0g-storage-node/node/pruner/src/lib.rs
0g-peterzhb d43a616b56
Some checks failed
abi-consistent-check / build-and-compare (push) Has been cancelled
code-coverage / unittest-cov (push) Has been cancelled
rust / check (push) Has been cancelled
rust / test (push) Has been cancelled
rust / lints (push) Has been cancelled
functional-test / test (push) Has been cancelled
fix shard config init (#354)
2025-03-12 22:27:34 +08:00

295 lines
10 KiB
Rust

use anyhow::{bail, Result};
use contract_interface::ChunkLinearReward;
use ethereum_types::Address;
use ethers::prelude::{Http, Provider};
use ethers::providers::{HttpRateLimitRetryPolicy, RetryClient, RetryClientBuilder};
use miner::MinerMessage;
use rand::Rng;
use std::cmp::Ordering;
use std::path::PathBuf;
use std::str::FromStr;
use std::sync::Arc;
use std::time::Duration;
use storage::config::{ShardConfig, SHARD_CONFIG_KEY};
use storage::log_store::log_manager::{DATA_DB_KEY, PORA_CHUNK_SIZE};
use storage_async::Store;
use task_executor::TaskExecutor;
use tokio::sync::{broadcast, mpsc};
use tracing::{debug, error, info};
use zgs_spec::SECTORS_PER_PRICING;
// Start pruning when the db directory size exceeds 0.9 * limit.
const PRUNE_THRESHOLD: f32 = 0.9;
const FIRST_REWARDABLE_CHUNK_KEY: &str = "first_rewardable_chunk";
const CHUNKS_PER_PRICING: u64 = (SECTORS_PER_PRICING / PORA_CHUNK_SIZE) as u64;
#[derive(Debug)]
pub struct PrunerConfig {
pub shard_config: ShardConfig,
pub db_path: PathBuf,
pub max_num_sectors: usize,
pub check_time: Duration,
pub batch_size: usize,
pub batch_wait_time: Duration,
pub rpc_endpoint_url: String,
pub reward_address: Address,
pub rate_limit_retries: u32,
pub timeout_retries: u32,
pub initial_backoff: u64,
}
impl PrunerConfig {
fn start_prune_size(&self) -> u64 {
(self.max_num_sectors as f32 * PRUNE_THRESHOLD) as u64
}
}
pub struct Pruner {
config: PrunerConfig,
first_rewardable_chunk: u64,
first_tx_seq: u64,
store: Arc<Store>,
sender: mpsc::UnboundedSender<PrunerMessage>,
miner_sender: Option<broadcast::Sender<MinerMessage>>,
reward_contract: ChunkLinearReward<Arc<Provider<RetryClient<Http>>>>,
}
impl Pruner {
pub async fn spawn(
executor: TaskExecutor,
mut config: PrunerConfig,
store: Arc<Store>,
miner_sender: Option<broadcast::Sender<MinerMessage>>,
) -> Result<mpsc::UnboundedReceiver<PrunerMessage>> {
if let Some(shard_config) = get_shard_config(store.as_ref()).await? {
config.shard_config = shard_config;
}
let (first_rewardable_chunk, first_tx_seq) = get_first_rewardable_chunk(store.as_ref())
.await?
.unwrap_or((0, 0));
let provider = Arc::new(Provider::new(
RetryClientBuilder::default()
.rate_limit_retries(config.rate_limit_retries)
.timeout_retries(config.timeout_retries)
.initial_backoff(Duration::from_millis(config.initial_backoff))
.build(
Http::from_str(&config.rpc_endpoint_url)?,
Box::new(HttpRateLimitRetryPolicy),
),
));
let reward_contract = ChunkLinearReward::new(config.reward_address, Arc::new(provider));
let (tx, rx) = mpsc::unbounded_channel();
let pruner = Pruner {
config,
first_rewardable_chunk,
first_tx_seq,
store,
sender: tx,
miner_sender,
reward_contract,
};
pruner.put_shard_config().await?;
executor.spawn(
async move {
pruner.start().await.expect("pruner error");
},
"pruner",
);
Ok(rx)
}
pub async fn start(mut self) -> Result<()> {
loop {
// Check shard config update and prune unneeded data.
if let Some(delete_list) = self.maybe_update().await? {
info!(new_config = ?self.config.shard_config, "new shard config");
self.put_shard_config().await?;
self.prune_in_batch(delete_list).await?;
}
// Check no reward chunks and prune.
match self.reward_contract.first_rewardable_chunk().call().await {
Ok(new_first_rewardable) => {
if let Some(no_reward_list) = self
.maybe_forward_first_rewardable(new_first_rewardable)
.await?
