2018-08-09 12:03:34 -07:00
|
|
|
//! The `retransmit_stage` retransmits blobs between validators
|
2018-06-13 21:52:23 -07:00
|
|
|
|
2019-02-21 11:19:45 -08:00
|
|
|
use crate::bank_forks::BankForks;
|
2019-02-07 20:52:39 -08:00
|
|
|
use crate::blocktree::Blocktree;
|
2019-02-11 16:20:31 -08:00
|
|
|
use crate::cluster_info::{
|
2019-04-19 22:31:40 -07:00
|
|
|
compute_retransmit_peers, ClusterInfo, GROW_LAYER_CAPACITY, NEIGHBORHOOD_SIZE,
|
2019-02-11 16:20:31 -08:00
|
|
|
};
|
2019-02-12 10:56:48 -08:00
|
|
|
use crate::packet::SharedBlob;
|
2018-12-07 19:16:27 -08:00
|
|
|
use crate::result::{Error, Result};
|
|
|
|
use crate::service::Service;
|
2019-02-28 13:15:25 -08:00
|
|
|
use crate::staking_utils;
|
2018-12-07 19:16:27 -08:00
|
|
|
use crate::streamer::BlobReceiver;
|
2019-02-07 15:10:54 -08:00
|
|
|
use crate::window_service::WindowService;
|
2019-02-18 22:26:22 -08:00
|
|
|
use solana_metrics::counter::Counter;
|
2018-11-16 08:45:59 -08:00
|
|
|
use solana_metrics::{influxdb, submit};
|
2018-06-13 21:52:23 -07:00
|
|
|
use std::net::UdpSocket;
|
2019-02-13 20:04:20 -08:00
|
|
|
use std::sync::atomic::AtomicBool;
|
2019-02-04 15:33:43 -08:00
|
|
|
use std::sync::mpsc::channel;
|
2018-08-09 13:41:21 -07:00
|
|
|
use std::sync::mpsc::RecvTimeoutError;
|
2018-06-13 21:52:23 -07:00
|
|
|
use std::sync::{Arc, RwLock};
|
2018-08-09 14:17:50 -07:00
|
|
|
use std::thread::{self, Builder, JoinHandle};
|
2018-08-09 13:41:21 -07:00
|
|
|
use std::time::Duration;
|
2018-06-13 21:52:23 -07:00
|
|
|
|
2019-02-11 16:20:31 -08:00
|
|
|
fn retransmit(
|
2019-02-21 11:19:45 -08:00
|
|
|
bank_forks: &Arc<RwLock<BankForks>>,
|
2019-02-11 16:20:31 -08:00
|
|
|
cluster_info: &Arc<RwLock<ClusterInfo>>,
|
|
|
|
r: &BlobReceiver,
|
|
|
|
sock: &UdpSocket,
|
|
|
|
) -> Result<()> {
|
|
|
|
let timer = Duration::new(1, 0);
|
|
|
|
let mut dq = r.recv_timeout(timer)?;
|
|
|
|
while let Ok(mut nq) = r.try_recv() {
|
|
|
|
dq.append(&mut nq);
|
|
|
|
}
|
|
|
|
|
|
|
|
submit(
|
|
|
|
influxdb::Point::new("retransmit-stage")
|
|
|
|
.add_field("count", influxdb::Value::Integer(dq.len() as i64))
|
|
|
|
.to_owned(),
|
|
|
|
);
|
2019-04-19 22:31:40 -07:00
|
|
|
let r_bank = bank_forks.read().unwrap().working_bank();
|
|
|
|
let bank_epoch = r_bank.get_stakers_epoch(r_bank.slot());
|
2019-02-12 10:56:48 -08:00
|
|
|
let (neighbors, children) = compute_retransmit_peers(
|
2019-04-19 22:31:40 -07:00
|
|
|
&staking_utils::delegated_stakes_at_epoch(&r_bank, bank_epoch).unwrap(),
|
2019-02-11 16:20:31 -08:00
|
|
|
cluster_info,
|
2019-04-19 22:31:40 -07:00
|
|
|
NEIGHBORHOOD_SIZE,
|
2019-02-11 16:20:31 -08:00
|
|
|
NEIGHBORHOOD_SIZE,
|
|
|
|
GROW_LAYER_CAPACITY,
|
|
|
|
);
