Issue #17933 has been updated by byroot (Jean Boussier).


> Are you talking about moving the write_timeout implementation from Net::BufferedIO to IO.copy_stream?

No. `IO.copy_stream` is a low level API that leverage `sendfile(2)` and other similar APIs so that the data transfer happens in the Kernel rather than in user space. For large uploads the performance difference can be quite massive.

However this low-level API has no timeout support. The discussion was about adding timeout support to that so that Net::HTTP could pass a timeout argument.

----------------------------------------
Bug #17933: `Net::HTTP#write_timeout` doesn't work with `body_stream`
https://bugs.ruby-lang.org/issues/17933#change-92300

* Author: miguelfteixeira (Miguel Teixeira)
* Status: Open
* Priority: Normal
* ruby -v: ruby 2.6.3p62 (2019-04-16 revision 67580) [universal.x86_64-darwin20]
* Backport: 2.6: UNKNOWN, 2.7: UNKNOWN, 3.0: UNKNOWN
----------------------------------------
While testing `Net::HTTP#write_timeout` using the server example from the [Feature Issue](https://bugs.ruby-lang.org/issues/13396), I've noticed that [Faraday](https://github.com/lostisland/faraday) multipart requests (with a file parameter) didn't trigger `WriteTimeout`. The root cause is that [Net::HTTPGenericRequest#send_request_with_body_stream](https://github.com/ruby/ruby/blob/v2_6_7/lib/net/http/generic_request.rb#L205-L207) is not using `Net::BufferedIO` (the class that implements the write timeout).

The following patch fixes the issue, but looking at the comments, it's unclear if it will break some existing functionality. Albeit all the tests pass.

```diff
--- a/lib/net/http/generic_request.rb
+++ b/lib/net/http/generic_request.rb
@@ -204,7 +204,7 @@ def send_request_with_body_stream(sock, ver, path, f)
     else
       # copy_stream can sendfile() to sock.io unless we use SSL.
       # If sock.io is an SSLSocket, copy_stream will hit SSL_write()
-      IO.copy_stream(f, sock.io)
+      IO.copy_stream(f, sock)
     end
   end
```

```bash
  ruby git:(fix-net-http-write-timeout-body-stream)  ../mspec/bin/mspec run library/net/
ruby 2.6.3p62 (2019-04-16 revision 67580) [universal.x86_64-darwin20]
[- | ==================100%================== | 00:00:00]      0F      0E

Finished in 2.029623 seconds

187 files, 876 examples, 1214 expectations, 0 failures, 0 errors, 0 tagged
```

Any thoughts on this? It looks like `Net::HTTP#write_timeout` is partially implemented, and we should ensure that it either works on all use cases or an error is raised when it's used in a non-supported use case.

I'm more than happy to help with the implementation. If this patch is the correct approach, I can also create new unit tests to assert the proper behaviour.



-- 
https://bugs.ruby-lang.org/

Unsubscribe: <mailto:ruby-core-request / ruby-lang.org?subject=unsubscribe>
<http://lists.ruby-lang.org/cgi-bin/mailman/options/ruby-core>