gaspersic/solana
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Weight repair slots based on vote stake (#10741)

Browse Source 
* Weight repair slots based on vote stake

* Add test
This commit is contained in:
sakridge
2020-06-22 20:27:45 -07:00
committed by GitHub
parent 2ba8fc5243
commit cabd0a09c3
5 changed files with 147 additions and 27 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
core/src/cluster_info_vote_listener.rs
View File

@@ -49,7 +49,7 @@ pub type VerifiedVoteTransactionsReceiver = CrossbeamReceiver<Vec<Transaction>>;
pub struct SlotVoteTracker {
voted: HashSet<Arc<Pubkey>>,
updates: Option<Vec<Arc<Pubkey>>>,
total_stake: u64,
pub total_stake: u64,
}
impl SlotVoteTracker {
@@ -62,7 +62,7 @@ impl SlotVoteTracker {
#[derive(Default)]
pub struct VoteTracker {
// Map from a slot to a set of validators who have voted for that slot
slot_vote_trackers: RwLock<HashMap<Slot, Arc<RwLock<SlotVoteTracker>>>>,
pub slot_vote_trackers: RwLock<HashMap<Slot, Arc<RwLock<SlotVoteTracker>>>>,
// Don't track votes from people who are not staked, acts as a spam filter
epoch_authorized_voters: RwLock<HashMap<Epoch, Arc<EpochAuthorizedVoters>>>,
leader_schedule_epoch: RwLock<Epoch>,

