Skip EnhancedAttributeValue wrapping in converters transformTo when t…

[RanVaknin]

The Enhanced Client processes every DynamoDB Enhanced client operation through a conversion pipeline that transforms between AttributeValue and Java POJOs. This optimization targets the deserialization path where responses from DynamoDB are converted back into Java objects.

Current Approach (Before Optimization)

Our current approach for deserialization codepaths (transformTo method) wraps every AttributeValue in an EnhancedAttributeValue, even when no type conversion is needed:

public class StringAttributeConverter implements AttributeConverter<String> {
    @Override
    public String transformTo(AttributeValue input) {
        // Always wraps, even for String -> String
        return EnhancedAttributeValue.fromAttributeValue(input).convert(TypeConvertingVisitor.INSTANCE);
    }
}

The problem is that for the common case where types already match (String→String, Bytes→byte[]), we're creating unnecessary wrapper objects and going through the visitor pattern.

Optimized Approach (After)

Short-circuit the common case where no type conversion is needed:

public class StringAttributeConverter implements AttributeConverter<String> {
    @Override
    public String transformTo(AttributeValue input) {
        if (input.s() != null) {
            return input.s();  // Fast path
        }
        // Slow path: type conversion (Number->String, Boolean->String, etc.)
        return EnhancedAttributeValue.fromAttributeValue(input).convert(TypeConvertingVisitor.INSTANCE);
    }
}

Scope of Optimization

This optimization applies to scalar type converters where the DynamoDB type directly maps to the Java type:

• StringAttributeConverter (String → String) (covered in this PR)
• ByteArrayAttributeConverter (Bytes → byte[]) (covered in this PR)

Not covered in the PR because not included in the Datasets. Will likely benefit the same way but harder to prove without more data:

• BooleanAttributeConverter (Boolean → Boolean)
• ByteBufferAttributeConverter (Bytes → ByteBuffer)
• SdkBytesAttributeConverter (Bytes → SdkBytes)

Cannot be optimized:

• Collection converters (List, Map, Set) - must recursively convert nested elements, requiring the visitor pattern to handle arbitrary nesting and type conversions
• Number converters (Integer, Long, Double, etc.) - DynamoDB's wire format stores numbers as strings (e.g., {"N": "123"}). The converter must parse this string representation regardless of whether we short-circuit the wrapper. Since input.n() returns "123" (a String), not 123 (an Integer)

Flamegraphs (Get SMALL):

Before:

After:

Benchmarks:

Operation	Size	V2 Master Mean (us/op)	optimized	Improvement
Delete	TINY	0.425	0.388	9.54%
Delete	SMALL	0.413	0.422	-2.13%
Delete	HUGE	0.411	0.402	2.24%
Delete	HUGE_FLAT	0.407	0.4	1.75%
Get	TINY	0.471	0.445	5.84%
Get	SMALL	1.001	0.896	11.72%
Get	HUGE	10.256	9.048	13.35%
Get	HUGE_FLAT	3.075	2.863	7.40%
Put	TINY	0.746	0.717	4.04%
Put	SMALL	1.346	1.37	-1.75%
Put	HUGE	13.449	13.194	1.93%
Put	HUGE_FLAT	8.236	7.855	4.85%
Query	TINY	2.544	2.371	7.30%
Query	SMALL	4.273	3.74	14.25%
Query	HUGE	30.658	30.639	0.06%
Query	HUGE_FLAT	8.884	9.05	-1.83%
Scan	TINY	1.088	0.928	17.24%
Scan	SMALL	2.604	2.274	14.51%
Scan	HUGE	29.882	27.949	6.92%
Scan	HUGE_FLAT	8.496	7.778	9.23%
Update	TINY	1.372	1.317	4.18%
Update	SMALL	9.324	9.118	2.26%
Update	HUGE	39.691	39.114	1.48%
Update	HUGE_FLAT	241.643	237.085	1.92%

[zoewangg]

Seems worth adding changelog entry?

[sonarqubecloud]

Quality Gate failed

Failed conditions
30.2% Coverage on New Code (required ≥ 80%)

See analysis details on SonarQube Cloud

[github-actions]

This pull request has been closed and the conversation has been locked. Comments on closed PRs are hard for our team to see. If you need more assistance, please open a new issue that references this one.