Fix Builder usage.

Long story short, builders are one per message.
This commit is contained in:
Bradlee Speice 2019-08-26 22:05:40 -04:00
parent bf71146152
commit b333fd3810
2 changed files with 181 additions and 110 deletions

View File

@ -1,19 +1,16 @@
use std::cmp::{max, min};
use std::collections::hash_map::{DefaultHasher, HashMap};
use std::hash::Hasher;
use std::io::{BufReader, Error, Read};
use std::str::from_utf8_unchecked;
use capnp::message::ReaderOptions;
use capnp::serialize::{read_message, write_message};
use capnp::Error;
use capnp::message::{Builder, ReaderOptions, ScratchSpace, ScratchSpaceHeapAllocator};
use capnp::serialize::write_message;
use capnp::serialize_packed::{read_message as read_message_packed, write_message as write_message_packed};
use nom::bytes::complete::take_until;
use nom::IResult;
use crate::iex::{IexMessage, IexParser};
use crate::{StreamVec, Summarizer};
use crate::iex::{IexMessage, IexPayload};
use crate::marketdata_capnp::{multi_message, Side};
use crate::marketdata_capnp::message;
use crate::SummaryStats;
fn __take_until<'a>(tag: &'static str, input: &'a [u8]) -> IResult<&'a [u8], &'a [u8]> {
take_until(tag)(input)
@ -25,22 +22,44 @@ fn parse_symbol(sym: &[u8; 8]) -> &str {
unsafe { from_utf8_unchecked(sym_bytes) }
}
pub fn serialize_capnp(parser: IexParser, size_hint: usize, packed: bool) -> Vec<u8> {
let write_fn = if packed { write_message_packed } else { write_message };
pub struct CapnpWriter<'a> {
// We have to be very careful with how messages are built, as running
// `init_root` and rebuilding will still accumulate garbage if using
// the standard HeapAllocator.
// https://github.com/capnproto/capnproto-rust/issues/111
words: Vec<capnp::Word>,
scratch: ScratchSpace<'a>,
}
// Because CapNProto builds messages in heap before serialization,
// we'll reserve memory up front and should avoid alloc calls later
let mut capnp_message = capnp::message::Builder::new_default();
let multimsg = capnp_message.init_root::<multi_message::Builder>();
multimsg.init_messages(256);
impl<'a> CapnpWriter<'a> {
pub fn new() -> CapnpWriter<'a> {
// Cap'n'Proto words are 8 bytes, MTU is 1500 bytes, theoretically need only 188 words.
// In practice, let's just make sure everything fits.
let mut words = capnp::Word::allocate_zeroed_vec(1024);
// Allocate our output buffer
let mut output: Vec<u8> = Vec::with_capacity(size_hint);
let mut scratch = ScratchSpace::new(unsafe {
std::mem::transmute(&mut words[..])
});
// Now to the actual work
for iex_msg in parser {
// Find the messages we actually care about in this context
let num_msgs = iex_msg.messages.iter().map(|m| {
CapnpWriter {
words,
scratch,
}
}
fn builder(&mut self) -> capnp::message::Builder<ScratchSpaceHeapAllocator<'a, 'a>> {
// Builders are only safe to use for serializing a single message. We can re-use the
// backing memory (unsafe because now both `self` and the returned builder have a
// mutable reference to `self.scratch), but Bad Things happen if we don't drop
// in between serialization.
capnp::message::Builder::new(ScratchSpaceHeapAllocator::new(unsafe {
std::mem::transmute(&mut self.scratch)
}))
}
pub fn serialize(&mut self, payload: &IexPayload, mut output: &mut Vec<u8>, packed: bool) {
// First, count the messages we actually care about.
let num_msgs = payload.messages.iter().map(|m| {
match m {
IexMessage::TradeReport(_) | IexMessage::PriceLevelUpdate(_) => 1,
_ => 0
@ -48,16 +67,24 @@ pub fn serialize_capnp(parser: IexParser, size_hint: usize, packed: bool) -> Vec
}).fold(0, |sum, i| sum + i);
if num_msgs == 0 {
continue;
return;
}
// And actually serialize the IEX payload to CapNProto format
