TODO Working Group K. Pugin Internet-Draft A. Frindell Intended status: Informational J. Cenzano Expires: 13 January 2022 J. Weissman Facebook 12 July 2021 RUSH - Reliable (unreliable) streaming protocol draft-kpugin-rush-00 Abstract RUSH is an application-level protocol for ingesting live video. This document describes core of the protocol and how it maps onto QUIC Discussion Venues This note is to be removed before publishing as an RFC. Discussion of this document takes place on the mailing list (), which is archived at . Source for this draft and an issue tracker can be found at https://github.com/afrind/draft-rush. Status of This Memo This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet- Drafts is at https://datatracker.ietf.org/drafts/current/. 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." This Internet-Draft will expire on 13 January 2022. Copyright Notice Copyright (c) 2021 IETF Trust and the persons identified as the document authors. All rights reserved. Pugin, et al. Expires 13 January 2022 [Page 1] Internet-Draft rush July 2021 This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://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. Table of Contents 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 2 2. Conventions and Definitions . . . . . . . . . . . . . . . . . 3 3. Theory of Operations . . . . . . . . . . . . . . . . . . . . 3 3.1. Connection establishment . . . . . . . . . . . . . . . . 3 3.2. Sending Video Data . . . . . . . . . . . . . . . . . . . 4 3.3. Receiving data . . . . . . . . . . . . . . . . . . . . . 4 3.4. Reconnect . . . . . . . . . . . . . . . . . . . . . . . . 5 4. Wire Format . . . . . . . . . . . . . . . . . . . . . . . . . 5 4.1. Frame Header . . . . . . . . . . . . . . . . . . . . . . 5 4.2. Frames . . . . . . . . . . . . . . . . . . . . . . . . . 8 4.2.1. Connect frame . . . . . . . . . . . . . . . . . . . . 8 4.2.2. Connect Ack frame . . . . . . . . . . . . . . . . . . 8 4.2.3. End of Video frame . . . . . . . . . . . . . . . . . 9 4.2.4. Error frame . . . . . . . . . . . . . . . . . . . . . 9 4.2.5. Video frame . . . . . . . . . . . . . . . . . . . . . 10 4.2.6. Audio frame . . . . . . . . . . . . . . . . . . . . . 11 4.2.7. GOAWAY frame . . . . . . . . . . . . . . . . . . . . 13 4.3. Quic Mapping . . . . . . . . . . . . . . . . . . . . . . 13 4.3.1. Normal mode . . . . . . . . . . . . . . . . . . . . . 13 4.3.2. Multi Stream Mode . . . . . . . . . . . . . . . . . . 13 5. Error Handling . . . . . . . . . . . . . . . . . . . . . . . 14 5.1. Connection Errors . . . . . . . . . . . . . . . . . . . . 14 5.2. Frame errors . . . . . . . . . . . . . . . . . . . . . . 14 6. Extensions . . . . . . . . . . . . . . . . . . . . . . . . . 15 7. Security Considerations . . . . . . . . . . . . . . . . . . . 15 8. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 15 9. Normative References . . . . . . . . . . . . . . . . . . . . 15 Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . 16 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 16 1. Introduction RUSH is bidirectional application level protocol designed for live video ingestion that runs on top of QUIC. Pugin, et al. Expires 13 January 2022 [Page 2] Internet-Draft rush July 2021 RUSH was built as a replacement for RTMP (Real-Time Messaging Protocol) with the goal to provide support for new audio and video codecs, extensibility in the form of new message types, and multi- track support. In addition, RUSH gives applications option to control data delivery guarantees by utilizing QUIC streams. This document describes the RUSH protocol, wire format, and QUIC mapping. 2. Conventions and Definitions The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here. Frame/Message: logical unit of information that client and server can exchange PTS: presentation timestamp DTS: decoding timestamp AAC: advanced audio codec NALU: network abstract layer unit VPS: video parameter set (H265 video specific NALU) SPS: sequence parameter set (H264/H265 video specific NALU) PPS: picture parameter set (H264/H265 video specific NALU) ADTS header: _Audio Data Transport Stream Header_ ASC: Audio specific config GOP: Group of pictures, specifies the order in which intra- and inter-frames are arranged. 3. Theory of Operations 3.1. Connection establishment In order to live stream using RUSH, the client establishes a QUIC connection using the ALPN token "rush". Pugin, et al. Expires 13 January 2022 [Page 3] Internet-Draft rush July 2021 After the QUIC connection is established, client creates a new bidirectional QUIC stream, choses starting frame ID and sends "Connect" frame Section 4.2.1 over that stream. This stream is called the Connect Stream. The client sends "mode of operation" setting in "Connect" frame payload, format of the payload is "TBD". One connection SHOULD only be used to send one video. 