163
core/src/repair_service.rs
View File

@@ -2,12 +2,16 @@
//! regularly finds missing shreds in the ledger and sends repair requests for those shreds
use crate::{
cluster_info::ClusterInfo,
cluster_info_vote_listener::VoteTracker,
cluster_slots::ClusterSlots,
commitment::VOTE_THRESHOLD_SIZE,
result::Result,
serve_repair::{RepairType, ServeRepair, DEFAULT_NONCE},
};
use crossbeam_channel::{Receiver as CrossbeamReceiver, Sender as CrossbeamSender};
use rand::distributions::{Distribution, WeightedIndex};
use rand::{thread_rng, Rng, SeedableRng};
use rand_chacha::ChaChaRng;
use solana_ledger::{
blockstore::{Blockstore, CompletedSlotsReceiver, SlotMeta},
shred::Nonce,
@@ -113,6 +117,7 @@ impl RepairService {
cluster_info: Arc<ClusterInfo>,
repair_info: RepairInfo,
cluster_slots: Arc<ClusterSlots>,
vote_tracker: Arc<VoteTracker>,
) -> Self {
let t_repair = Builder::new()
.name("solana-repair-service".to_string())
@@ -124,6 +129,7 @@ impl RepairService {
cluster_info,
repair_info,
&cluster_slots,
vote_tracker,
)
})
.unwrap();
@@ -138,6 +144,7 @@ impl RepairService {
cluster_info: Arc<ClusterInfo>,
repair_info: RepairInfo,
cluster_slots: &ClusterSlots,
vote_tracker: Arc<VoteTracker>,
) {
let serve_repair = ServeRepair::new(cluster_info.clone());
let id = cluster_info.id();
@@ -190,6 +197,7 @@ impl RepairService {
root_bank.slot(),
MAX_REPAIR_LENGTH,
&duplicate_slot_repair_statuses,
&vote_tracker,
)
};
@@ -271,6 +279,7 @@ impl RepairService {
root: Slot,
max_repairs: usize,
duplicate_slot_repair_statuses: &HashMap<Slot, DuplicateSlotRepairStatus>,
vote_tracker: &Arc<VoteTracker>,
) -> Result<Vec<RepairType>> {
// Slot height and shred indexes for shreds we want to repair
let mut repairs: Vec<RepairType> = vec![];
@@ -280,10 +289,9 @@ impl RepairService {
max_repairs,
root,
duplicate_slot_repair_statuses,
vote_tracker,
);
// TODO: Incorporate gossip to determine priorities for repair?
// Try to resolve orphans in blockstore
let orphans = blockstore.orphans_iterator(root + 1).unwrap();
Self::generate_repairs_for_orphans(orphans, &mut repairs);
@@ -526,27 +534,64 @@ impl RepairService {
max_repairs: usize,
slot: Slot,
duplicate_slot_repair_statuses: &HashMap<Slot, DuplicateSlotRepairStatus>,
vote_tracker: &Arc<VoteTracker>,
) {
let mut seed = [0u8; 32];
thread_rng().fill(&mut seed);
let rng = &mut ChaChaRng::from_seed(seed);
let mut pending_slots = vec![slot];
while repairs.len() < max_repairs && !pending_slots.is_empty() {
let slot = pending_slots.pop().unwrap();
if duplicate_slot_repair_statuses.contains_key(&slot) {
// These are repaired through a different path
continue;
pending_slots.retain(|slot| !duplicate_slot_repair_statuses.contains_key(slot));
let mut next_pending_slots = vec![];
let mut level_repairs = HashMap::new();
for slot in &pending_slots {
if let Some(slot_meta) = blockstore.meta(*slot).unwrap() {
let new_repairs = Self::generate_repairs_for_slot(
blockstore,
*slot,
&slot_meta,
std::usize::MAX,
);
if !new_repairs.is_empty() {
level_repairs.insert(*slot, new_repairs);
}
next_pending_slots.extend(slot_meta.next_slots);
}
}
if let Some(slot_meta) = blockstore.meta(slot).unwrap() {
let new_repairs = Self::generate_repairs_for_slot(
blockstore,
slot,
&slot_meta,
max_repairs - repairs.len(),
);
repairs.extend(new_repairs);
let next_slots = slot_meta.next_slots;
pending_slots.extend(next_slots);
} else {
break;
if !level_repairs.is_empty() {
let mut slots_to_repair: Vec<_> = level_repairs.keys().cloned().collect();
let mut weights: Vec<_> = {
let r_vote_tracker = vote_tracker.slot_vote_trackers.read().unwrap();
slots_to_repair
.iter()
.map(|slot| {
if let Some(slot_vote_tracker) = r_vote_tracker.get(slot) {
std::cmp::max(slot_vote_tracker.read().unwrap().total_stake, 1)
} else {
// should it be something else?
1
}
})
.collect()
};
let mut weighted_index = WeightedIndex::new(weights.clone()).unwrap();
while repairs.len() < max_repairs && !level_repairs.is_empty() {
let index = weighted_index.sample(rng);
let slot_repairs = level_repairs.get_mut(&slots_to_repair[index]).unwrap();
repairs.push(slot_repairs.remove(0));
if slot_repairs.is_empty() {
level_repairs.remove(&slots_to_repair[index]);
slots_to_repair.remove(index);
weights.remove(index);
if !weights.is_empty() {
weighted_index = WeightedIndex::new(weights.clone()).unwrap();
}
}
}
}
pending_slots = next_pending_slots;
}
}
@@ -633,8 +678,10 @@ mod test {
let (shreds2, _) = make_slot_entries(5, 2, 1);
shreds.extend(shreds2);
blockstore.insert_shreds(shreds, None, false).unwrap();
let vote_tracker = Arc::new(VoteTracker::default());
assert_eq!(
RepairService::generate_repairs(&blockstore, 0, 2, &HashMap::new()).unwrap(),
RepairService::generate_repairs(&blockstore, 0, 2, &HashMap::new(), &vote_tracker)
.unwrap(),
vec![RepairType::HighestShred(0, 0), RepairType::Orphan(2)]
);
}
@@ -654,9 +701,11 @@ mod test {
// any shreds for
blockstore.insert_shreds(shreds, None, false).unwrap();
let vote_tracker = Arc::new(VoteTracker::default());
// Check that repair tries to patch the empty slot
assert_eq!(
RepairService::generate_repairs(&blockstore, 0, 2, &HashMap::new()).unwrap(),
RepairService::generate_repairs(&blockstore, 0, 2, &HashMap::new(), &vote_tracker)
.unwrap(),
vec![RepairType::HighestShred(0, 0)]
);
}
@@ -701,9 +750,16 @@ mod test {
})
.collect();
let vote_tracker = Arc::new(VoteTracker::default());
assert_eq!(
RepairService::generate_repairs(&blockstore, 0, std::usize::MAX, &HashMap::new())
.unwrap(),
RepairService::generate_repairs(
&blockstore,
0,
std::usize::MAX,
&HashMap::new(),
&vote_tracker
)
.unwrap(),
expected
);
@@ -712,7 +768,8 @@ mod test {
&blockstore,
0,
expected.len() - 2,
&HashMap::new()
&HashMap::new(),
&vote_tracker,
)
.unwrap()[..],
expected[0..expected.len() - 2]
@@ -721,6 +778,55 @@ mod test {
Blockstore::destroy(&blockstore_path).expect("Expected successful database destruction");
}
#[test]
pub fn test_repairs_distributed_across_slots() {
solana_logger::setup();
let blockstore_path = get_tmp_ledger_path!();
{
let blockstore = Blockstore::open(&blockstore_path).unwrap();
let num_entries_per_slot = 100;
// Create some shreds
for i in 1..10 {
let (shreds, _) = make_slot_entries(i, 0, num_entries_per_slot as u64);
// Only insert the first shred
blockstore
.insert_shreds(shreds[..1].to_vec(), None, false)
.unwrap();
}
let vote_tracker = Arc::new(VoteTracker::default());
let repairs = RepairService::generate_repairs(
&blockstore,
0,
num_entries_per_slot,
&HashMap::new(),
&vote_tracker,
)
.unwrap();
let mut repairs_slots = HashMap::new();
for repair in repairs {
match repair {
RepairType::Shred(slot, _shred_index) => {
*repairs_slots.entry(slot).or_insert(0) += 1;
}
RepairType::HighestShred(slot, _shred_index) => {
*repairs_slots.entry(slot).or_insert(0) += 1;
}
RepairType::Orphan(slot) => {
*repairs_slots.entry(slot).or_insert(0) += 1;
}
}
}
for i in 1..10 {
assert!(repairs_slots.contains_key(&i));
}
}
Blockstore::destroy(&blockstore_path).expect("Expected successful database destruction");
}
#[test]
pub fn test_generate_highest_repair() {
let blockstore_path = get_tmp_ledger_path!();
@@ -742,9 +848,16 @@ mod test {
let expected: Vec<RepairType> =
vec![RepairType::HighestShred(0, num_shreds_per_slot - 1)];
let vote_tracker = Arc::new(VoteTracker::default());
assert_eq!(
RepairService::generate_repairs(&blockstore, 0, std::usize::MAX, &HashMap::new())
.unwrap(),
RepairService::generate_repairs(
&blockstore,
0,
std::usize::MAX,
&HashMap::new(),
&vote_tracker
)
.unwrap(),
expected
);
}

