters More Than Performance

Problem

While building the Gift API clone at Kakao Tech Campus, I needed to implement a product update API. In a JDBC environment, there were two main approaches:

Method 1: Single Update Query

• One UPDATE product SET ... WHERE id = ? and done
• Just 1 query

Method 2: Select then Update

• Fetch first, modify the object, then save
• 2 queries

From a performance standpoint, Method 1 wins. But I went with Method 2.

---

Limitation of Method 1: Ambiguity of affected rows

A single Update query returns affected rows—the number of rows changed. If it’s 0, you might assume the update failed.

The problem? You can’t tell why it’s 0.

• Does the ID not exist?
• Does the user lack permission?
• Did some business condition (status, deleted_at, etc.) fail?

From the client’s perspective, “Product does not exist” and “You don’t have permission” mean completely different things—but affected rows alone can’t tell them apart.

---

Advantage of Method 2: Clear Exception Separation

By fetching the domain object first, you can separate validation logic at the application level.

1. Check existence → EntityNotFoundException if missing
2. Check permission → AccessDeniedException if denied
3. Check state → IllegalStateException if already deleted

This way, each exception is clearly separated, making debugging easier and API responses more meaningful.

---

Is the Performance Overhead Acceptable?

You might worry about the extra query, but here’s my reasoning:

1. PK lookups are cheap — B-Tree index lookups are O(log n), usually just 3–4 disk I/Os.
2. Most update requests succeed — Editing a nonexistent product is the exception, not the rule.
3. Optimize at the bottleneck — There’s no rush to optimize before there’s an actual performance issue.

---

Summary

Criteria	Single Update	Select-then-Update
Performance	Faster	1 extra query
Exception handling	Impossible	Clear
Maintainability	Harder	Easier
Debugging	Inconvenient	Convenient

Takeaway: Designing predictable and safe systems matters more than raw performance.

---

From Kakao Tech Campus 3rd cohort Gift API clone coding.