3.2. Sending Video Data The client can choose to wait for the "ConnectAck" frame Section 4.2.2 or it can start sending data immediately after sending the "Connect" frame. A track is a logical organization of the data, for example, video can have one video track, and two audio tracks (for two languages). The client can send data for multiple tracks simultaneously. The encoded audio or video data of each track is serialized into frames (see Section 4.2.6 or Section 4.2.5) and transmitted from the client to the server. Each track has its own monotonically increasing frame ID sequence. The client MUST start with initial frame ID = 1. Depending on mode of operation (Section 4.3), the client sends audio and video frames on the Connect stream or on a new QUIC stream for each frame. In "Multi Stream Mode" (Section 4.3.2), the client can stop sending a frame by resetting the corresponding QUIC stream. In this case, there is no guarantee that the frame was received by the server. 3.3. Receiving data Upon receiving "Connect" frame, the server replies with "ConnectAck" frame Section 4.2.2 and prepares to receive audio/video data. It's possible that in "Multi Stream Mode" (Section 4.3.2), the server receives audio or video data before it receives the "Connect" frame. The implementation can choose whether to buffer or drop the data. The audio/video data cannot be interpreted correctly before the arrival of the "Connect" frame. In "Normal Mode" (Section 4.3.1), it is guaranteed by the transport that frames arrive into the application layer in order they were sent. Pugin, et al. Expires 13 January 2022 [Page 4] Internet-Draft rush July 2021 In "Multi Stream Mode", it's possible that frames arrive at the application layer in a different order than they were sent, therefore the server MUST keep track of last received frame ID for every track that it receives. A gap in the frame sequence ID on a given track can indicate out of order delivery and the server MAY wait until missing frames arrive. The server must consider frame lost if the corresponding QUIC stream was reset. Upon detecting a gap in the frame sequence, the server MAY wait for the missing frames to arrive for an implementation defined time. If missing frames don't arrive, the server SHOULD consider them lost and continue processing rest of the frames. For example if the server receives the following frames for track 1: "1 2 3 5 6" and frame "#4" hasn't arrived after implementation defined timeout, thee server SHOULD continue processing frames "5" and "6". When the client is done streaming, it sends the "End of Video" frame (Section 4.2.3) to indicate to the server that there won't be any more data sent. 3.4. Reconnect If the QUIC connection is closed at any point, client MAY reconnect by simply repeat the "Connection establishment" process (Section 3.1) and resume sending the same video where it left off. In order to support termination of the new connection by a different server, the client SHOULD resume sending video frames starting with I-frame, to guarantee that the video track can be decoded. Reconnect can be initiated by the server if it needs to "go away" for maintenance. In this case, the server sends a "GOAWAY" frame (Section 4.2.7) to advise the client to gracefully close the connection. This allows client to finish sending some data and establish new connection to continue sending without interruption. 4. Wire Format 4.1. Frame Header The client and server exchange information using frames. There are different types of frames and the payload of each frame depends on its type. Generic frame format: Pugin, et al. Expires 13 January 2022 [Page 5] Internet-Draft rush July 2021 0 1 2 3 4 5 6 7 +--------------------------------------------------------------+ | Length (64) | +--------------------------------------------------------------+ | ID (64) | +-------+------------------------------------------------------+ |Type(8)| Payload ... | +-------+------------------------------------------------------+ Length(64)`: Each frame starts with length field, 64 bit size that tells size of the frame in bytes (including predefined fields, so if LENGTH is 100 bytes, then PAYLOAD length is 100 - 8 - 8 - 1 = 82 bytes). ID(64): 64 bit frame sequence number, every new frame MUST have a sequence ID greater than that of the previous frame within the same track. Track ID would be specified in each frame. If track ID is not specified it's 0 implicitly. Type(8): 1 byte representing the type