akoshelev
commented
2 weeks ago Why do we go the route of having a BodyStream wrapper over BodyStreamInner?
Slicing must be done for both in-memory and axum streams, so BodyStream is where it is happening in one place. I could've implemented it twice separately, but I didn't see any benefit of duplicating the logic. Also, it is useful to have it done transparently for clients, so we don't need to worry about how the streams are constructed - chunking will apply regardless.
codecov[bot]
This fixes #1141 and adds a test that verifies that clients can submit inputs larger than 2GB ($2^{31}$ - 1). The test is quite slow (40 seconds), so I gated it behind an environment variable and only run it on Github. The probability of breaking it is quite low, so it is fine if it fails on CI only.
IPA uses two different stream types for in-memory and real-world infrastructure to submit bodies - one is Axum type and another one is just boxed byte stream. The fix must be applied to both because report_collector uses in-memory streams and helper binary and integration tests use Axum stream.
In order to have only one place that splits the incoming byte, both streams had to be hidden behind an umbrella type
BodyStreamthat works for both.