web3-proxy/web3_proxy/src/rpcs/connection.rs

748 lines
29 KiB
Rust
Raw Normal View History

2022-08-27 02:44:25 +03:00
///! Rate-limited communication with a web3 provider.
2022-08-30 23:01:42 +03:00
use super::blockchain::{ArcBlock, BlockHashesMap, BlockId};
use super::provider::Web3Provider;
2022-08-30 23:01:42 +03:00
use super::request::{OpenRequestHandle, OpenRequestResult};
use crate::app::{flatten_handle, AnyhowJoinHandle};
use crate::config::BlockAndRpc;
2022-06-16 05:53:37 +03:00
use anyhow::Context;
2022-07-19 04:31:12 +03:00
use ethers::prelude::{Block, Bytes, Middleware, ProviderError, TxHash, H256, U64};
2022-06-16 05:53:37 +03:00
use futures::future::try_join_all;
2022-05-05 22:07:09 +03:00
use futures::StreamExt;
2022-08-10 05:37:34 +03:00
use parking_lot::RwLock;
use redis_rate_limit::{RedisPool, RedisRateLimit, ThrottleResult};
2022-05-21 01:16:15 +03:00
use serde::ser::{SerializeStruct, Serializer};
use serde::Serialize;
2022-05-05 22:07:09 +03:00
use std::fmt;
use std::hash::{Hash, Hasher};
2022-07-19 04:31:12 +03:00
use std::sync::atomic::{self, AtomicU32, AtomicU64};
2022-05-05 22:07:09 +03:00
use std::{cmp::Ordering, sync::Arc};
2022-06-29 22:15:05 +03:00
use tokio::sync::broadcast;
2022-08-10 05:37:34 +03:00
use tokio::sync::RwLock as AsyncRwLock;
2022-08-24 03:14:49 +03:00
use tokio::time::{interval, sleep, sleep_until, Duration, MissedTickBehavior};
use tracing::{error, info, instrument, trace, warn};
2022-06-14 08:43:28 +03:00
2022-08-24 02:13:56 +03:00
/// An active connection to a Web3Rpc
pub struct Web3Connection {
2022-08-24 03:11:49 +03:00
pub name: String,
2022-08-24 02:13:56 +03:00
/// TODO: can we get this from the provider? do we even need it?
2022-08-26 20:26:17 +03:00
url: String,
2022-08-24 02:13:56 +03:00
/// keep track of currently open requests. We sort on this
2022-08-24 03:11:49 +03:00
pub(super) active_requests: AtomicU32,
2022-08-24 02:13:56 +03:00
/// provider is in a RwLock so that we can replace it if re-connecting
/// it is an async lock because we hold it open across awaits
2022-08-24 03:11:49 +03:00
pub(super) provider: AsyncRwLock<Option<Arc<Web3Provider>>>,
2022-08-24 02:13:56 +03:00
/// rate limits are stored in a central redis so that multiple proxies can share their rate limits
hard_limit: Option<RedisRateLimit>,
/// used for load balancing to the least loaded server
2022-08-26 20:26:17 +03:00
pub(super) soft_limit: u32,
2022-09-05 08:53:58 +03:00
/// TODO: have an enum for this so that "no limit" prints pretty
2022-08-24 02:13:56 +03:00
block_data_limit: AtomicU64,
2022-08-26 20:26:17 +03:00
/// Lower weight are higher priority when sending requests
pub(super) weight: u32,
// TODO: async lock?
pub(super) head_block: RwLock<BlockId>,
2022-08-24 02:13:56 +03:00
}
2022-05-05 22:07:09 +03:00
impl Web3Connection {
/// Connect to a web3 rpc
2022-06-14 08:43:28 +03:00
// #[instrument(name = "spawn_Web3Connection", skip(hard_limit, http_client))]
// TODO: have this take a builder (which will have channels attached)
2022-06-14 08:43:28 +03:00
#[allow(clippy::too_many_arguments)]
pub async fn spawn(
2022-08-10 08:56:09 +03:00
name: String,
chain_id: u64,
2022-05-05 22:07:09 +03:00
url_str: String,
2022-05-15 04:51:24 +03:00
// optional because this is only used for http providers. websocket providers don't use it
http_client: Option<reqwest::Client>,
2022-06-29 22:15:05 +03:00
http_interval_sender: Option<Arc<broadcast::Sender<()>>>,
// TODO: have a builder struct for this.
