Support for Streaming Large File Uploads (GB-level) in Crow 1.2.0 (C++14 Environment)

[elecfreaks1]

Hi, I'm currently using Crow 1.2.0 in an embedded environment with C++14 support, and I'm trying to implement a file upload feature that supports large files (up to GB-level).

However, I noticed that the crow::request::body is returned as a std::string, which implies that the entire request body is read into memory at once. This behavior is problematic when dealing with large file uploads, as it may easily lead to OOM (Out of Memory) errors on resource-constrained embedded systems.

❓ My Questions:
Does Crow currently support streaming or chunked reading of the request body?
If not, are there any recommended workarounds or design patterns to implement streaming behavior manually?
Is there a plan to support asynchronous or stream-based request body handling in future versions?

[gittiver]

Does Crow currently support streaming or chunked reading of the request body?

No, this is currently not supported.

If not, are there any recommended workarounds or design patterns to implement streaming behavior manually?

The only way to have streaming of requests would be the web socket part.

Is there a plan to support asynchronous or stream-based request body handling in future versions?

its hard to give concrete dates for that. http parsing and chunking/ streaming are more complex items, currently I fix only smaller things due to lack of time for working on crow. But I have the feeling that http-parsing and http-connection need refactoring to be able to implement such requests.

So, no, there is no concrete plan.

[anton-n-petrov]

Work on this is in progress, split across two PRs (the original single PR, #1233, was superseded per review):

• Add a request body size limit that closes over-limit connections #1234 — a general request body size limit (app.max_body_size() / per-route override), so an oversized request is rejected at headers-complete instead of being buffered into req.body at all.
• Add a request body sink, with FileBodySink as a file-backed wrapper #1235 — a .body_sink() primitive (write()/finish()) that the parser feeds as body bytes arrive, with .body_file() as a file-backed wrapper built on it. The sink interface has no filesystem dependency, which should cover the embedded/resource-constrained case described above — a custom BodySink can hold the bytes however makes sense for the target instead of requiring a file.

Leaving this open until one or both land.