garage/src/table_sync.rs

737 lines
18 KiB
Rust
Raw Normal View History

2020-04-16 12:50:49 +00:00
use rand::Rng;
use std::collections::{BTreeMap, VecDeque};
2020-04-16 12:50:49 +00:00
use std::sync::Arc;
2020-04-16 16:41:10 +00:00
use std::time::{Duration, Instant};
2020-04-16 12:50:49 +00:00
2020-04-16 16:41:10 +00:00
use futures::future::BoxFuture;
use futures::{pin_mut, select};
2020-04-16 12:50:49 +00:00
use futures_util::future::*;
use futures_util::stream::*;
use serde::{Deserialize, Serialize};
use serde_bytes::ByteBuf;
2020-04-16 12:50:49 +00:00
use tokio::sync::Mutex;
use tokio::sync::{mpsc, watch};
2020-04-16 12:50:49 +00:00
use crate::data::*;
use crate::error::Error;
2020-04-16 15:04:28 +00:00
use crate::membership::Ring;
2020-04-16 12:50:49 +00:00
use crate::table::*;
2020-04-16 17:28:02 +00:00
const MAX_DEPTH: usize = 16;
const SCAN_INTERVAL: Duration = Duration::from_secs(60);
2020-04-16 16:41:10 +00:00
const CHECKSUM_CACHE_TIMEOUT: Duration = Duration::from_secs(1800);
const TABLE_SYNC_RPC_TIMEOUT: Duration = Duration::from_secs(10);
pub struct TableSyncer<F: TableSchema, R: TableReplication> {
table: Arc<Table<F, R>>,
todo: Mutex<SyncTodo>,
cache: Vec<Mutex<BTreeMap<SyncRange, RangeChecksum>>>,
2020-04-16 12:50:49 +00:00
}
#[derive(Serialize, Deserialize)]
pub enum SyncRPC {
2020-04-17 19:59:07 +00:00
GetRootChecksumRange(Hash, Hash),
RootChecksumRange(SyncRange),
Checksums(Vec<RangeChecksum>, bool),
2020-04-17 16:27:29 +00:00
Difference(Vec<SyncRange>, Vec<Arc<ByteBuf>>),
}
2020-04-16 12:50:49 +00:00
pub struct SyncTodo {
todo: Vec<TodoPartition>,
2020-04-16 12:50:49 +00:00
}
#[derive(Debug, Clone)]
struct TodoPartition {
begin: Hash,
end: Hash,
retain: bool,
2020-04-16 12:50:49 +00:00
}
2020-04-16 16:41:10 +00:00
#[derive(Hash, PartialEq, Eq, Debug, Clone, Serialize, Deserialize)]
pub struct SyncRange {
begin: Vec<u8>,
end: Vec<u8>,
level: usize,
2020-04-16 16:41:10 +00:00
}
2020-04-16 17:28:02 +00:00
impl std::cmp::PartialOrd for SyncRange {
fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
Some(self.cmp(other))
}
}
impl std::cmp::Ord for SyncRange {
fn cmp(&self, other: &Self) -> std::cmp::Ordering {
self.begin.cmp(&other.begin)
}
}
2020-04-16 16:41:10 +00:00
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct RangeChecksum {
bounds: SyncRange,
children: Vec<(SyncRange, Hash)>,
found_limit: Option<Vec<u8>>,
2020-04-16 16:41:10 +00:00
#[serde(skip, default = "std::time::Instant::now")]
time: Instant,
2020-04-16 16:41:10 +00:00
}
impl<F, R> TableSyncer<F, R>
where
F: TableSchema + 'static,
R: TableReplication + 'static,
{
pub async fn launch(table: Arc<Table<F, R>>) -> Arc<Self> {
2020-04-16 12:50:49 +00:00
let todo = SyncTodo { todo: Vec::new() };
let syncer = Arc::new(TableSyncer {
