* refactor(sse): decouple encryption from ecstore * feat(kms): enhance KMS service manager with runtime state and persistence support * feat(kms): add local key export functionality for SSE-S3 migration tests * fix(kms): keep local key export narrowly scoped * fix(sse): validate copy source customer algorithm --------- Co-authored-by: Zhengchao An <anzhengchao@gmail.com>
MinIO Fixture Lab
This lab captures real MinIO backend artifacts into a repeatable local layout for RustFS compatibility tests.
It now supports two workflows:
- manual capture from an already exported backend tree
- automated capture from a disposable local MinIO run
Scope
Use the manual path after you already have:
- a running MinIO instance
- one or more uploaded objects you want to preserve as fixtures
- the backend object tree or object-version directory you want to export
Use the automated path when you want the lab to:
- locate a local
miniobinary - start a disposable MinIO instance
- upload a predefined SSE fixture case
- export the generated backend tree into the lab layout
Layout
The default root is artifacts/minio-fixture-lab, which is already ignored by the repository.
Each case is stored under:
artifacts/minio-fixture-lab/
cases/
<case-id>/
backend/
request.json
head.json
plaintext.sha256
manifest.json
manifest.json is the source of truth for the captured case.
Commands
Initialize the lab root:
uv run python D:\Github\rustfs\crates\rio-v2\tests\minio_fixture_lab\lab.py init
Capture one case from an existing MinIO backend tree:
uv run python D:\Github\rustfs\crates\rio-v2\tests\minio_fixture_lab\lab.py add-case `
--case-id sse-kms-singlepart-64k `
--bucket demo `
--object dir/object.bin `
--source-tree D:\minio-data-export\case-tree `
--head-json D:\minio-data-export\head.json `
--request-json D:\minio-data-export\request.json `
--plaintext-sha256 D:\minio-data-export\plaintext.sha256
Capture the default automated matrix:
uv run python D:\Github\rustfs\crates\rio-v2\tests\minio_fixture_lab\lab.py capture-matrix `
--root D:\Github\rustfs\artifacts\minio-fixture-lab `
--minio-binary D:\go\bin\minio.exe `
--endpoint https://127.0.0.1:19000
Capture only one automated case:
uv run python D:\Github\rustfs\crates\rio-v2\tests\minio_fixture_lab\lab.py capture-matrix `
--root D:\Github\rustfs\artifacts\minio-fixture-lab `
--minio-binary D:\Github\rustfs\tmp\minio.windows-amd64.RELEASE.2025-09-07T16-13-09Z.exe `
--endpoint https://127.0.0.1:19000 `
--case-id sse-s3-singlepart-64k
The automated default matrix is intentionally small:
sse-s3-singlepart-64ksse-kms-singlepart-64ksse-c-singlepart-64ksse-s3-multipart-8msse-kms-multipart-8msse-c-multipart-8m
64 KiB multipart is intentionally excluded because S3 multipart semantics require a larger non-final part size.
Automated Runner Prerequisites
The automated runner expects:
- either
minioinPATH, the bundledD:\Github\rustfs\tmp\minio.windows-amd64.RELEASE.2025-09-07T16-13-09Z.exe,--minio-binary, or--minio-rootpointing at a directory containingminio.exe - a free local endpoint port
When the selected matrix includes SSE-C cases, use an https:// endpoint. The lab will mint a short-lived local self-signed certificate and use its built-in SigV4 S3 client with certificate verification disabled for the disposable local MinIO run.
The runner provisions backend directories under --work-root and exports the resulting backend tree into the lab.
For local static-KMS runs, pass --kms-secret-key or set
MINIO_FIXTURE_LAB_KMS_SECRET_KEY using MinIO's
<key-id>:<base64-32byte-key> format. The runner derives the SSE-KMS request
key id from that configured key name automatically.
On some Windows MinIO builds, a multi-disk backend may not come online when all disk directories live on the same volume. If you only want a local smoke run of the upload/export pipeline, try:
uv run python D:\Github\rustfs\crates\rio-v2\tests\minio_fixture_lab\lab.py capture-matrix `
--root D:\Github\rustfs\artifacts\minio-fixture-lab `
--minio-binary D:\go\bin\minio.exe `
--endpoint https://127.0.0.1:19000 `
--disk-count 1 `
--case-id sse-s3-singlepart-64k
That is a runner smoke path only. Real compatibility fixtures should still prefer the intended multi-disk backend layout when the local environment can support it.
Via Docker (Linux/macOS, no local minio install)
When you don't have a minio binary on the host, capture_via_docker.sh
generates the fixtures the ignored round-trip tests consume using Docker only.
It builds a throwaway image (the official MinIO server binary pulled from
minio/minio plus this lab on a small Python base — lab.py drives MinIO's S3
API directly, so no mc is needed) and runs capture-matrix inside it, writing
fixtures under crates/rio-v2/tests/fixtures/minio-generated/ (the root the Rust
tests read):
# Default: the two 8 MiB multipart cases the round-trip tests need.
# Pass case ids to override, or "all" for the full default matrix.
./capture_via_docker.sh
RUSTFS_MINIO_STATIC_KMS_KEY_B64=IyqsU3kMFloCNup4BsZtf/rmfHVcTgznO2F25CkEH1g= \
cargo test -p rustfs --features rio-v2 storage::minio_generated_read_test --lib -- --ignored
This is exactly what the nightly minio-interop GitHub Actions workflow runs
(.github/workflows/minio-interop.yml), so the local and CI paths stay in sync.
SSE-C cases still need the host-minio + TLS path above; the Docker helper
targets the SSE-S3 / SSE-KMS multipart cases the interop tests assert on.
Capture Guidance
For each case, preserve these inputs when possible:
- the exact backend tree that contains
xl.metaand part files - the request shape used to create the object
- the API metadata returned by
HEAD Object - the plaintext digest used for byte-for-byte read verification
Recommended early matrix:
- singlepart SSE-S3
- singlepart SSE-KMS
- multipart SSE-S3
- multipart SSE-KMS
- compressed + encrypted
- range-sensitive sizes around
64 KiBand8 MiB
Next Stage
The current runner covers launch, upload, HEAD capture, and backend export.
The next iteration should focus on:
- proving the multi-disk backend path in the target local environment
- adding compressed fixtures to the automated matrix
- tightening exported tree selection if a narrower object-level slice becomes practical