Route BankForks into the ReplayStage

This commit is contained in:
Michael Vines
2019-02-21 11:37:48 -08:00
committed by Grimes
parent f0f55af35b
commit b501090443
6 changed files with 209 additions and 201 deletions

View File

@@ -13,13 +13,11 @@ use crate::rpc_subscriptions::RpcSubscriptions;
use crate::service::Service;
use crate::storage_stage::StorageState;
use crate::tpu::Tpu;
use crate::tvu::{Sockets, Tvu, TvuRotationReceiver};
use crate::tvu::{Sockets, Tvu, TvuRotationInfo, TvuRotationReceiver};
use crate::voting_keypair::VotingKeypair;
use log::Level;
use solana_metrics::counter::Counter;
use solana_runtime::bank::Bank;
use solana_sdk::genesis_block::GenesisBlock;
use solana_sdk::hash::Hash;
use solana_sdk::pubkey::Pubkey;
use solana_sdk::signature::{Keypair, KeypairUtil};
use solana_sdk::timing::timestamp;
@@ -100,7 +98,6 @@ pub struct Fullnode {
rpc_service: Option<JsonRpcService>,
rpc_pubsub_service: Option<PubSubService>,
gossip_service: GossipService,
bank_forks: Arc<RwLock<BankForks>>,
sigverify_disabled: bool,
tpu_sockets: Vec<UdpSocket>,
broadcast_socket: UdpSocket,
@@ -224,46 +221,10 @@ impl Fullnode {
// Setup channel for rotation indications
let (rotation_sender, rotation_receiver) = channel();
// TODO: All this into Tpu/ReplayStage....
if bank_forks_info.len() != 1 {
warn!("TODO: figure out what to do with multiple bank forks");
}
let (bank, entry_height, last_entry_id) = {
let mut bank_forks = bank_forks.write().unwrap();
bank_forks.set_working_bank_id(bank_forks_info[0].bank_id);
(
bank_forks.working_bank(),
bank_forks_info[0].entry_height,
bank_forks_info[0].last_entry_id,
)
};
// Figure which node should generate the next tick
let (next_leader, next_slot) = {
let next_tick = bank.tick_height() + 1;
let leader_scheduler = leader_scheduler.read().unwrap();
let next_slot = leader_scheduler.tick_height_to_slot(next_tick);
let next_leader = leader_scheduler
.get_leader_for_slot(next_slot)
.expect("Leader not known after processing bank");
trace!(
"node {:?} scheduled as leader for slot {}",
next_leader,
next_slot,
);
(next_leader, next_slot)
};
// END TODO
let tvu = Tvu::new(
voting_keypair_option,
&bank_forks,
entry_height,
last_entry_id,
&bank_forks_info,
&cluster_info,
sockets,
blocktree.clone(),
@@ -277,9 +238,9 @@ impl Fullnode {
);
let tpu = Tpu::new(id, &cluster_info);
let mut fullnode = Self {
inc_new_counter_info!("fullnode-new", 1);
Self {
id,
bank_forks,
sigverify_disabled: config.sigverify_disabled,
gossip_service,
rpc_service: Some(rpc_service),
@@ -291,22 +252,19 @@ impl Fullnode {
rotation_receiver,
blocktree,
leader_scheduler,
};
// TODO: This first rotate should come from the Tvu/ReplayStage
fullnode.rotate(&bank, next_leader, next_slot, &last_entry_id);
inc_new_counter_info!("fullnode-new", 1);
fullnode
}
}
fn rotate(
&mut self,
bank: &Arc<Bank>,
leader: Pubkey,
slot: u64,
last_entry_id: &Hash,
) -> FullnodeReturnType {
if leader == self.id {
fn rotate(&mut self, rotation_info: TvuRotationInfo) -> FullnodeReturnType {
trace!(
"{:?}: rotate for slot={} to leader={:?} using last_entry_id={:?}",
self.id,
rotation_info.slot,
rotation_info.leader_id,
rotation_info.last_entry_id,
);
if rotation_info.leader_id == self.id {
let transition = match self.node_services.tpu.is_leader() {
Some(was_leader) => {
if was_leader {
@@ -319,9 +277,8 @@ impl Fullnode {
}
None => FullnodeReturnType::LeaderToLeaderRotation, // value doesn't matter here...
};
let tpu_bank = Arc::new(Bank::new_from_parent(bank, &leader));
self.node_services.tpu.switch_to_leader(
&tpu_bank,
Arc::new(rotation_info.bank),
PohServiceConfig::default(),
self.tpu_sockets
.iter()
@@ -331,8 +288,8 @@ impl Fullnode {
.try_clone()
.expect("Failed to clone broadcast socket"),
self.sigverify_disabled,
slot,
last_entry_id,
rotation_info.slot,
rotation_info.last_entry_id,
&self.blocktree,
&self.leader_scheduler,
);
@@ -340,7 +297,7 @@ impl Fullnode {
} else {
debug!("{:?} rotating to validator role", self.id);
self.node_services.tpu.switch_to_forwarder(
leader,
rotation_info.leader_id,
self.tpu_sockets
.iter()
.map(|s| s.try_clone().expect("Failed to clone TPU sockets"))