table: table.clone(),
todo: Mutex::new(todo),
cache: (0..MAX_DEPTH)
.map(|_| Mutex::new(BTreeMap::new()))
.collect::<Vec<_>>(),
2020-04-16 12:50:49 +00:00
});
let (busy_tx, busy_rx) = mpsc::unbounded_channel();
2020-04-16 12:50:49 +00:00
let s1 = syncer.clone();
table
.system
.background
2020-04-19 21:33:38 +00:00
.spawn_worker(
format!("table sync watcher for {}", table.name),
move |must_exit: watch::Receiver<bool>| s1.watcher_task(must_exit, busy_rx),
)
2020-04-16 12:50:49 +00:00
.await;
let s2 = syncer.clone();
table
.system
.background
2020-04-19 21:33:38 +00:00
.spawn_worker(
format!("table syncer for {}", table.name),
move |must_exit: watch::Receiver<bool>| s2.syncer_task(must_exit, busy_tx),
)
2020-04-16 12:50:49 +00:00
.await;
syncer
}
async fn watcher_task(
self: Arc<Self>,
mut must_exit: watch::Receiver<bool>,
mut busy_rx: mpsc::UnboundedReceiver<bool>,
2020-04-16 12:50:49 +00:00
) -> Result<(), Error> {
let mut prev_ring: Arc<Ring> = self.table.system.ring.borrow().clone();
let mut ring_recv: watch::Receiver<Arc<Ring>> = self.table.system.ring.clone();
let mut nothing_to_do_since = Some(Instant::now());
2020-04-16 12:50:49 +00:00
2020-04-16 17:28:02 +00:00
while !*must_exit.borrow() {
2020-04-16 12:50:49 +00:00
let s_ring_recv = ring_recv.recv().fuse();
let s_busy = busy_rx.recv().fuse();
2020-04-16 12:50:49 +00:00
let s_must_exit = must_exit.recv().fuse();
let s_timeout = tokio::time::delay_for(Duration::from_secs(1)).fuse();
pin_mut!(s_ring_recv, s_busy, s_must_exit, s_timeout);
2020-04-16 12:50:49 +00:00
select! {
new_ring_r = s_ring_recv => {
if let Some(new_ring) = new_ring_r {
2020-04-21 12:54:55 +00:00
debug!("({}) Adding ring difference to syncer todo list", self.table.name);
2020-04-16 12:50:49 +00:00
self.todo.lock().await.add_ring_difference(&self.table, &prev_ring, &new_ring);
prev_ring = new_ring;
}
}
busy_opt = s_busy => {
if let Some(busy) = busy_opt {
if busy {
nothing_to_do_since = None;
} else {
if nothing_to_do_since.is_none() {
nothing_to_do_since = Some(Instant::now());
}
}
}
}
2020-04-16 12:50:49 +00:00
must_exit_v = s_must_exit => {
if must_exit_v.unwrap_or(false) {
2020-04-16 17:28:02 +00:00
break;
2020-04-16 12:50:49 +00:00
}
}
_ = s_timeout => {
if nothing_to_do_since.map(|t| Instant::now() - t >= SCAN_INTERVAL).unwrap_or(false) {
nothing_to_do_since = None;
2020-04-21 12:54:55 +00:00
debug!("({}) Adding full scan to syncer todo list", self.table.name);
self.todo.lock().await.add_full_scan(&self.table);
}
}
2020-04-16 12:50:49 +00:00
}
}
2020-04-16 17:28:02 +00:00
Ok(())
2020-04-16 12:50:49 +00:00
}
async fn syncer_task(
self: Arc<Self>,
mut must_exit: watch::Receiver<bool>,
busy_tx: mpsc::UnboundedSender<bool>,
2020-04-16 12:50:49 +00:00
) -> Result<(), Error> {
2020-04-16 16:41:10 +00:00