let mut multimsg = capnp_message.init_root::<multi_message::Builder>();
multimsg.set_seq_no(iex_msg.first_seq_no);
// This is the unsafe (but faster) version
let mut builder = self.builder();
let mut multimsg = builder.init_root::<multi_message::Builder>();
// And the safe version used for testing
//let mut builder = capnp::message::Builder::new_default();
//let mut multimsg = builder.init_root::<multi_message::Builder>();
multimsg.set_seq_no(payload.first_seq_no);
let mut messages = multimsg.init_messages(num_msgs as u32);
let mut current_msg_no = 0;
for iex_msg in iex_msg.messages {
for iex_msg in payload.messages.iter() {
match iex_msg {
IexMessage::TradeReport(tr) => {
let mut message = messages.reborrow().get(current_msg_no);
@ -91,76 +118,47 @@ pub fn serialize_capnp(parser: IexParser, size_hint: usize, packed: bool) -> Vec
}
}
write_fn(&mut output, &capnp_message).unwrap();
}
let write_fn = if packed { write_message_packed } else { write_message };
output
}
struct AdvancingVec<'a> {
pos: usize,
inner: &'a Vec<u8>,
}
impl<'a> Read for AdvancingVec<'a> {
fn read(&mut self, buf: &mut [u8]) -> Result<usize, Error> {
// TODO: There's *got* to be a better way to handle this
let end = self.pos + buf.len();
let end = if end > self.inner.len() { self.inner.len() } else { end };
let read_size = end - self.pos;
buf[..read_size].copy_from_slice(&self.inner[self.pos..end]);
self.pos = end;
Ok(read_size)
write_fn(&mut output, &builder).unwrap();
}
}
pub fn read_capnp(buffer: &Vec<u8>, packed: bool) -> HashMap<u64, SummaryStats> {
let read_fn = if packed { read_message_packed } else { read_message };
let unbuffered = AdvancingVec {
pos: 0,
inner: buffer,
};
let mut buffered = BufReader::new(unbuffered);
let read_opts = ReaderOptions::new();
pub struct CapnpReader {
read_opts: ReaderOptions
}
let mut stats = HashMap::new();
impl CapnpReader {
pub fn new() -> CapnpReader {
CapnpReader {
read_opts: ReaderOptions::new()
}
}
while let Ok(msg) = read_fn(&mut buffered, read_opts) {
let multimsg = msg.get_root::<multi_message::Reader>().unwrap();
pub fn deserialize_packed<'a>(&self, buf: &'a mut StreamVec, stats: &mut Summarizer) -> Result<(), Error> {
// Because `capnp::serialize_packed::PackedRead` is hidden from us, packed reads
// *have* to both allocate new segments every read, and copy the buffer into
// those same segments. Un-packed reading can use `SliceSegments` for true zero-copy
let reader = read_message_packed(buf, self.read_opts)?;
let multimsg = reader.get_root::<multi_message::Reader>().unwrap();
for msg in multimsg.get_messages().unwrap().iter() {
// Hash the symbol name since we can't return a HashMap containing
// string pointers as the keys
let sym = msg.get_symbol().unwrap();
let mut h = DefaultHasher::new();
h.write(sym.as_bytes());
let key = h.finish();
let mut sym_stats = stats.entry(key)
.or_insert(SummaryStats::new(sym));
match msg.which() {
Ok(message::Trade(tr)) => {
let tr = tr.unwrap();
sym_stats.trade_volume += tr.get_size() as u64;
}
stats.append_trade_volume(msg.get_symbol().unwrap(), tr.get_size() as u64);
},
Ok(message::Quote(q)) => {
let q = q.unwrap();
if q.get_side().unwrap() == Side::Buy {
sym_stats.bid_high = max(sym_stats.bid_high, q.get_price());
sym_stats.bid_low = min(sym_stats.bid_low, q.get_price());
} else {
sym_stats.ask_high = max(sym_stats.ask_high, q.get_price());
sym_stats.ask_low = min(sym_stats.ask_low, q.get_price());
let is_bid = match q.get_side().unwrap() {
Side::Buy => true,
_ => false
};
stats.update_quote_prices(msg.get_symbol().unwrap(), q.get_price(), is_bid);
},
_ => panic!("Unrecognized message type!")
}
};
Ok(())
}
}
_ => {
panic!("Unrecognized message type")
}
}
}
}
stats
}