of the frame. Predefined frame types: Pugin, et al. Expires 13 January 2022 [Page 6] Internet-Draft rush July 2021 +============+====================+ | Frame Type | Frame | +============+====================+ | 0x0 | connect frame | +------------+--------------------+ | 0x1 | connect ack frame | +------------+--------------------+ | 0x2 | reserved | +------------+--------------------+ | 0x3 | reserved | +------------+--------------------+ | 0x4 | end of video frame | +------------+--------------------+ | 0x5 | error frame | +------------+--------------------+ | 0x6 | reserved | +------------+--------------------+ | 0x7 | reserved | +------------+--------------------+ | 0x8 | reserved | +------------+--------------------+ | 0x9 | reserved | +------------+--------------------+ | 0xA | reserved | +------------+--------------------+ | 0XB | reserved | +------------+--------------------+ | 0xC | reserved | +------------+--------------------+ | 0xD | video frame | +------------+--------------------+ | 0xE | audio frame | +------------+--------------------+ | 0XF | reserved | +------------+--------------------+ | 0X10 | reserved | +------------+--------------------+ | 0x11 | reserved | +------------+--------------------+ | 0x12 | reserved | +------------+--------------------+ | 0x13 | reserved | +------------+--------------------+ | 0x14 | GOAWAY frame | +------------+--------------------+ Table 1 Pugin, et al. Expires 13 January 2022 [Page 7] Internet-Draft rush July 2021 4.2. Frames 4.2.1. Connect frame +--------------------------------------------------------------+ | Length (64) | +--------------------------------------------------------------+ | ID (64) | +-------+-------+---------------+---------------+--------------+ | 0x0 |Version|Video Timescale|Audio Timescale| | +-------+-------+---------------+---------------+--------------+ | Live Session ID(64) | +--------------------------------------------------------------+ | Payload ... | +--------------------------------------------------------------+ Version: version of the protocol (initial version is 0x0). Video Timescale: timescale for all video frame timestamps on this connection. Recommended value 30000 Audio Timescale: timescale for all audio samples timestamps on this connection, recommended value same as audio sample rate, for example 44100 Live Session ID: identifier of broadcast, when reconnect, client MUST use the same live session ID Payload: application and version specific data that can be used by the server. OPTIONAL This frame is used by the client to initiate broadcasting. The client can start sending other frames immediately after "Connect frame" without waiting acknowledgement from the server. If server doesn't support VERSION sent by the client, the server sends an Error frame with code "UNSUPPORTED VERSION". If audio timescale or video timescale are 0, the server sends error frame with error code "INVALID FRAME FORMAT" and closes connection. If the client receives a Connect frame from the server, the client sends an Error frame with code "TBD". 4.2.2. Connect Ack frame Pugin, et al. Expires 13 January 2022 [Page 8] Internet-Draft rush July 2021 0 1 2 3 4 5 6 7 +--------------------------------------------------------------+ | 17 | +--------------------------------------------------------------+ | ID (64) | +-------+------------------------------------------------------+ | 0x1 | +-------+ The server sends the "Connect Ack" frame in response to "Connect" frame indicating that server accepts "version" and is ready to receive data. If the client doesn't receive "Connect Ack" frame from the server within a timeout, it will close the connection. The timeout value is chosen by the implementation. There can be only one "Connect Ack" frame sent over lifetime of the QUIC connection. If the server receives a Connect Ack frame from the client, the client sends an Error frame with code "TBD". 4.2.3. End of Video frame +--------------------------------------------------------------+ | 17 | +--------------------------------------------------------------+ | ID (64) | +-------+------------------------------------------------------+ | 0x4 | +-------+ End of Video frame is sent by a client when it's done sending data and is about to close the connection. The server SHOULD ignore all frames sent after that. 4.2.4. Error frame Pugin, et al. Expires 13 January 2022 [Page 9] Internet-Draft rush July 2021 +--------------------------------------------------------------+ | 29 | +--------------------------------------------------------------+ | ID (64) | +-------+------------------------------------------------------+ | 0x5 | +-------+------------------------------------------------------+ | Sequence ID (64) | +------------------------------+-------------------------------+ | Error Code (32) | +------------------------------+ Sequence ID: ID of the frame sent by the client that error is generated for, ID=0x0 indicates connection level error. Error Code: 32 bit unsigned integer Error frame can be sent by the client or the server to indicate that an error occurred. Some errors are fatal and the connection will be closed after sending the Error frame. 