3
core/src/retransmit_stage.rs
View File

@@ -1,5 +1,6 @@
//! The `retransmit_stage` retransmits shreds between validators
use crate::cluster_info_vote_listener::VoteTracker;
use crate::{
cluster_info::{compute_retransmit_peers, ClusterInfo, DATA_PLANE_FANOUT},
cluster_slots::ClusterSlots,
@@ -413,6 +414,7 @@ impl RetransmitStage {
shred_version: u16,
cluster_slots: Arc<ClusterSlots>,
duplicate_slots_reset_sender: DuplicateSlotsResetSender,
vote_tracker: Arc<VoteTracker>,
) -> Self {
let (retransmit_sender, retransmit_receiver) = channel();
@@ -457,6 +459,7 @@ impl RetransmitStage {
rv && is_connected
},
cluster_slots,
vote_tracker,
);
let thread_hdls = t_retransmit;

1
core/src/tvu.rs
View File

@@ -145,6 +145,7 @@ impl Tvu {
tvu_config.shred_version,
cluster_slots.clone(),
duplicate_slots_reset_sender,
vote_tracker.clone(),
);
let (ledger_cleanup_slot_sender, ledger_cleanup_slot_receiver) = channel();

3
core/src/window_service.rs
View File

@@ -3,6 +3,7 @@
//!
use crate::{
cluster_info::ClusterInfo,
cluster_info_vote_listener::VoteTracker,
cluster_slots::ClusterSlots,
repair_response,
repair_service::{RepairInfo, RepairService},
@@ -300,6 +301,7 @@ impl WindowService {
leader_schedule_cache: &Arc<LeaderScheduleCache>,
shred_filter: F,
cluster_slots: Arc<ClusterSlots>,
vote_tracker: Arc<VoteTracker>,
) -> WindowService
where
F: 'static
@@ -316,6 +318,7 @@ impl WindowService {
cluster_info.clone(),
repair_info,
cluster_slots,
vote_tracker,
);
let (insert_sender, insert_receiver) = unbounded();