{
info!(
?new_first_rewardable,
"first rewardable chunk moves forward, start pruning"
);
self.prune_tx(
self.first_rewardable_chunk * SECTORS_PER_PRICING as u64,
new_first_rewardable * SECTORS_PER_PRICING as u64,
)
.await?;
self.prune_in_batch(no_reward_list).await?;
self.first_rewardable_chunk = new_first_rewardable;
self.put_first_rewardable_chunk_index(
self.first_rewardable_chunk,
self.first_tx_seq,
)
.await?;
}
}
e => {
error!("handle reward contract read fails, e={:?}", e);
}
};
tokio::time::sleep(self.config.check_time).await;
}
}
async fn maybe_update(&mut self) -> Result<Option<Box<dyn Send + Iterator<Item = u64>>>> {
let current_size = self.store.get_num_entries().await?;
debug!(
current_size = current_size,
config = ?self.config.shard_config,
"maybe_update"
);
if current_size < self.config.start_prune_size() {
Ok(None)
} else {
// Update config and generate delete list should be done in a single lock to ensure
// the list is complete.
let config = &mut self.config.shard_config;
let old_shard_id = config.shard_id;
let old_num_shard = config.num_shard;
// Update new config
let rand_bit = {
let mut rng = rand::thread_rng();
rng.gen::<bool>()
};
config.shard_id = old_shard_id + rand_bit as usize * old_num_shard;
config.num_shard *= 2;
// Generate delete list
let flow_len = self
.store
.get_context()
.await?
.1
.div_ceil(PORA_CHUNK_SIZE as u64);
let start_index = old_shard_id + (!rand_bit) as usize * old_num_shard;
Ok(Some(Box::new(
(start_index as u64..flow_len).step_by(config.num_shard),
)))
}
}
async fn maybe_forward_first_rewardable(
&mut self,
new_first_rewardable: u64,
) -> Result<Option<Box<dyn Send + Iterator<Item = u64>>>> {
match self.first_rewardable_chunk.cmp(&new_first_rewardable) {
Ordering::Less => Ok(Some(Box::new(
self.first_rewardable_chunk * CHUNKS_PER_PRICING
..new_first_rewardable * CHUNKS_PER_PRICING,
))),
Ordering::Equal => Ok(None),
Ordering::Greater => {
error!(
"Unexpected first_rewardable_chunk revert: old={} new={}",
self.first_rewardable_chunk, new_first_rewardable
);
Ok(None)
}
}
}
async fn prune_in_batch(&self, to_prune: Box<dyn Send + Iterator<Item = u64>>) -> Result<()> {
let mut batch = Vec::with_capacity(self.config.batch_size);
let mut iter = to_prune.peekable();
while let Some(index) = iter.next() {
batch.push(index);
if batch.len() == self.config.batch_size || iter.peek().is_none() {
debug!(start = batch.first(), end = batch.last(), "prune batch");
self.store.remove_chunks_batch(&batch).await?;
batch = Vec::with_capacity(self.config.batch_size);
tokio::time::sleep(self.config.batch_wait_time).await;
}
}
Ok(())
}
async fn prune_tx(&mut self, start_sector: u64, end_sector: u64) -> Result<()> {
loop {
if let Some(tx) = self.store.get_tx_by_seq_number(self.first_tx_seq).await? {
// If a part of the tx data is pruned, we mark the tx as pruned.
if tx.start_entry_index() >= start_sector && tx.start_entry_index() < end_sector {
self.store.prune_tx(tx.seq).await?;
} else if tx.start_entry_index() >= end_sector {
break;
} else {
bail!(
"prune tx out of range: tx={:?}, start={} end={}",
tx,
start_sector,
end_sector
);
}
self.first_tx_seq += 1;
} else {
// Wait for `first_tx_seq` to be processed.
tokio::time::sleep(Duration::from_secs(60)).await;
}
}
Ok(())
}
async fn put_shard_config(&self) -> Result<()> {
if let Some(sender) = &self.miner_sender {
sender.send(MinerMessage::SetShardConfig(self.config.shard_config))?;
}
self.sender
.send(PrunerMessage::ChangeShardConfig(self.config.shard_config))?;
self.store
.update_shard_config(self.config.shard_config)
.await;
self.store
.set_config_encoded(&SHARD_CONFIG_KEY, &self.config.shard_config, DATA_DB_KEY)
.await
}
async fn put_first_rewardable_chunk_index(
&self,
new_first_rewardable_chunk: u64,
new_first_tx_seq: u64,
) -> Result<()> {
self.store
.set_config_encoded(
&FIRST_REWARDABLE_CHUNK_KEY,
&(new_first_rewardable_chunk, new_first_tx_seq),
DATA_DB_KEY,
)
.await
}
}
pub async fn get_shard_config(store: &Store) -> Result<Option<ShardConfig>> {
store
.get_config_decoded(&SHARD_CONFIG_KEY, DATA_DB_KEY)
.await
}
async fn get_first_rewardable_chunk(store: &Store) -> Result<Option<(u64, u64)>> {
store
.get_config_decoded(&FIRST_REWARDABLE_CHUNK_KEY, DATA_DB_KEY)
.await
}
#[derive(Debug)]
pub enum PrunerMessage {
ChangeShardConfig(ShardConfig),
}