hard_limit: Option<(u64, RedisPool)>,
2022-05-05 22:07:09 +03:00
// TODO: think more about this type
soft_limit: u32,
2022-08-28 02:49:41 +03:00
block_map: BlockHashesMap,
2022-07-22 08:11:26 +03:00
block_sender: Option<flume::Sender<BlockAndRpc>>,
2022-06-14 08:43:28 +03:00
tx_id_sender: Option<flume::Sender<(TxHash, Arc<Self>)>>,
reconnect: bool,
2022-08-08 22:57:54 +03:00
weight: u32,
2022-06-14 08:43:28 +03:00
) -> anyhow::Result<(Arc<Web3Connection>, AnyhowJoinHandle<()>)> {
2022-05-22 21:39:06 +03:00
let hard_limit = hard_limit.map(|(hard_rate_limit, redis_conection)| {
2022-08-06 05:29:55 +03:00
// TODO: allow configurable period and max_burst
RedisRateLimit::new(
redis_conection,
2022-08-06 08:46:33 +03:00
"web3_proxy",
2022-08-06 05:29:55 +03:00
&format!("{}:{}", chain_id, url_str),
2022-05-22 02:34:05 +03:00
hard_rate_limit,
2022-08-30 23:01:42 +03:00
60.0,
2022-05-22 02:34:05 +03:00
)
});
2022-05-05 22:07:09 +03:00
2022-05-17 05:26:47 +03:00
let provider = Web3Provider::from_str(&url_str, http_client).await?;
2022-05-05 22:07:09 +03:00
2022-07-09 07:25:59 +03:00
let new_connection = Self {
2022-08-10 08:56:09 +03:00
name,
2022-05-05 22:07:09 +03:00
url: url_str.clone(),
2022-05-15 04:51:24 +03:00
active_requests: 0.into(),
2022-08-10 05:37:34 +03:00
provider: AsyncRwLock::new(Some(Arc::new(provider))),
2022-05-22 02:34:05 +03:00
hard_limit,
2022-05-05 22:07:09 +03:00
soft_limit,
2022-07-19 04:31:12 +03:00
block_data_limit: Default::default(),
2022-08-26 20:26:17 +03:00
head_block: RwLock::new(Default::default()),
2022-08-08 22:57:54 +03:00
weight,
2022-05-12 21:49:57 +03:00
};
2022-07-09 07:25:59 +03:00
let new_connection = Arc::new(new_connection);
2022-05-12 21:49:57 +03:00
2022-05-15 04:51:24 +03:00
// check the server's chain_id here
2022-07-19 04:31:12 +03:00
// TODO: move this outside the `new` function and into a `start` function or something. that way we can do retries from there
2022-05-15 04:51:24 +03:00
// TODO: some public rpcs (on bsc and fantom) do not return an id and so this ends up being an error
2022-06-17 01:23:41 +03:00
// TODO: this will wait forever. do we want that?
let found_chain_id: Result<U64, _> = new_connection
.wait_for_request_handle()
2022-08-06 08:26:43 +03:00
.await?
