Introduction

In a few cases while working on Go software, I have needed to generate a cryptographic hash (mainly SHA-256) from a string or JSON value. In all cases up to this point, I needed to convert the resulting []byte to a string representation for storage or viewing. In my investigation, I found two popular ways to approach this problem. I’ll describe them below as well as explore the performance implications of each choice.

Generating the Hash

For all intents and purposes, this process will yield the same result for all common hashes such as MD5, SHA-1, SHA-256, and SHA-512. In my preliminary investigation, I only ran benchmarks on SHA-256. That benchmark data will be what we take a look at today. The process used is quite straightforward, and can be found here.

Of note, the line shaBytes := sha256.Sum256([]byte(a)) is supplied with a string of 20 random characters in each test.

Method #1 (fmt.Sprintf)

func MethodOne(a string) string {
	shaBytes := sha256.Sum256([]byte(a))
	return fmt.Sprintf("%x", shaBytes)
}

This was the first method that I encountered in production code at work. Using the %x (hexadecimal) format specifier in conjunction with fmt.Sprintf() is a quick way to return the string representation of a hash. In my opinion, this method is more useful when printing to the console with fmt.Printf() as it requires one less line of code than the second method.

Method #2 (hex.EncodeToString)

func MethodTwo(a string) string {
	shaBytes := sha256.Sum256([]byte(a))
	return hex.EncodeToString(shaBytes[:])
}

This was the method I settled upon to generate over 300,000 hashes-per-minute in multiple production services. When using this method it is important to remember that the EncodeToString() function requires a byte slice as it’s argument. To convert the byte array to a byte slice is as simple as using shaBytes[:], which can be seen above.

Performance Statistics

Below is the direct output of the benchmarking process.

λ  sha_hex_string_bench main ✗  go test -run=XXX -bench=. -benchtime=100000x
goos: darwin
goarch: amd64
pkg: github.com/michaelpeterswa/sha_hex_string_bench
cpu: Intel(R) Core(TM) i7-9750H CPU @ 2.60GHz
BenchmarkShaHexStringWithHexLib-12     	  100000	       729.5 ns/op	     176 B/op	       4 allocs/op
BenchmarkShaHexStringWithSprintf-12    	  100000	       966.3 ns/op	     176 B/op	       5 allocs/op
BenchmarkRandStringBytes-12            	  100000	       382.2 ns/op	      48 B/op	       2 allocs/op
PASS
ok  	github.com/michaelpeterswa/sha_hex_string_bench	0.336s

And here is the simplified (and arguably more readable) table.

Benchmarksiterationsns/opB/opallocs/op
Method One (fmt.Sprintf)100,000966.31765
Method Two (hex.EncodeToString)100,000729.51764
Input String Generation100,000382.2482

Conclusion

As seen in the above data, hex.EncodeToString()is significantly faster. This is because the EncodeToString function does not rely on reflection internally like Sprintf does. It also has one less allocation to the heap for the same memory footprint, which is a benefit at scale. Although in most cases the added performance benefit wouldn’t matter, it’s important to take a step back and reflect on why certain design choices were made. Even small additions such as these SHA-256 hashes may negatively impact performance or increase cost in the long term. We want to avoid that! 😁

For more info about allocations and what they translate to under the hood check out this article: Understanding Allocations in Go