max request size limitation without buffer

[wbpcode]

Title: One line description

Now, all the the exist way to limit the max request body in the Envoy need to buffer the request body. But I won't whether make sense to add a featue to limit the request body without buffer body.

The request body will be proxied to upstream. But if the limitation is reached then the 413 will be returned and whole connection will be closed directly.

From the clients' view, the Envoy responded a 413, from the servers' view, Envoy cancelled the requests by close connection. This could limit the max request body size and need keep the body in the memory buffer.

🤔

[wufanqqfsc]

It's reasonable to return 413 ,but why we need close the connection ? The max request body size reached for this request but not mean other requests via this connection will also over size right ?

[wbpcode]

It's reasonable to return 413 ,but why we need close the connection ? The max request body size reached for this request but not mean other requests via this connection will also over size right ?

It depends on the protocol. HTTP1.1 will close the connection because the reponse will be complete before the response. For the HTTP2, the connection may be kept.

[KBaichoo]

To confirm the problem: we want to have a max request payload size while streaming the request to avoid buffering the entire request in memory at the proxy?

[wbpcode]

To confirm the problem: we want to have a max request payload size while streaming the request to avoid buffering the entire request in memory at the proxy?

yeah.

[wufanqqfsc]

It's reasonable to return 413 ,but why we need close the connection ? The max request body size reached for this request but not mean other requests via this connection will also over size right ?

It depends on the protocol. HTTP1.1 will close the connection because the reponse will be complete before the response. For the HTTP2, the connection may be kept.

What about the http1.1 with keep-alive?

[wbpcode]

It's reasonable to return 413 ,but why we need close the connection ? The max request body size reached for this request but not mean other requests via this connection will also over size right ?

It depends on the protocol. HTTP1.1 will close the connection because the reponse will be complete before the response. For the HTTP2, the connection may be kept.

What about the http1.1 with keep-alive?

HTTP1.1 has no frame or stream like HTTP2 on the top of the connection. So, if the response cannot end as normally, to avoid possible dirty data, the connection should be closed anyway.
All sendLocalReply of Envoy before requests or responses are completed as normally, may result in HTTP1.1 connection closing.

[wufanqqfsc]

OK, the proposed solution here is:
1.For downstream: envoy count and sum the received buffer size. And then compare with our target max body size. If over size then envoy sendLocalrepay to downstream with 413.
2.For upstream you want to close the connection to avoid more dirty data transfer ed to upstream service right? But anyway there is already some dirty data send to the upstream. So we want envoy to close the
connection. And for the upstream service should handle the connection lost event and clean the buffer data.

I just wonder how this body limit was done by other proxy like nginx . Is it possible to have the same behavior ? So that it will much easier to do proxy replacement and no sence for upstream service.

[github-actions]

This issue has been automatically marked as stale because it has not had activity in the last 30 days. It will be closed in the next 7 days unless it is tagged "help wanted" or "no stalebot" or other activity occurs. Thank you for your contributions.