|
2019-02-12 10:56:48 -08:00
|
|
|
for b in &dq {
|
|
|
|
if b.read().unwrap().should_forward() {
|
|
|
|
ClusterInfo::retransmit_to(&cluster_info, &neighbors, ©_for_neighbors(b), sock)?;
|
|
|
|
}
|
|
|
|
// Always send blobs to children
|
|
|
|
ClusterInfo::retransmit_to(&cluster_info, &children, b, sock)?;
|
2018-08-09 13:41:21 -07:00
|
|
|
}
|
|
|
|
Ok(())
|
|
|
|
}
|
|
|
|
|
2019-02-12 10:56:48 -08:00
|
|
|
/// Modifies a blob for neighbors nodes
|
|
|
|
#[inline]
|
|
|
|
fn copy_for_neighbors(b: &SharedBlob) -> SharedBlob {
|
|
|
|
let mut blob = b.read().unwrap().clone();
|
|
|
|
// Disable blob forwarding for neighbors
|
|
|
|
blob.forward(false);
|
|
|
|
Arc::new(RwLock::new(blob))
|
|
|
|
}
|
|
|
|
|
2019-01-02 00:46:15 -08:00
|
|
|
/// Service to retransmit messages from the leader or layer 1 to relevant peer nodes.
|
2018-10-08 19:55:54 -07:00
|
|
|
/// See `cluster_info` for network layer definitions.
|
2018-08-09 13:41:21 -07:00
|
|
|
/// # Arguments
|
|
|
|
/// * `sock` - Socket to read from. Read timeout is set to 1.
|
|
|
|
/// * `exit` - Boolean to signal system exit.
|
2018-10-08 19:55:54 -07:00
|
|
|
/// * `cluster_info` - This structure needs to be updated and populated by the bank and via gossip.
|
2018-08-09 13:41:21 -07:00
|
|
|
/// * `recycler` - Blob recycler.
|
|
|
|
/// * `r` - Receive channel for blobs to be retransmitted to all the layer 1 nodes.
|
2018-10-08 19:55:54 -07:00
|
|
|
fn retransmitter(
|
|
|
|
sock: Arc<UdpSocket>,
|
2019-02-21 11:19:45 -08:00
|
|
|
bank_forks: Arc<RwLock<BankForks>>,
|
2018-10-08 19:55:54 -07:00
|
|
|
cluster_info: Arc<RwLock<ClusterInfo>>,
|
|
|
|
r: BlobReceiver,
|
|
|
|
) -> JoinHandle<()> {
|
2018-08-09 13:41:21 -07:00
|
|
|
Builder::new()
|
|
|
|
.name("solana-retransmitter".to_string())
|
|
|
|
.spawn(move || {
|
|
|
|
trace!("retransmitter started");
|
|
|
|
loop {
|
2019-02-21 11:19:45 -08:00
|
|
|
if let Err(e) = retransmit(&bank_forks, &cluster_info, &r, &sock) {
|
2018-08-09 13:41:21 -07:00
|
|
|
match e {
|
|
|
|
Error::RecvTimeoutError(RecvTimeoutError::Disconnected) => break,
|
|
|
|
Error::RecvTimeoutError(RecvTimeoutError::Timeout) => (),
|
|
|
|
_ => {
|
|
|
|
inc_new_counter_info!("streamer-retransmit-error", 1, 1);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
trace!("exiting retransmitter");
|
2018-12-07 19:01:28 -08:00
|
|
|
})
|
|
|
|
.unwrap()
|
2018-08-09 13:41:21 -07:00
|
|
|
}
|
|
|
|
|
2018-08-09 12:03:34 -07:00
|
|
|
pub struct RetransmitStage {
|
2018-10-10 16:49:41 -07:00
|
|
|
thread_hdls: Vec<JoinHandle<()>>,
|
2019-02-07 15:10:54 -08:00
|
|
|
window_service: WindowService,
|
2018-06-13 21:52:23 -07:00
|
|
|
}
|
|
|
|
|
2018-08-09 12:03:34 -07:00
|
|