@@ -368,20 +325,9 @@ impl Fullnode {
}
match self.rotation_receiver.recv_timeout(timeout) {
Ok((bank_id, slot, leader)) => {
trace!(
"{:?}: rotate for slot={} to leader={:?} using bank={}",
self.id,
slot,
leader,
bank_id
);
// TODO: Uncomment next line once `bank_id` has a valid id in it
//self.bank_forks.write().set_working_bank_id(bank_id);
let bank = self.bank_forks.read().unwrap().working_bank();
let transition = self.rotate(&bank, leader, slot, &bank.last_id());
Ok(rotation_info) => {
let slot = rotation_info.slot;
let transition = self.rotate(rotation_info);
debug!("role transition complete: {:?}", transition);
if let Some(ref rotation_notifier) = rotation_notifier {
rotation_notifier.send((transition, slot)).unwrap();
@@ -417,7 +363,7 @@ impl Fullnode {
}
#[allow(clippy::trivially_copy_pass_by_ref)]
fn new_banks_from_blocktree(
pub fn new_banks_from_blocktree(
blocktree_path: &str,
blocktree_config: &BlocktreeConfig,
leader_scheduler: &Arc<RwLock<LeaderScheduler>>,
@@ -440,34 +386,6 @@ fn new_banks_from_blocktree(
)
}
// TODO: Remove this function from tests
#[allow(clippy::trivially_copy_pass_by_ref)]
pub fn new_bank_from_ledger(
ledger_path: &str,
ledger_config: &BlocktreeConfig,
leader_scheduler: &Arc<RwLock<LeaderScheduler>>,
) -> (Arc<Bank>, u64, Hash, Blocktree, Receiver<bool>) {
let (mut bank_forks, bank_forks_info, blocktree, ledger_signal_receiver) =
new_banks_from_blocktree(ledger_path, ledger_config, leader_scheduler);
// This helper won't handle multiple banks
assert_eq!(bank_forks_info.len(), 1);
bank_forks.set_working_bank_id(bank_forks_info[0].bank_id);
let (working_bank, entry_height, last_entry_id) = (
bank_forks.working_bank(),
bank_forks_info[0].entry_height,
bank_forks_info[0].last_entry_id,
);
(
working_bank,
entry_height,
last_entry_id,
blocktree,
ledger_signal_receiver,
)
}
impl Service for Fullnode {
type JoinReturnType = ();
@@ -494,6 +412,7 @@ mod tests {
use crate::entry::make_consecutive_blobs;
use crate::leader_scheduler::make_active_set_entries;
use crate::streamer::responder;
use solana_sdk::hash::Hash;
use std::fs::remove_dir_all;
#[test]
@@ -634,6 +553,10 @@ mod tests {
// Wait for the bootstrap leader to transition. Since there are no other nodes in the
// cluster it will continue to be the leader
assert_eq!(
rotation_receiver.recv().unwrap(),
(FullnodeReturnType::LeaderToLeaderRotation, 0)
);
assert_eq!(
rotation_receiver.recv().unwrap(),
(FullnodeReturnType::LeaderToLeaderRotation, 1)
@@ -689,8 +612,12 @@ mod tests {
Some(&bootstrap_leader_info),
&fullnode_config,
);
assert!(!bootstrap_leader.node_services.tpu.is_leader().unwrap());
let (rotation_sender, rotation_receiver) = channel();
let bootstrap_leader_exit = bootstrap_leader.run(Some(rotation_sender));
assert_eq!(
rotation_receiver.recv().unwrap(),
(FullnodeReturnType::LeaderToValidatorRotation, 2)
);
// Test that a node knows to transition to a leader based on parsing the ledger
let validator = Fullnode::new(
@@ -702,11 +629,15 @@ mod tests {
&fullnode_config,
);
assert!(validator.node_services.tpu.is_leader().unwrap());
validator.close().expect("Expected leader node to close");
bootstrap_leader
.close()
.expect("Expected validator node to close");
let (rotation_sender, rotation_receiver) = channel();
let validator_exit = validator.run(Some(rotation_sender));
assert_eq!(
rotation_receiver.recv().unwrap(),
(FullnodeReturnType::LeaderToLeaderRotation, 2)
);
validator_exit();
bootstrap_leader_exit();
}
for path in ledger_paths {
Blocktree::destroy(&path).expect("Expected successful database destruction");
@@ -792,11 +723,13 @@ mod tests {
// Close the validator so that rocksdb has locks available
validator_exit();
let leader_scheduler = Arc::new(RwLock::new(LeaderScheduler::default()));
let (bank, entry_height, _, _, _) = new_bank_from_ledger(
let (bank_forks, bank_forks_info, _, _) = new_banks_from_blocktree(
&validator_ledger_path,
&BlocktreeConfig::default(),
&leader_scheduler,
);
let bank = bank_forks.working_bank();
let entry_height = bank_forks_info[0].entry_height;
assert!(bank.tick_height() >= leader_scheduler.read().unwrap().ticks_per_epoch);