while !*must_exit.borrow() {
if let Some(partition) = self.todo.lock().await.pop_task() {
busy_tx.send(true)?;
let res = self
.clone()
.sync_partition(&partition, &mut must_exit)
.await;
2020-04-16 16:41:10 +00:00
if let Err(e) = res {
2020-04-21 12:54:55 +00:00
warn!(
"({}) Error while syncing {:?}: {}",
self.table.name, partition, e
);
2020-04-16 16:41:10 +00:00
}
} else {
busy_tx.send(false)?;
2020-04-16 16:41:10 +00:00
tokio::time::delay_for(Duration::from_secs(1)).await;
}
}
Ok(())
}
2020-04-16 12:50:49 +00:00
async fn sync_partition(
self: Arc<Self>,
partition: &TodoPartition,
must_exit: &mut watch::Receiver<bool>,
) -> Result<(), Error> {
2020-04-17 16:27:29 +00:00
let my_id = self.table.system.id.clone();
let nodes = self
.table
.replication
2020-04-21 14:07:15 +00:00
.write_nodes(&partition.begin, &self.table.system)
.into_iter()
.filter(|node| *node != my_id)
.collect::<Vec<_>>();
debug!(
"({}) Preparing to sync {:?} with {:?}...",
self.table.name, partition, nodes
);
let root_cks = self
.root_checksum(&partition.begin, &partition.end, must_exit)
.await?;
let mut sync_futures = nodes
.iter()
.map(|node| {
self.clone().do_sync_with(
2020-04-17 19:59:07 +00:00
partition.clone(),
root_cks.clone(),
node.clone(),
partition.retain,
must_exit.clone(),
)
})
2020-04-16 16:41:10 +00:00
.collect::<FuturesUnordered<_>>();
2020-04-19 20:36:36 +00:00
let mut n_errors = 0;
2020-04-16 16:41:10 +00:00
while let Some(r) = sync_futures.next().await {
if let Err(e) = r {
2020-04-19 20:36:36 +00:00
n_errors += 1;
2020-04-21 12:54:55 +00:00
warn!("({}) Sync error: {}", self.table.name, e);
2020-04-16 16:41:10 +00:00
}
}
2020-04-19 20:36:36 +00:00
if n_errors > self.table.replication.max_write_errors() {
2020-04-21 14:07:15 +00:00
return Err(Error::Message(format!(
"Sync failed with too many nodes (should have been: {:?}).",
nodes
)));
2020-04-19 20:36:36 +00:00
}
2020-04-16 16:41:10 +00:00
if !partition.retain {
self.table
.delete_range(&partition.begin, &partition.end)
.await?;
2020-04-16 16:41:10 +00:00
}
Ok(())
}
async fn root_checksum(
self: &Arc<Self>,
begin: &Hash,
end: &Hash,
must_exit: &mut watch::Receiver<bool>,
) -> Result<RangeChecksum, Error> {
2020-04-16 17:28:02 +00:00
for i in 1..MAX_DEPTH {
let rc = self
.range_checksum(
&SyncRange {
begin: begin.to_vec(),
end: end.to_vec(),
level: i,
},
must_exit,
)
.await?;
2020-04-16 16:41:10 +00:00
if rc.found_limit.is_none() {
return Ok(rc);
}
}
Err(Error::Message(format!(
"Unable to compute root checksum (this should never happen"
)))
2020-04-16 16:41:10 +00:00
}
fn range_checksum<'a>(
self: &'a Arc<Self>,
range: &'a SyncRange,
must_exit: &'a mut watch::Receiver<bool>,
) -> BoxFuture<'a, Result<RangeChecksum, Error>> {
2020-04-16 16:41:10 +00:00
async move {