2022-05-15 04:51:24 +03:00
.request("eth_chainId", Option::None::<()>)
2022-05-13 00:20:33 +03:00
.await;
match found_chain_id {
Ok(found_chain_id) => {
2022-06-29 22:15:05 +03:00
// TODO: there has to be a cleaner way to do this
if chain_id != found_chain_id.as_u64() {
2022-05-13 00:20:33 +03:00
return Err(anyhow::anyhow!(
"incorrect chain id! Expected {}. Found {}",
chain_id,
found_chain_id
)
2022-07-19 09:41:04 +03:00
.context(format!("failed @ {}", new_connection)));
2022-05-13 00:20:33 +03:00
}
}
Err(e) => {
2022-07-19 09:41:04 +03:00
let e = anyhow::Error::from(e).context(format!("failed @ {}", new_connection));
2022-05-13 00:20:33 +03:00
return Err(e);
}
2022-05-12 21:49:57 +03:00
}
let will_subscribe_to_blocks = block_sender.is_some();
2022-07-19 04:31:12 +03:00
// subscribe to new blocks and new transactions
// TODO: make transaction subscription optional (just pass None for tx_id_sender)
2022-06-14 08:43:28 +03:00
let handle = {
2022-07-09 07:25:59 +03:00
let new_connection = new_connection.clone();
2022-06-14 08:43:28 +03:00
tokio::spawn(async move {
2022-07-09 07:25:59 +03:00
new_connection
2022-08-26 20:26:17 +03:00
.subscribe(
http_interval_sender,
block_map,
block_sender,
tx_id_sender,
reconnect,
)
2022-06-14 08:43:28 +03:00
.await
})
};
2022-07-19 04:31:12 +03:00
// we could take "archive" as a parameter, but we would want a safety check on it regardless
2022-07-19 07:22:02 +03:00
// check common archive thresholds
2022-07-23 03:19:13 +03:00
// TODO: would be great if rpcs exposed this
// TODO: move this to a helper function so we can recheck on errors or as the chain grows
2022-08-03 03:27:26 +03:00
// TODO: move this to a helper function that checks
if will_subscribe_to_blocks {
// TODO: make sure the server isn't still syncing
2022-07-19 04:31:12 +03:00
// TODO: don't sleep. wait for new heads subscription instead
// TODO: i think instead of atomics, we could maybe use a watch channel
sleep(Duration::from_millis(250)).await;
2022-07-19 04:31:12 +03:00
2022-08-27 05:13:36 +03:00
new_connection.check_block_data_limit().await?;
}
2022-07-19 04:31:12 +03:00
2022-08-27 05:13:36 +03:00
Ok((new_connection, handle))
}
2022-07-19 04:31:12 +03:00
2022-08-27 05:13:36 +03:00
#[instrument]
async fn check_block_data_limit(self: &Arc<Self>) -> anyhow::Result<Option<u64>> {
let mut limit = None;
2022-08-27 05:13:36 +03:00
for block_data_limit in [u64::MAX, 90_000, 128, 64, 32] {
let mut head_block_num = self.head_block.read().num;
2022-08-27 05:13:36 +03:00
// TODO: wait until head block is set outside the loop? if we disconnect while starting we could actually get 0 though
while head_block_num == U64::zero() {
warn!("no head block");
2022-07-19 04:31:12 +03:00
2022-08-27 05:13:36 +03:00
// TODO: subscribe to a channel instead of polling? subscribe to http_interval_sender?
sleep(Duration::from_secs(1)).await;
2022-08-27 05:13:36 +03:00
head_block_num = self.head_block.read().num;
}
2022-08-27 05:13:36 +03:00
// TODO: subtract 1 from block_data_limit for safety?
let maybe_archive_block = head_block_num
.saturating_sub((block_data_limit).into())
.max(U64::one());
let archive_result: Result<Bytes, _> = self
.wait_for_request_handle()
.await?