4.2.5. Video frame +--------------------------------------------------------------+ | Length (64) | +--------------------------------------------------------------+ | ID (64) | +-------+-------+----------------------------------------------+ | 0xD | Codec | +-------+-------+----------------------------------------------+ | PTS (64) | +--------------------------------------------------------------+ | Track ID (64) | +---------------+----------------------------------------------+ | I-Frame ID Offset | Video Data ... | +---------------+----------------------------------------------+ Codec: specifies codec that was used to encode this frame. PTS: presentation timestamp in connection video timescale DTS: decoding timestamp in connection video timescale Supported type of codecs: Pugin, et al. Expires 13 January 2022 [Page 10] Internet-Draft rush July 2021 +======+=======+ | Type | Codec | +======+=======+ | 0x1 | H264 | +------+-------+ | 0x2 | H265 | +------+-------+ | 0x3 | VP8 | +------+-------+ | 0x4 | VP9 | +------+-------+ Table 2 Track ID: ID of the track that this frame is on I-Frame ID Offset: Distance from sequence ID of the I-frame that is required before this frame can be decoded. This can be useful to decide if frame can be dropped. Video Data: variable length field, that carries actual video frame data that is codec dependent For h264/h265 codec, "Video Data" are 1 or more NALUs in AVCC format: 0 1 2 3 4 5 6 7 +--------------------------------------------------------------+ | NALU Length (64) | +--------------------------------------------------------------+ | NALU Data ... +--------------------------------------------------------------+ EVERY h264 video key-frame MUST start with SPS/PPS NALUs. EVERY h265 video key-frame MUST start with VPS/SPS/PPS NALUs. Binary concatenation of "video data" from consecutive video frames, without data loss MUST produce VALID h264/h265 bitstream. 4.2.6. Audio frame Pugin, et al. Expires 13 January 2022 [Page 11] Internet-Draft rush July 2021 +--------------------------------------------------------------+ | Length (64) | +--------------------------------------------------------------+ | ID (64) | +-------+------------------------------------------------------+ | 0xE | Codec | +-------+-------+----------------------------------------------+ | Timestamp (64) | +-------+------------------------------------------------------+ |TrackID| +-------+------------------------------------------------------+ | Audio Data ... +--------------------------------------------------------------+ Codec: specifies codec that was used to encode this frame. Supported type of codecs: +======+=======+ | Type | Codec | +======+=======+ | 0x1 | AAC | +------+-------+ | 0x2 | OPUS | +------+-------+ Table 3 Timestamp: timestamp of first audio sample in Audio Data. Track ID: ID of the track that this frame is on Audio Data: variable length field, that carries 1 or more audio frames that is codec dependent. For AAC codec, "Audio Data" are 1 or more AAC samples, prefixed with ADTS HEADER: 152 158 ... N +---+---+---+---+---+---+---+... | ADTS(56) | AAC SAMPLE | +---+---+---+---+---+---+---+... Binary concatenation of all AAC samples in "Audio Data" from consecutive audio frames, without data loss MUST produce VALID AAC bitstream. Pugin, et al. Expires 13 January 2022 [Page 12] Internet-Draft rush July 2021 For OPUS codec, "Audio Data" are 1 or more OPUS samples, prefixed with OPUS header as defined in [RFC7845] 4.2.7. GOAWAY frame 0 1 2 3 4 5 6 7 +--------------------------------------------------------------+ | 17 | +--------------------------------------------------------------+ | ID (64) | +-------+------------------------------------------------------+ | 0x14 | +-------+ The GOAWAY frame is used by the server to initiate graceful shutdown of a connection, for example, for server maintenance. Upon receiving GOAWAY, the client MUST send frames remaining in current GOP and stop sending new frames on this connection. The client SHOULD establish a new connection and resume sending frames there. After sending a GOAWAY frame, the server continues processing arriving frames for an implementation defined time, after which the server SHOULD close the connection. 