let mut cache = self.cache[range.level].lock().await;
if let Some(v) = cache.get(&range) {
if Instant::now() - v.time < CHECKSUM_CACHE_TIMEOUT {
return Ok(v.clone());
2020-04-16 12:50:49 +00:00
}
2020-04-16 16:41:10 +00:00
}
cache.remove(&range);
drop(cache);
let v = self.range_checksum_inner(&range, must_exit).await?;
2020-04-21 12:54:55 +00:00
trace!(
"({}) New checksum calculated for {}-{}/{}, {} children",
self.table.name,
hex::encode(&range.begin[..]),
hex::encode(&range.end[..]),
range.level,
v.children.len()
);
2020-04-16 16:41:10 +00:00
let mut cache = self.cache[range.level].lock().await;
cache.insert(range.clone(), v.clone());
Ok(v)
}
.boxed()
2020-04-16 16:41:10 +00:00
}
async fn range_checksum_inner(
self: &Arc<Self>,
range: &SyncRange,
must_exit: &mut watch::Receiver<bool>,
) -> Result<RangeChecksum, Error> {
2020-04-16 16:41:10 +00:00
if range.level == 1 {
let mut children = vec![];
for item in self
.table
.store
.range(range.begin.clone()..range.end.clone())
{
2020-04-16 16:41:10 +00:00
let (key, value) = item?;
let key_hash = hash(&key[..]);
if key != range.begin && key_hash.as_slice()[0..range.level].iter().all(|x| *x == 0)
{
return Ok(RangeChecksum {
2020-04-16 16:41:10 +00:00
bounds: range.clone(),
children,
found_limit: Some(key.to_vec()),
time: Instant::now(),
});
2020-04-16 16:41:10 +00:00
}
let item_range = SyncRange {
2020-04-16 16:41:10 +00:00
begin: key.to_vec(),
end: vec![],
level: 0,
};
children.push((item_range, hash(&value[..])));
}
Ok(RangeChecksum {
2020-04-16 16:41:10 +00:00
bounds: range.clone(),
children,
found_limit: None,
time: Instant::now(),
})
} else {
let mut children = vec![];
let mut sub_range = SyncRange {
2020-04-16 16:41:10 +00:00
begin: range.begin.clone(),
end: range.end.clone(),
level: range.level - 1,
};
let mut time = Instant::now();
while !*must_exit.borrow() {
let sub_ck = self.range_checksum(&sub_range, must_exit).await?;
if sub_ck.children.len() > 0 {
let sub_ck_hash = hash(&rmp_to_vec_all_named(&sub_ck)?[..]);
children.push((sub_range.clone(), sub_ck_hash));
if sub_ck.time < time {
time = sub_ck.time;
2020-04-16 12:50:49 +00:00
}
}
2020-04-16 16:41:10 +00:00
if sub_ck.found_limit.is_none() || sub_ck.children.len() == 0 {
return Ok(RangeChecksum {
2020-04-16 16:41:10 +00:00
bounds: range.clone(),
children,
found_limit: None,
time,
});
}
let found_limit = sub_ck.found_limit.unwrap();
let actual_limit_hash = hash(&found_limit[..]);
if actual_limit_hash.as_slice()[0..range.level]
.iter()
.all(|x| *x == 0)
{
return Ok(RangeChecksum {
2020-04-16 16:41:10 +00:00
bounds: range.clone(),
children,
found_limit: Some(found_limit.clone()),
time,
});
}
sub_range.begin = found_limit;
2020-04-16 12:50:49 +00:00
}
2020-04-16 16:41:10 +00:00
Err(Error::Message(format!("Exiting.")))
