feat(dogstatsd): add configurable SO_RCVBUF for UDP socket

[duncanista]

What does this PR do?

Use the socket2 crate to create the UDP socket with a configurable receive buffer size (SO_RCVBUF).

On constrained environments like AWS Lambda, the kernel caps SO_RCVBUF at ~416 KiB (2x rmem_max), which can cause silent packet loss under burst traffic.

This change lets operators tune the buffer via so_rcvbuf.

Meant to be used with agents by DD_DOGSTATSD_SO_RCVBUF, matching the Go agent's configuration option.

Motivation

Avoid losing data.
SVLS-8549

Describe how to test/QA your changes

Manual testing and plugging it into other agents.

[Copilot]

Pull request overview

This PR adds an optional SO_RCVBUF configuration to the DogStatsD UDP listener by switching socket creation to socket2, enabling operators/consumers to increase the UDP receive buffer and reduce packet loss under burst traffic (e.g., constrained environments like AWS Lambda).

Changes:

• Add so_rcvbuf: Option<usize> to DogStatsDConfig and use a custom UDP socket creation path that can set SO_RCVBUF.
• Update call sites/tests to populate the new config field.
• Add socket2 dependency (and update lockfile accordingly).

Reviewed changes

Copilot reviewed 4 out of 5 changed files in this pull request and generated 8 comments.

Show a summary per file

File	Description
crates/dogstatsd/src/dogstatsd.rs	Adds so_rcvbuf config and new UDP socket creation using socket2 + logging.
crates/dogstatsd/tests/integration_test.rs	Updates test config structs to include the new so_rcvbuf field.
crates/dogstatsd/Cargo.toml	Adds socket2 dependency and expands Tokio features.
crates/datadog-serverless-compat/src/main.rs	Updates DogStatsD config construction to include so_rcvbuf.
Cargo.lock	Records new dependency resolution (adds socket2 0.5.x alongside existing socket2 0.6.x).

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The PR description mentions configuring this via DD_DOGSTATSD_SO_RCVBUF, but so_rcvbuf is still hard-coded to None here. If this binary is expected to expose the knob, wire so_rcvbuf from env/config (with parsing/validation) so the option is actually usable.

[duncanista]

Not adding an option a project I don't manage – ensuring that it works as before

[Copilot]

The socket is always created as IPv4 (Domain::IPV4). If addr is an IPv6 SocketAddr, bind will fail (and tokio::net::UdpSocket::bind previously supported IPv6). Choose the socket Domain based on whether the parsed SocketAddr is v4 or v6.

Suggested change

	let socket = Socket::new(Domain::IPV4, Type::DGRAM, Some(Protocol::UDP))?;
	let domain = if socket_addr.is_ipv4() {
	Domain::IPV4
	} else {
	Domain::IPV6
	};
	let socket = Socket::new(domain, Type::DGRAM, Some(Protocol::UDP))?;

[duncanista]

For now, we are always sending localhost addresses, might add in another PR

[Lewis-E]

Possibility of not-very-specific errors here, since the raw io::Error from bind() and set_nonblocking()
could look identical. Could be more specific with something like

socket.set_recv_buffer_size(buf_size).map_err(|e| {                                                       
      std::io::Error::new(e.kind(), format!("failed to set SO_RCVBUF to {}: {}", buf_size, e))              
  })?;                     
                                                                                                            
  socket.set_nonblocking(true).map_err(|e| {
      std::io::Error::new(e.kind(), format!("failed to set nonblocking: {}", e))
  })?;

  socket.bind(&socket_addr.into()).map_err(|e| {
      std::io::Error::new(e.kind(), format!("failed to bind to '{}': {}", addr, e))
  })?;

Claude Comment:

failed to bind to '127.0.0.1:8125': Address already in use: couldn't create UDP socket
vs the current output which would just be:
Address already in use: couldn't create UDP socket

Non-blocking comment, low change of things breaking, but the socket2 docs do specifically mention they don't do error handling.

[duncanista]

Too verbose so far, might improve this in another PR with a cleaner approach