perf(storage): use google_crc32c.value for checksums

[zhixiangli]

Updates _ReadResumptionStrategy to use google_crc32c.value(data) instead of manually converting Checksum(data).digest() to an integer.

Updated unit tests to mock google_crc32c.value accordingly.

Test code

import timeit
import os
import google_crc32c
from google_crc32c import Checksum

def method1(data):
    return int.from_bytes(Checksum(data).digest(), "big")

def method2(data):
    return google_crc32c.value(data)

def benchmark():
    # Testing larger sizes and more iterations
    # 1KB, 1MB, 10MB, 100MB
    data_sizes = [1024, 1024 * 1024, 10 * 1024 * 1024, 100 * 1024 * 1024] 
    
    for size in data_sizes:
        data = os.urandom(size)
        print(f"\nData size: {size / (1024*1024):.2f} MB" if size >= 1024*1024 else f"\nData size: {size / 1024:.2f} KB")
        
        # Correctness check
        res1 = method1(data)
        res2 = method2(data)
        assert res1 == res2, f"Failed for size {size}: {res1} != {res2}"
        print("Assertion passed: results are equal.")
        
        # Increase iterations for more stable results, with a minimum of 1000
        if size <= 1024:
            number = 100000
        elif size <= 1024 * 1024:
            number = 10000
        else:
            number = 1000
        
        t1 = timeit.timeit(lambda: method1(data), number=number)
        t2 = timeit.timeit(lambda: method2(data), number=number)
        
        avg1 = t1 / number
        avg2 = t2 / number
        
        print(f"Method 1 (Checksum(data).digest()): {t1:.6f} s total ({avg1:.8f} s/call)")
        print(f"Method 2 (google_crc32c.value(data)): {t2:.6f} s total ({avg2:.8f} s/call)")
        print(f"Improvement: {(t1 - t2) / t1 * 100:.2f}%")

if __name__ == "__main__":
    print(f"google_crc32c implementation: {google_crc32c.implementation}")
    benchmark()

Output

google_crc32c implementation: c

Data size: 1.00 KB
Assertion passed: results are equal.
Method 1 (Checksum(data).digest()): 0.046088 s total (0.00000046 s/call)
Method 2 (google_crc32c.value(data)): 0.016062 s total (0.00000016 s/call)
Improvement: 65.15%

Data size: 1.00 MB
Assertion passed: results are equal.
Method 1 (Checksum(data).digest()): 0.464044 s total (0.00004640 s/call)
Method 2 (google_crc32c.value(data)): 0.439121 s total (0.00004391 s/call)
Improvement: 5.37%

Data size: 10.00 MB
Assertion passed: results are equal.
Method 1 (Checksum(data).digest()): 0.450953 s total (0.00045095 s/call)
Method 2 (google_crc32c.value(data)): 0.445793 s total (0.00044579 s/call)
Improvement: 1.14%

Data size: 100.00 MB
Assertion passed: results are equal.
Method 1 (Checksum(data).digest()): 6.287893 s total (0.00628789 s/call)
Method 2 (google_crc32c.value(data)): 6.095833 s total (0.00609583 s/call)
Improvement: 3.05%

[gemini-code-assist]

Code Review

This pull request refactors the CRC32C checksum calculation in the asyncio reads resumption strategy by replacing the Checksum class usage with the more direct google_crc32c.value function. The associated unit tests have been updated to mock this new function and validate integer results instead of byte digests. I have no feedback to provide.