|
impl RetransmitStage {
|
2019-02-07 15:10:54 -08:00
|
|
|
#[allow(clippy::new_ret_no_self)]
|
2018-06-13 21:52:23 -07:00
|
|
|
pub fn new(
|
2019-02-21 11:19:45 -08:00
|
|
|
bank_forks: &Arc<RwLock<BankForks>>,
|
2019-02-07 20:52:39 -08:00
|
|
|
blocktree: Arc<Blocktree>,
|
2018-10-08 19:55:54 -07:00
|
|
|
cluster_info: &Arc<RwLock<ClusterInfo>>,
|
2018-08-28 16:32:40 -07:00
|
|
|
retransmit_socket: Arc<UdpSocket>,
|
2018-08-30 12:07:54 -07:00
|
|
|
repair_socket: Arc<UdpSocket>,
|
2018-06-27 11:33:56 -07:00
|
|
|
fetch_stage_receiver: BlobReceiver,
|
2019-03-04 20:50:02 -08:00
|
|
|
exit: &Arc<AtomicBool>,
|
2019-02-04 15:33:43 -08:00
|
|
|
) -> Self {
|
2018-06-13 21:52:23 -07:00
|
|
|
let (retransmit_sender, retransmit_receiver) = channel();
|
|
|
|
|
2019-01-02 00:46:15 -08:00
|
|
|
let t_retransmit = retransmitter(
|
|
|
|
retransmit_socket,
|
2019-02-21 11:19:45 -08:00
|
|
|
bank_forks.clone(),
|
2019-01-02 00:46:15 -08:00
|
|
|
cluster_info.clone(),
|
|
|
|
retransmit_receiver,
|
|
|
|
);
|
2019-02-07 15:10:54 -08:00
|
|
|
let window_service = WindowService::new(
|
2019-02-07 20:52:39 -08:00
|
|
|
blocktree,
|
2018-10-08 19:55:54 -07:00
|
|
|
cluster_info.clone(),
|
2018-06-13 21:52:23 -07:00
|
|
|
fetch_stage_receiver,
|
|
|
|
retransmit_sender,
|
2018-08-30 12:07:54 -07:00
|
|
|
repair_socket,
|
2019-01-31 13:43:22 -08:00
|
|
|
exit,
|
2019-04-08 12:46:23 -07:00
|
|
|
None,
|
2018-06-13 21:52:23 -07:00
|
|
|
);
|
|
|
|
|
2019-02-07 15:10:54 -08:00
|
|
|
let thread_hdls = vec![t_retransmit];
|
|
|
|
Self {
|
|
|
|
thread_hdls,
|
|
|
|
window_service,
|
|
|
|
}
|
2018-06-13 21:52:23 -07:00
|
|
|
}
|
|
|
|
}
|
2018-07-03 21:14:08 -07:00
|
|
|
|
2018-08-09 12:03:34 -07:00
|
|
|
impl Service for RetransmitStage {
|
2018-10-10 16:49:41 -07:00
|
|
|
type JoinReturnType = ();
|
2018-07-03 21:14:08 -07:00
|
|
|
|
2018-10-10 16:49:41 -07:00
|
|
|
fn join(self) -> thread::Result<()> {
|
|
|
|
for thread_hdl in self.thread_hdls {
|
|
|
|
thread_hdl.join()?;
|
2018-07-03 21:14:08 -07:00
|
|
|
}
|
2019-02-07 15:10:54 -08:00
|
|
|
self.window_service.join()?;
|
2018-10-10 16:49:41 -07:00
|
|
|
Ok(())
|
2018-07-03 21:14:08 -07:00
|
|
|
}
|
|
|
|
}
|
2019-02-11 16:20:31 -08:00
|
|
|
|
|
|
|
#[cfg(test)]
|
|
|
|
mod tests {
|
|
|
|
use super::*;
|
|
|
|
|
2019-02-12 10:56:48 -08:00
|
|
|
// Test that blobs always come out with forward unset for neighbors
|
|
|
|
#[test]
|
|
|
|
fn test_blob_for_neighbors() {
|
|
|
|
let blob = SharedBlob::default();
|
|
|
|
blob.write().unwrap().forward(true);
|
|
|
|
let for_hoodies = copy_for_neighbors(&blob);
|
|
|
|
assert!(!for_hoodies.read().unwrap().should_forward());
|
|
|
|
}
|
2019-02-11 16:20:31 -08:00
|
|
|
}
|