2020-04-16 12:50:49 +00:00
}
}
async fn do_sync_with(
self: Arc<Self>,
partition: TodoPartition,
root_ck: RangeChecksum,
who: UUID,
retain: bool,
mut must_exit: watch::Receiver<bool>,
) -> Result<(), Error> {
2020-04-16 16:41:10 +00:00
let mut todo = VecDeque::new();
2020-04-17 19:59:07 +00:00
// If their root checksum has level > than us, use that as a reference
let root_cks_resp = self
.table
2020-04-18 17:21:34 +00:00
.rpc_client
.call(
2020-04-17 19:59:07 +00:00
&who,
&TableRPC::<F>::SyncRPC(SyncRPC::GetRootChecksumRange(
partition.begin.clone(),
partition.end.clone(),
)),
TABLE_SYNC_RPC_TIMEOUT,
2020-04-17 19:59:07 +00:00
)
.await?;
if let TableRPC::<F>::SyncRPC(SyncRPC::RootChecksumRange(range)) = root_cks_resp {
if range.level > root_ck.bounds.level {
let their_root_range_ck = self.range_checksum(&range, &mut must_exit).await?;
todo.push_back(their_root_range_ck);
} else {
todo.push_back(root_ck);
}
}
2020-04-16 12:50:49 +00:00
2020-04-16 16:41:10 +00:00
while !todo.is_empty() && !*must_exit.borrow() {
let total_children = todo.iter().map(|x| x.children.len()).fold(0, |x, y| x + y);
2020-04-21 12:54:55 +00:00
trace!(
"({}) Sync with {:?}: {} ({}) remaining",
self.table.name,
who,
todo.len(),
total_children
);
2020-04-16 17:28:02 +00:00
2020-04-16 16:41:10 +00:00
let end = std::cmp::min(16, todo.len());
let step = todo.drain(..end).collect::<Vec<_>>();
2020-04-16 17:28:02 +00:00
let rpc_resp = self
.table
2020-04-18 17:21:34 +00:00
.rpc_client
.call(
&who,
&TableRPC::<F>::SyncRPC(SyncRPC::Checksums(step, retain)),
TABLE_SYNC_RPC_TIMEOUT,
)
.await?;
if let TableRPC::<F>::SyncRPC(SyncRPC::Difference(mut diff_ranges, diff_items)) =
rpc_resp
{
2020-04-19 19:38:45 +00:00
if diff_ranges.len() > 0 || diff_items.len() > 0 {
2020-04-21 12:54:55 +00:00
info!(
2020-04-19 19:38:45 +00:00
"({}) Sync with {:?}: difference {} ranges, {} items",
self.table.name,
who,
diff_ranges.len(),
diff_items.len()
);
}
2020-04-17 16:27:29 +00:00
let mut items_to_send = vec![];
for differing in diff_ranges.drain(..) {
2020-04-16 17:28:02 +00:00
if differing.level == 0 {
2020-04-17 16:27:29 +00:00
items_to_send.push(differing.begin);
2020-04-16 17:28:02 +00:00
} else {
let checksum = self.range_checksum(&differing, &mut must_exit).await?;
todo.push_back(checksum);
}
}
if retain && diff_items.len() > 0 {
2020-04-17 16:27:29 +00:00
self.table.handle_update(diff_items).await?;
}
if items_to_send.len() > 0 {
self.table
.system
.background
2020-04-17 16:27:29 +00:00
.spawn(self.clone().send_items(who.clone(), items_to_send));
2020-04-16 17:28:02 +00:00
}
} else {
return Err(Error::BadRequest(format!(
"Unexpected response to sync RPC checksums: {}",
debug_serialize(&rpc_resp)
)));
2020-04-16 17:28:02 +00:00
}
2020-04-16 16:41:10 +00:00
}
2020-04-16 15:04:28 +00:00
Ok(())
2020-04-16 12:50:49 +00:00
}
2020-04-16 16:41:10 +00:00
2020-04-16 17:28:02 +00:00
async fn send_items(self: Arc<Self>, who: UUID, item_list: Vec<Vec<u8>>) -> Result<(), Error> {
2020-04-21 12:54:55 +00:00
info!(
"({}) Sending {} items to {:?}",
self.table.name,
item_list.len(),
who
);
2020-04-16 17:28:02 +00:00
let mut values = vec![];
for item in item_list.iter() {
if let Some(v) = self.table.store.get(&item[..])? {
