Network Working Group W. Zhu INTERNET-DRAFT Google, Inc. Intended Status: Informational M. Jennings Expires: November 18, 2010 May 17, 2010 Implications of Full-Duplex HTTP draft-zhu-http-fullduplex-00.txt Status of this Memo This Internet-Draft is submitted to IETF in full conformance with the provisions of BCP 78 and BCP 79. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet-Drafts. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress." The list of current Internet-Drafts can be accessed at http://www.ietf.org/1id-abstracts.html The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html Copyright and License Notice Copyright (c) 2010 IETF Trust and the persons identified as the document authors. All rights reserved. This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License. Zhu & Jennings Expires November 18, 2010 [Page 1] INTERNET DRAFT Implications of Full-Duplex HTTP May 17, 2010 Abstract Full-duplex HTTP follows the basic HTTP request-response semantics but also allows the server to send the response to the client at the same time when the client is transmitting the request to the server. Requirements for Full-duplex HTTP are under-specified in the existing HTTP specification, and this memo intends to clarify the behaviors of full-duplex HTTP under the standard HTTP protocol. Table of Contents 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.1 Terminology . . . . . . . . . . . . . . . . . . . . . . . . 3 2 Streaming in HTTP . . . . . . . . . . . . . . . . . . . . . . . 3 2.1 Request and Response Streaming . . . . . . . . . . . . . . 3 2.2 Full-Duplex Streaming . . . . . . . . . . . . . . . . . . . 4 3 Protocol Considerations . . . . . . . . . . . . . . . . . . . . 4 3.1 Initialization . . . . . . . . . . . . . . . . . . . . . . 4 3.2 Termination . . . . . . . . . . . . . . . . . . . . . . . . 4 3.3 Persistent Connections . . . . . . . . . . . . . . . . . . 5 3.4 Time-Out . . . . . . . . . . . . . . . . . . . . . . . . . 5 3.5 Proxies . . . . . . . . . . . . . . . . . . . . . . . . . . 5 3.6 Errors . . . . . . . . . . . . . . . . . . . . . . . . . . 5 4 Application Considerations . . . . . . . . . . . . . . . . . . . 6 4.1 Compatibility . . . . . . . . . . . . . . . . . . . . . . . 6 4.2 Fall Back and Buffering . . . . . . . . . . . . . . . . . . 6 4.2 Messaging . . . . . . . . . . . . . . . . . . . . . . . . . 6 5 Security Considerations . . . . . . . . . . . . . . . . . . . . 7 6 IANA Considerations . . . . . . . . . . . . . . . . . . . . . . 7 7 Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . 7 8 Informative References . . . . . . . . . . . . . . . . . . . . . 7 Author's Addresses . . . . . . . . . . . . . . . . . . . . . . . . 7 Zhu & Jennings Expires November 18, 2010 [Page 2] INTERNET DRAFT Implications of Full-Duplex HTTP May 17, 2010 1 Introduction HTTP [RFC2616], being a RPC style protocol, requires communication between client and server follows the strict request-response pattern. HTTP may also be used to stream data from either client or server. When bi-directional streaming is required, two connections are often used to stream client and server data separately. Using two separate connections not only introduce overhead, but also make HTTP insufficient to be used as a standalone protocol, i.e. application- level protocols are required to handle the two connections. However, if the server is allowed to send response to the client at the same time when request is being transmitted from the client to the server, then effectively full-duplex streaming becomes possible under the standard HTTP protocol [RFC2616]. Full-duplex streaming requires end-to-end support from both the client and server. More specifically, the client has to be explicitly designed to support such capability. Given the unique properties of full-duplex HTTP, special requirements exist for both the client and server. And those requirements need be clearly identified so that implementations are able to follow standard behaviors in adopting full-duplex HTTP. 1.1 Terminology The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119 [RFC2119]. 2 Streaming in HTTP 2.1 Request and Response Streaming Request streaming requires the server to deliver to the application any data that has been received by the server. Response streaming requires the server to send to the client any data that has been generated by the application. Chunked transfer-encoding [RFC2616] is expected for both request and response streaming. However, client or server should not design the underlying streaming or messaging API based on the chunked transfer- encoding (which is generated hop-by-hop). Zhu & Jennings Expires November 18, 2010 [Page 3] INTERNET DRAFT Implications of Full-Duplex HTTP May 17, 2010 Most browsers don't support chunked transfer-encoding for requests. 2.2 Full-Duplex Streaming When full-duplex streaming is enabled, request and response are transmitted between client and server simultaneously over the same HTTP connection. Full-duplex streaming may be applied to any resource that is designed to concurrently stream request and response, for example a voice translator. Full-duplex streaming should still follow the standard request- response semantics and maintain the basic temporal dependency between request and response: 1 Server SHOULD NOT generate any response until a request has been received. That is, unsolicited server-initiated response is not allowed. 