4.3. Quic Mapping One of the main goals of the RUSH protocol was ability to provide applications a way to control reliability of delivering audio/video data. This is achieved by using a special mode Section 4.3.2. 4.3.1. Normal mode In normal mode, RUSH uses one bidirectional QUIC stream to send data and receive data. Using one stream guarantees reliable, in-order delivery - applications can rely on QUIC transport layer to retransmit lost packets. The performance characteristics of this mode are similar to RTMP over TCP. 4.3.2. Multi Stream Mode In normal mode, if packet belonging to video frame is lost, all packets sent after it will not be delivered to application, even though those packets may have arrived at the server. This introduces head of line blocking and can negatively impact latency. Pugin, et al. Expires 13 January 2022 [Page 13] Internet-Draft rush July 2021 To address this problem, RUSH defines "Multi Stream Mode", in which one QUIC stream is used per audio/video frame. Connection establishment follows the normal procedure by client sending Connect frame, after that Video and Audio frames are sent using following rules: * Each new frame is sent on new bidirectional QUIC stream * Frames within same track must have IDs that are monotonically increasing, such that ID(n) = ID(n-1) + 1 The receiver reconstructs the track using the frames IDs. Response Frames (Connect Ack and Error), will be in the response stream of the stream that sent it. The client MAY control delivery reliability by setting a delivery timer for every audio or video frame and reset the QUIC stream when the timer fires. This will effectively stop retransmissions if the frame wasn't fully delivered in time. Timeout is implementation defined, however future versions of the draft will define a way to negotiate it. 5. Error Handling An endpoint that detects an error SHOULD signal the existence of that error to its peer. Errors can affect an entire connection (see Section 5.1), or a single frame (see Section 5.2). The most appropriate error code SHOULD be included in the error frame that signals the error. 5.1. Connection Errors There is one error code defined in core of the protocol that indicates connection error: 1 - UNSUPPORTED VERSION - indicates that the server doesn't support version specified in Connect frame 5.2. Frame errors There are two error codes defined in core protocol that indicate a problem with a particular frame: Pugin, et al. Expires 13 January 2022 [Page 14] Internet-Draft rush July 2021 2 - UNSUPPORTED CODEC - indicates that the server doesn't support the given audio or video codec 3 - INVALID FRAME FORMAT - indicates that the receiver was not able to parse the frame or there was an issue with a field's value. 6. Extensions RUSH permits extension of the protocol. Extensions are permitted to use new frame types (Section 4), new error codes (Section 4.2.4), or new audio and video codecs (Section 4.2.6, Section 4.2.5). Implementations MUST ignore unknown or unsupported values in all extensible protocol elements, except "codec id", which returns an UNSUPPORTED CODEC error. Implementations MUST discard frames that have unknown or unsupported types. 7. Security Considerations RUSH protocol relies on security guarantees provided by the transport. Implementation SHOULD be prepare to handle cases when sender deliberately sends frames with gaps in sequence IDs. Implementation SHOULD be prepare to handle cases when server never receives Connect frame (Section 4.2.1). A frame parser MUST ensure that value of frame length field (see Section 4.1) matches actual length of the frame, including the frame header. Implementation SHOULD be prepare to handle cases when sender sends a frame with large frame length field value. 8. IANA Considerations TODO: add frame type registry, error code registry, audio/video codecs registry 9. Normative References [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997, . Pugin, et al. Expires 13 January 2022 [Page 15] Internet-Draft rush July 2021 [RFC7845] Terriberry, T., Lee, R., and R. Giles, "Ogg Encapsulation for the Opus Audio Codec", RFC 7845, DOI 10.17487/RFC7845, April 2016, . [RFC8174] Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174, May 2017, . Acknowledgments This draft is the work of many people: Vlad Shubin, Nitin Garg, Milen Lazarov, Benny Luo, Nick Ruff, Konstantin Tsoy, Nick Wu. Authors' Addresses Kirill Pugin Facebook Email: ikir@fb.com Alan Frindell Facebook Email: afrind@fb.com Jordi Cenzano Facebook Email: jcenzano@fb.com Jake Weissman Facebook Email: jakeweissman@fb.com Pugin, et al. Expires 13 January 2022 [Page 16]