Optimizing Kubernetes API `DeepEquals` with Go String Interning

comparison.md

Optimizing Kubernetes API DeepEquals with Go String Interning

Code used: jpbetz/kubernetes@intern-bench

Methodology

This measures performance impact of interning strings (via Go's unique.Make) on DeepEquals.

For each API type, the test:

1. Fuzzes an object, populating it with strings of a specific length.
2. Compares the performance of DeepEquals on objects with distinct string allocations versus the exact same objects with interned strings.

Results

String Length	Distinct (ns/op)	Interned (ns/op)	Speedup
Default (~10B)	~3,740K	~3,780K	~0%
100B	~3,860K	~3,700K	~4%
1KB	~4,220K	~3,630K	~14%
10KB	~8,590K	~3,690K	~57%

String Equality

Mostly for my own benefit as a first time user of unique.Make, I wanted to see how Go handles string equality.

This is for 10kb strings:

Comparison Type	Time (ns/op)	Explanation
Normal String s1 == s2	127.7000	Standard byte-by-byte comparison overhead.
Interned String h1.Value() == h2.Value()	0.9098	Comparing strings generated by unique.Make. Go checks string memory pointers first, skipping the expensive byte scan.
Interned Handle h1 == h2	0.2440	Comparing unique.Handle[string] types directly. This is the fastest approach, bypassing string header unpacking completely.

Conclusion

String interning provides no measurable benefit for very short strings, but dramatically accelerates DeepEquals comparisons (up to 57%) for objects containing large, repetitive strings. Where applicable, comparing raw unique.Handle types is the absolute fastest approach.