values.push(Arc::new(ByteBuf::from(v.as_ref())));
}
}
let rpc_resp = self
.table
2020-04-18 17:21:34 +00:00
.rpc_client
.call(&who, &TableRPC::<F>::Update(values), TABLE_SYNC_RPC_TIMEOUT)
.await?;
2020-04-16 17:28:02 +00:00
if let TableRPC::<F>::Ok = rpc_resp {
Ok(())
} else {
Err(Error::Message(format!(
"Unexpected response to RPC Update: {}",
debug_serialize(&rpc_resp)
)))
2020-04-16 17:28:02 +00:00
}
}
pub async fn handle_rpc(
self: &Arc<Self>,
message: &SyncRPC,
mut must_exit: watch::Receiver<bool>,
) -> Result<SyncRPC, Error> {
2020-04-17 19:59:07 +00:00
match message {
SyncRPC::GetRootChecksumRange(begin, end) => {
let root_cks = self.root_checksum(&begin, &end, &mut must_exit).await?;
Ok(SyncRPC::RootChecksumRange(root_cks.bounds))
}
SyncRPC::Checksums(checksums, retain) => {
self.handle_checksums_rpc(&checksums[..], *retain, &mut must_exit)
.await
}
_ => Err(Error::Message(format!("Unexpected sync RPC"))),
}
}
2020-04-17 16:27:29 +00:00
async fn handle_checksums_rpc(
2020-04-17 19:59:07 +00:00
self: &Arc<Self>,
checksums: &[RangeChecksum],
retain: bool,
must_exit: &mut watch::Receiver<bool>,
) -> Result<SyncRPC, Error> {
let mut ret_ranges = vec![];
let mut ret_items = vec![];
for ckr in checksums.iter() {
let our_ckr = self.range_checksum(&ckr.bounds, must_exit).await?;
for (range, hash) in ckr.children.iter() {
// Only consider items that are in the intersection of the two ranges
// (other ranges will be exchanged at some point)
if our_ckr
.found_limit
.as_ref()
.map(|x| range.begin.as_slice() >= x.as_slice())
.unwrap_or(false)
{
break;
}
let differs = match our_ckr
.children
.binary_search_by(|(our_range, _)| our_range.begin.cmp(&range.begin))
{
Err(_) => true,
Ok(i) => our_ckr.children[i].1 != *hash,
};
if differs {
ret_ranges.push(range.clone());
if retain && range.level == 0 {
if let Some(item_bytes) = self.table.store.get(range.begin.as_slice())? {
ret_items.push(Arc::new(ByteBuf::from(item_bytes.to_vec())));
2020-04-16 16:41:10 +00:00
}
2020-04-17 16:27:29 +00:00
}
}
2020-04-17 19:59:07 +00:00
}
for (range, _hash) in our_ckr.children.iter() {
if ckr
.found_limit
.as_ref()
.map(|x| range.begin.as_slice() >= x.as_slice())
.unwrap_or(false)
{
break;
}
2020-04-17 16:27:29 +00:00
2020-04-17 19:59:07 +00:00
let not_present = ckr
.children
.binary_search_by(|(their_range, _)| their_range.begin.cmp(&range.begin))
.is_err();
if not_present {
if range.level > 0 {
ret_ranges.push(range.clone());
}
if retain && range.level == 0 {
if let Some(item_bytes) = self.table.store.get(range.begin.as_slice())? {
ret_items.push(Arc::new(ByteBuf::from(item_bytes.to_vec())));
}
2020-04-16 16:41:10 +00:00
}
}
}
}
2020-04-17 19:59:07 +00:00
let n_checksums = checksums
.iter()
.map(|x| x.children.len())
.fold(0, |x, y| x + y);
2020-04-19 19:38:45 +00:00
if ret_ranges.len() > 0 || ret_items.len() > 0 {
2020-04-21 12:54:55 +00:00
trace!(
2020-04-19 19:38:45 +00:00
"({}) Checksum comparison RPC: {} different + {} items for {} received",
self.table.name,
ret_ranges.len(),
ret_items.len(),
n_checksums
);
}
2020-04-17 19:59:07 +00:00
Ok(SyncRPC::Difference(ret_ranges, ret_items))
2020-04-16 16:41:10 +00:00
}
2020-04-16 17:28:02 +00:00
pub async fn invalidate(self: Arc<Self>, item_key: Vec<u8>) -> Result<(), Error> {
