vortex_ipc/messages/decoder.rs
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271
use std::fmt::Debug;
use bytes::Buf;
use flatbuffers::{root, root_unchecked};
use itertools::Itertools;
use vortex_array::parts::ArrayParts;
use vortex_buffer::{AlignedBuf, Alignment, ByteBuffer};
use vortex_dtype::DType;
use vortex_error::{vortex_bail, vortex_err, VortexExpect, VortexResult};
use vortex_flatbuffers::message::{MessageHeader, MessageVersion};
use vortex_flatbuffers::{message as fb, FlatBuffer};
use crate::ALIGNMENT;
/// A message decoded from an IPC stream.
///
/// Note that the `Array` variant cannot fully decode into an [`vortex_array::Array`] without
/// a [`vortex_array::ContextRef`] and a [`DType`]. As such, we partially decode into an
/// [`ArrayParts`] and allow the caller to finish the decoding.
#[derive(Debug)]
pub enum DecoderMessage {
Array(ArrayParts),
Buffer(ByteBuffer),
DType(DType),
}
#[derive(Default)]
enum State {
#[default]
Length,
Header(usize),
Array(ReadingArray),
Buffer(ReadingBuffer),
}
struct ReadingArray {
header: FlatBuffer,
buffers_length: usize,
}
struct ReadingBuffer {
length: usize,
length_with_padding: usize,
alignment: Alignment,
}
#[derive(Debug)]
pub enum PollRead {
Some(DecoderMessage),
/// Returns the _total_ number of bytes needed to make progress.
/// Note this is _not_ the incremental number of bytes needed to make progress.
NeedMore(usize),
}
// NOTE(ngates): we should design some trait that the Decoder can take that doesn't require unique
// ownership of the underlying bytes. The decoder needs to split out bytes, and advance a cursor,
// but it doesn't need to mutate any bytes. So in theory, we should be able to do this zero-copy
// over a shared buffer of bytes, instead of requiring a `BytesMut`.
/// A stateful reader for decoding IPC messages from an arbitrary stream of bytes.
pub struct MessageDecoder {
/// The minimum alignment to use when reading a data `Buffer`.
alignment: Alignment,
/// The current state of the decoder.
state: State,
}
impl Default for MessageDecoder {
fn default() -> Self {
Self {
alignment: ALIGNMENT.into(),
state: Default::default(),
}
}
}
impl MessageDecoder {
/// Attempt to read the next message from the bytes object.
///
/// If the message is incomplete, the function will return `NeedMore` with the _total_ number
/// of bytes needed to make progress. The next call to read_next _should_ provide at least
/// this number of bytes otherwise it will be given the same `NeedMore` response.
pub fn read_next<B: AlignedBuf>(&mut self, bytes: &mut B) -> VortexResult<PollRead> {
loop {
match &self.state {
State::Length => {
if bytes.remaining() < 4 {
return Ok(PollRead::NeedMore(4));
}
let msg_length = bytes.get_u32_le();
self.state = State::Header(msg_length as usize);
}
State::Header(msg_length) => {
if bytes.remaining() < *msg_length {
return Ok(PollRead::NeedMore(*msg_length));
}
let msg_bytes = bytes.copy_to_const_aligned(*msg_length);
let msg = root::<fb::Message>(msg_bytes.as_ref())?;
if msg.version() != MessageVersion::V0 {
vortex_bail!("Unsupported message version {:?}", msg.version());
}
match msg.header_type() {
MessageHeader::ArrayMessage => {
let array_msg = msg
.header_as_array_message()
.vortex_expect("array message header");
let buffers_length: u64 = array_msg
.buffers()
.unwrap_or_default()
.iter()
.map(|buffer| buffer.length() + (buffer.padding() as u64))
.sum();
let buffers_length = usize::try_from(buffers_length).map_err(|_| {
vortex_err!("buffers length is too large for usize")
})?;
self.state = State::Array(ReadingArray {
header: msg_bytes,
buffers_length,
});
}
MessageHeader::Buffer => {
let buffer = msg.header_as_buffer().vortex_expect("buffer header");
let length = usize::try_from(buffer.length())
.vortex_expect("Buffer length is too large for usize");
let length_with_padding = length + buffer.padding() as usize;
self.state = State::Buffer(ReadingBuffer {
length,
length_with_padding,
alignment: buffer.alignment().max(1).into(),
});
}
MessageHeader::DType => {
let msg_dtype = msg.header_as_dtype().vortex_expect("dtype header");
let dtype = DType::try_from_view(msg_dtype, msg_bytes.clone())?;
// Nothing else to read, so we reset the state to Length
self.state = Default::default();
return Ok(PollRead::Some(DecoderMessage::DType(dtype)));
}
_ => {
vortex_bail!("Unsupported message header type {:?}", msg.header_type());
}
}
}
State::Buffer(ReadingBuffer {
length,
length_with_padding,
alignment,
}) => {
// Ensure the buffer is read with maximum of reader and message alignment.