View File

@ -1,11 +1,15 @@
use std::cmp::{max, min};
use std::collections::hash_map::{DefaultHasher, HashMap};
use std::fs::File;
use std::io::Read;
use std::hash::Hasher;
use std::io::{BufRead, Read};
use std::io::Error;
use std::path::Path;
use std::time::SystemTime;
use clap::{App, Arg};
use crate::iex::{IexMessage, IexParser};
use crate::iex::IexParser;
// Cap'n'Proto and Flatbuffers typically ask that you generate code on the fly to match
// the schemas. For purposes of auto-complete and easy browsing in the repository,
@ -39,19 +43,28 @@ fn main() {
file.read_to_end(&mut buf)
.expect(&format!("Unable to read file={}", path.display()));
let start = SystemTime::now();
let _start = SystemTime::now();
let mut summarizer = Summarizer::default();
let mut parser = IexParser::new(&buf[..]);
// Try with Capnproto for now
let parser = IexParser::new(&buf[..]);
let capnp_buf = capnp_runner::serialize_capnp(parser, buf.len(), true);
let stats = capnp_runner::read_capnp(&capnp_buf, true);
let mut capnp_writer = capnp_runner::CapnpWriter::new();
let capnp_reader = capnp_runner::CapnpReader::new();
let mut output_buf = Vec::new();
dbg!(stats);
for iex_payload in parser {
//let iex_payload = parser.next().unwrap();
capnp_writer.serialize(&iex_payload, &mut output_buf, true);
}
println!(
"Parse time seconds={}",
SystemTime::now().duration_since(start).unwrap().as_secs()
)
let mut read_buf = StreamVec::new(output_buf);
let mut parsed_msgs: u64 = 0;
while let Ok(_) = capnp_reader.deserialize_packed(&mut read_buf, &mut summarizer) {
parsed_msgs += 1;
}
assert_eq!(read_buf.pos, read_buf.inner.len());
dbg!(parsed_msgs);
dbg!(summarizer);
}
#[derive(Debug)]
@ -64,15 +77,75 @@ pub struct SummaryStats {
ask_low: u64,
}
impl SummaryStats {
fn new(sym: &str) -> SummaryStats {
SummaryStats {
#[derive(Default, Debug)]
pub struct Summarizer {
data: HashMap<u64, SummaryStats>
}
impl Summarizer {
fn entry(&mut self, sym: &str) -> &mut SummaryStats {
let mut hasher = DefaultHasher::new();
hasher.write(sym.as_bytes());
self.data.entry(hasher.finish())
.or_insert(SummaryStats {
symbol: sym.to_string(),
trade_volume: 0,
bid_high: 0,
bid_low: u64::max_value(),
ask_high: 0,
ask_low: u64::max_value(),
})
}
pub fn append_trade_volume(&mut self, sym: &str, volume: u64) {
self.entry(sym).trade_volume += volume;
}
pub fn update_quote_prices(&mut self, sym: &str, price: u64, is_buy: bool) {
let entry = self.entry(sym);
if is_buy {
entry.bid_low = min(entry.bid_low, price);
entry.bid_high = max(entry.bid_high, price);
} else {
entry.ask_low = min(entry.ask_low, price);
entry.ask_high = max(entry.ask_high, price);
}
}
}
pub struct StreamVec {
pos: usize,
inner: Vec<u8>,
}
impl StreamVec {
pub fn new(buf: Vec<u8>) -> StreamVec {
StreamVec {
pos: 0,
inner: buf,
}
}
}
impl Read for StreamVec {
fn read(&mut self, buf: &mut [u8]) -> Result<usize, Error> {
// TODO: There's *got* to be a better way to handle this
let end = self.pos + buf.len();
let end = if end > self.inner.len() { self.inner.len() } else { end };
let read_size = end - self.pos;
buf[..read_size].copy_from_slice(&self.inner[self.pos..end]);
self.pos = end;
Ok(read_size)
}
}
impl BufRead for StreamVec {
fn fill_buf(&mut self) -> Result<&[u8], Error> {
Ok(&self.inner[self.pos..])
}
fn consume(&mut self, amt: usize) {
self.pos += amt;
}
}