Skip to content

Lab: The Multipart ETag Mystery

中文版

Run:

uv run python labs/lab_multipart_etag.py

The lab stores same-bytes once with a single PUT and once as two multipart parts. Both GET bodies compare equal, but their ETags do not:

same body: True
single PUT ETag: "e1d44c23b69953b35433ff067798318a"
multipart ETag: "05888a49b792dfb72298daafe3807667-2"
ETags differ: True

The multipart formula is not md5(complete body). It is md5(part-1 MD5 binary digest || part-2 MD5 binary digest)-2. Hashing the hexadecimal digest text would produce another value and is the classic bug this experiment is designed to expose.

The lab lowers MiniS3's configurable minimum part size to three bytes so it can remain tiny. Production S3's normal minimum for every completed part except the last is 5 MiB; MiniS3 uses that value by default.