let read_alignment = self.alignment.max(*alignment);
if bytes.remaining() < *length_with_padding {
return Ok(PollRead::NeedMore(*length_with_padding));
}
let buffer = bytes.copy_to_aligned(*length, read_alignment);
// Then use the buffer-requested alignment for the result.
let msg = DecoderMessage::Buffer(buffer.aligned(*alignment));
bytes.advance(length_with_padding - length);
// Nothing else to read, so we reset the state to Length
self.state = Default::default();
return Ok(PollRead::Some(msg));
}
State::Array(ReadingArray {
header,
buffers_length,
}) => {
if bytes.remaining() < *buffers_length {
return Ok(PollRead::NeedMore(*buffers_length));
}
// SAFETY: we've already validated the header
let msg = unsafe { root_unchecked::<fb::Message>(header.as_ref()) };
let array_msg = msg
.header_as_array_message()
.vortex_expect("array message header");
let array = array_msg
.array()
.ok_or_else(|| vortex_err!("array data message missing array"))?;
let buffers = array_msg
.buffers()
.unwrap_or_default()
.iter()
.map(|buffer_msg| {
let buffer_len = usize::try_from(buffer_msg.length())
.vortex_expect("buffer length is too large for usize");
let buffer_alignment = Alignment::from(buffer_msg.alignment().max(1));
// Ensure the buffer is read with maximum of reader and message alignment.
let read_alignment = self.alignment.max(buffer_alignment);
let buffer = bytes.copy_to_aligned(buffer_len, read_alignment);
bytes.advance(buffer_msg.padding() as usize);
// But use the buffer-requested alignment for the result.
buffer.aligned(buffer_alignment)
})
.collect_vec();
let row_count = usize::try_from(array_msg.row_count())
.map_err(|_| vortex_err!("row count is too large for usize"))?;
let msg = DecoderMessage::Array(ArrayParts::new(
row_count,
array,
header.clone(),
buffers,
));
self.state = Default::default();
return Ok(PollRead::Some(msg));
}
}
}
}
}
#[cfg(test)]
mod test {
use bytes::BytesMut;
use vortex_array::array::ConstantArray;
use vortex_array::{Array, IntoArray};
use vortex_buffer::buffer;
use vortex_error::vortex_panic;
use super::*;
use crate::messages::{EncoderMessage, MessageEncoder};
fn write_and_read(expected: Array) {
let mut ipc_bytes = BytesMut::new();
let mut encoder = MessageEncoder::default();
for buf in encoder.encode(EncoderMessage::Array(&expected)) {
ipc_bytes.extend_from_slice(buf.as_ref());
}
let mut decoder = MessageDecoder::default();
// Since we provide all bytes up-front, we should never hit a NeedMore.
let mut buffer = BytesMut::from(ipc_bytes.as_ref());
let array_parts = match decoder.read_next(&mut buffer).unwrap() {
PollRead::Some(DecoderMessage::Array(array_parts)) => array_parts,
otherwise => vortex_panic!("Expected an array, got {:?}", otherwise),
};
// Decode the array parts with the context
let actual = array_parts
.decode(Default::default(), expected.dtype().clone())
.unwrap();
assert_eq!(expected.len(), actual.len());
assert_eq!(expected.encoding(), actual.encoding());
}
#[test]
fn array_ipc() {
write_and_read(buffer![0i32, 1, 2, 3].into_array());
}
#[test]
fn array_no_buffers() {
// Constant arrays have a single buffer
let array = ConstantArray::new(10i32, 20).into_array();
assert_eq!(array.nbuffers(), 1, "Array should have a single buffer");
write_and_read(array);
}
}