2 Server SHOULD NOT complete the response before the client completes the request. That is, the server can only complete the response after the server has received the complete request from the client 3 Protocol Considerations 3.1 Initialization Full-duplex streaming may be started as soon as the server receives the first byte of the request body. This behavior is significantly different from the technique commonly-known as Hanging GET, in which a separate GET request is issued to initiate streaming to the client. 3.2 Termination The client terminates request streaming by completing the request, i.e. sending out the last-chunk [RFC2616]. And the server may choose to continue streaming the response after it receives the complete request from the client, as decided by the application. Eventually the server terminates response streaming by sending out the last- chunk of response. Before the server receives the complete request from the client, the only way for the server to terminate the response streaming is to close the connection. It is considered an illegal state for an HTTP connection to have a pending request when the response has already been completed. The client should close the connection immediately if Zhu & Jennings Expires November 18, 2010 [Page 4] INTERNET DRAFT Implications of Full-Duplex HTTP May 17, 2010 the client receives a complete response when the client is still streaming the request. 3.3 Persistent Connections It's important that HTTP keep-alive and pipelining still work with full-duplex HTTP. To achieve that, the client and server should respect the original semantics of HTTP persistent connections [RFC2616]. 3.4 Time-Out HTTP server or client may time-out connections while waiting for request or response. Full-duplex HTTP should not override this behavior. A connection may be closed due to time-out if either request or response streaming is inactive. Full-duplex HTTP introduces no special requirement on timeout of the underlying TPC connection. When time-out does happen both request and response streaming will be terminated. 3.5 Proxies HTTP proxies may not support concurrent responses, and one of the purposes of this document is to increase awareness of full-duplex HTTP communication in proxies. Proxies may also buffer streamed requests or responses, or have problems to handle chunked transfer-encoding, especially for requests. 3.6 Errors According to RFC2616 [RFC2616], a client or proxy should close the connection if the client receives an error response as it is transmitting the request. Full-duplex HTTP must respect this requirement. In cases when a server is incapable of streaming response or decides to timeout, the server should close the connection. This is also true for a client when it is streaming request. A client or server should stop streaming any new data after it notices that the underlying TCP connection has been closed by the other party. In-flight data will be discarded under this kind of half-close behavior. Zhu & Jennings Expires November 18, 2010 [Page 5] INTERNET DRAFT Implications of Full-Duplex HTTP May 17, 2010 4 Application Considerations 4.1 Compatibility Full-duplex streaming is completely controlled by the server application, and should only be enabled to clients that have been explicitly identified by the server. It is not sufficient to enable full-duplex HTTP solely based on User- Agent. For non-controlled client applications, the client needs advise its capability of full-duplex streaming via URL parameters or headers (for example, "X-Content-Streaming: full-duplex, timeout=30"). Otherwise, full-duplex streaming should be disabled, or the server should return 404 (Not Found) status if full-duplex streaming is mandatory for the requested resource. 4.2 Fall Back and Buffering Proxies may disallow early responses, or buffer requests or responses. In such a case, the application may have to switch to a different protocol that uses two connections and rely on polling techniques. One efficient way to trigger a fall back will be for the client to wait for initial response for a short time-out period. When a proxy or client is unable to consume early response data, the server may suffer from buffer overflow. And the server should close the connection as soon as it detects such a condition. Generally there is no reliable way for a server to distinguish between ill- behaved clients and non-compatible proxies. Full-duplex HTTP expects minimized buffering from both client and server. However, applications may choose to buffer certain amount of data for optimization or application specific purposes. 4.2 Messaging Since chunked transfer-encoding isn't a reliable way to provide message framing, messaging support has to be provided by the application stack, along with any required message delivery guarantee. Zhu & Jennings Expires November 18, 2010 [Page 6] INTERNET DRAFT Implications of Full-Duplex HTTP May 17, 2010 5 Security Considerations Full-duplex HTTP introduces no new security concerns beyond those known with regular HTTP communication. 6 IANA Considerations This document does not require any actions by the IANA. 7 Acknowledgments Thanks to Henrik Nordstrom, Jamie Lokier, and Mark Nottingham for their feedbacks when the subject of this document was originally brought up on httpbis. 8 Informative References [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997. [RFC2616] Fielding, R., Gettys, J., Mogul, J., Frystyk, H., Masinter, L., Leach, P., and T. Berners-Lee, "Hypertext Transfer Protocol -- HTTP/1.1", RFC 2616, June 1999. Author's Addresses Wenbo Zhu Google, Inc. 1600 Amphitheatre Parkway Mountain View, CA 94043 US Phone: +1 650 214 5894 Email: wenboz@google.com Mike Jennings Email: mike.c.jennings@gmail.com Zhu & Jennings Expires November 18, 2010 [Page 7]