for i in 1..MAX_DEPTH {
let needle = SyncRange {
2020-04-16 17:28:02 +00:00
begin: item_key.to_vec(),
end: vec![],
level: i,
};
let mut cache = self.cache[i].lock().await;
if let Some(cache_entry) = cache.range(..=needle).rev().next() {
if cache_entry.0.begin <= item_key && cache_entry.0.end > item_key {
let index = cache_entry.0.clone();
drop(cache_entry);
cache.remove(&index);
}
}
}
Ok(())
}
2020-04-16 12:50:49 +00:00
}
impl SyncTodo {
fn add_full_scan<F: TableSchema, R: TableReplication>(&mut self, table: &Table<F, R>) {
2020-04-16 12:50:49 +00:00
let my_id = table.system.id.clone();
self.todo.clear();
let ring = table.system.ring.borrow().clone();
let split_points = table.replication.split_points(&ring);
for i in 0..split_points.len() - 1 {
let begin = split_points[i].clone();
let end = split_points[i + 1].clone();
let nodes = table.replication.replication_nodes(&begin, &ring);
let retain = nodes.contains(&my_id);
if !retain {
// Check if we have some data to send, otherwise skip
if table
.store
.range(begin.clone()..end.clone())
.next()
.is_none()
{
continue;
}
2020-04-16 15:04:28 +00:00
}
self.todo.push(TodoPartition { begin, end, retain });
2020-04-16 12:50:49 +00:00
}
}
fn add_ring_difference<F: TableSchema, R: TableReplication>(
2020-04-16 12:50:49 +00:00
&mut self,
table: &Table<F, R>,
old_ring: &Ring,
new_ring: &Ring,
2020-04-16 12:50:49 +00:00
) {
2020-04-16 15:04:28 +00:00
let my_id = table.system.id.clone();
let mut all_points = None
.into_iter()
.chain(table.replication.split_points(old_ring).drain(..))
.chain(table.replication.split_points(new_ring).drain(..))
.chain(self.todo.iter().map(|x| x.begin.clone()))
.chain(self.todo.iter().map(|x| x.end.clone()))
.collect::<Vec<_>>();
all_points.sort();
all_points.dedup();
let mut old_todo = std::mem::replace(&mut self.todo, vec![]);
old_todo.sort_by(|x, y| x.begin.cmp(&y.begin));
2020-04-16 15:04:28 +00:00
let mut new_todo = vec![];
2020-04-16 15:04:28 +00:00
for i in 0..all_points.len() - 1 {
let begin = all_points[i].clone();
let end = all_points[i + 1].clone();
let was_ours = table
.replication
.replication_nodes(&begin, &old_ring)
2020-04-16 15:04:28 +00:00
.contains(&my_id);
let is_ours = table
.replication
.replication_nodes(&begin, &new_ring)
2020-04-16 15:04:28 +00:00
.contains(&my_id);
let was_todo = match old_todo.binary_search_by(|x| x.begin.cmp(&begin)) {
2020-04-16 15:04:28 +00:00
Ok(_) => true,
Err(j) => {
(j > 0 && old_todo[j - 1].begin < end && begin < old_todo[j - 1].end)
|| (j < old_todo.len()
&& old_todo[j].begin < end && begin < old_todo[j].end)
2020-04-16 15:04:28 +00:00
}
};
if was_todo || (is_ours && !was_ours) || (was_ours && !is_ours) {
new_todo.push(TodoPartition {
2020-04-16 15:04:28 +00:00
begin,
end,
retain: is_ours,
});
}
}
self.todo = new_todo;
2020-04-16 12:50:49 +00:00
}
fn pop_task(&mut self) -> Option<TodoPartition> {
2020-04-16 12:50:49 +00:00
if self.todo.is_empty() {
return None;
}
let i = rand::thread_rng().gen_range::<usize, _, _>(0, self.todo.len());
if i == self.todo.len() - 1 {
self.todo.pop()
} else {
let replacement = self.todo.pop().unwrap();
let ret = std::mem::replace(&mut self.todo[i], replacement);
Some(ret)
}
}
}