fix(grpc-web): err when > client_body_buffer_size

[StarlightIbuki]

Summary

grpc-web uses get_raw_body() to fetch the request body, and does not handle the error when the body exceeds client_body_buffer_size.

Checklist

• The Pull Request has tests
• A changelog file has been created under changelog/unreleased/kong or skip-changelog label added on PR if changelog is unnecessary. README.md

Issue reference

Fix FTI-5624

[catbro666]

We should be careful with such blocking file io operations. If the file is large, it will block the whole nginx event loop significantly.

[StarlightIbuki]

That's the reason for emitting the warning message. And there's no better solution: we are not using a streaming JSON parser and the pb_unframe.

[StarlightIbuki]

A possible solution is to read with limited length and yield after every read, but it increase the memory and CPU usage.

[catbro666]

Yeah, this is a possible approach, or perhaps throwing it to another OS thread to avoid blocking the main worker thread.

[bungle]

A good candidate for:
https://github.com/openresty/lua-nginx-module#ngxrun_worker_thread

But that needs an addition to nginx.conf, such as:

thread_pool default threads=32 max_queue=65536;

[bungle]

grpc-web uses get_raw_body() to fetch the request body, and does not handle the error when the body exceeds client_body_buffer_size.

This is currently the generic design choice of Kong. Aka plugins and PDK does not read file buffered bodies. I think we should not start making specific plugin related changes, but perhaps allow reading files or continue not to allowl

e.g. should we just enable it with worker thread. Though buffering to file and then reading back to memory feels a bit counter intuitive. Aka why not then just grow the buffers.

[StarlightIbuki]

@catbro666 @bungle I agree that this is not an ideal solution. Even if we read from the file, it still uses that amount of memory. I'm confirming the reason why the customer does not want to increase the buffer size.

[catbro666]

@catbro666 @bungle I agree that this is not an ideal solution. Even if we read from the file, it still uses that amount of memory. I'm confirming the reason why the customer does not want to increase the buffer size.

These two are still different. The client body buffer will be allocated for every request as long as the whole request body was not pre-read in the header buffer(s), while reading from the file only happens when the client_body_buffer_size is not enough.

[StarlightIbuki]

Closing as the customer seems satisfied with the solution.