.request(
"eth_getCode",
(
"0xdead00000000000000000000000000000000beef",
maybe_archive_block,
),
)
.await;
trace!(?archive_result, %self);
if archive_result.is_ok() {
limit = Some(block_data_limit);
break;
2022-07-19 04:31:12 +03:00
}
}
2022-08-27 05:13:36 +03:00
if let Some(limit) = limit {
self.block_data_limit
.store(limit, atomic::Ordering::Release);
}
2022-07-19 04:31:12 +03:00
2022-08-27 05:13:36 +03:00
Ok(limit)
2022-07-09 07:25:59 +03:00
}
2022-07-19 04:31:12 +03:00
/// TODO: this might be too simple. different nodes can prune differently
2022-07-25 03:27:00 +03:00
pub fn block_data_limit(&self) -> U64 {
2022-07-19 04:31:12 +03:00
self.block_data_limit.load(atomic::Ordering::Acquire).into()
}
2022-07-22 22:30:39 +03:00
pub fn has_block_data(&self, needed_block_num: &U64) -> bool {
2022-07-25 03:27:00 +03:00
let block_data_limit: U64 = self.block_data_limit();
2022-07-19 04:31:12 +03:00
2022-08-26 20:26:17 +03:00
let newest_block_num = self.head_block.read().num;
2022-07-19 04:31:12 +03:00
let oldest_block_num = newest_block_num
.saturating_sub(block_data_limit)
.max(U64::one());
2022-07-22 22:30:39 +03:00
needed_block_num >= &oldest_block_num && needed_block_num <= &newest_block_num
2022-05-05 22:07:09 +03:00
}
#[instrument(skip_all)]
pub async fn reconnect(
self: &Arc<Self>,
2022-07-22 08:11:26 +03:00
block_sender: Option<flume::Sender<BlockAndRpc>>,
) -> anyhow::Result<()> {
// websocket doesn't need the http client
let http_client = None;
2022-09-03 05:59:30 +03:00
info!(%self, "reconnecting");
2022-07-19 04:31:12 +03:00
// since this lock is held open over an await, we use tokio's locking
2022-07-19 04:31:12 +03:00
// TODO: timeout on this lock. if its slow, something is wrong
let mut provider = self.provider.write().await;
*provider = None;
2022-07-25 03:27:00 +03:00
// tell the block subscriber that we don't have any blocks
2022-06-16 05:53:37 +03:00
if let Some(block_sender) = block_sender {
2022-09-05 06:40:13 +03:00
// TODO: maybe it would be better do send a "None" or an Option<Arc<Block<TxHash>>>
2022-06-16 05:53:37 +03:00
block_sender
2022-07-22 08:11:26 +03:00
.send_async((Arc::new(Block::default()), self.clone()))
2022-06-16 05:53:37 +03:00
.await
.context("block_sender at 0")?;
}
2022-06-16 05:53:37 +03:00
// TODO: if this fails, keep retrying
let new_provider = Web3Provider::from_str(&self.url, http_client).await?;
*provider = Some(Arc::new(new_provider));
Ok(())
}
2022-05-06 09:07:01 +03:00
#[inline]
2022-05-05 22:07:09 +03:00
pub fn active_requests(&self) -> u32 {
self.active_requests.load(atomic::Ordering::Acquire)
}
2022-06-04 01:22:55 +03:00
#[inline]
pub async fn has_provider(&self) -> bool {
self.provider.read().await.is_some()
}
2022-05-17 03:56:56 +03:00
#[instrument(skip_all)]
2022-08-26 20:26:17 +03:00
async fn send_head_block_result(
self: &Arc<Self>,
2022-08-30 23:01:42 +03:00
new_head_block: Result<ArcBlock, ProviderError>,
2022-07-22 08:11:26 +03:00
block_sender: &flume::Sender<BlockAndRpc>,
2022-08-28 02:49:41 +03:00
block_map: BlockHashesMap,
2022-05-30 07:30:13 +03:00
) -> anyhow::Result<()> {
2022-08-26 20:26:17 +03:00
match new_head_block {
2022-08-30 23:01:42 +03:00
Ok(mut new_head_block) => {
2022-08-26 20:26:17 +03:00
// TODO: is unwrap_or_default ok? we might have an empty block
let new_hash = new_head_block.hash.unwrap_or_default();
// if we already have this block saved, we don't need to store this copy
2022-09-05 08:53:58 +03:00
// TODO: small race here
new_head_block = if let Some(existing_block) = block_map.get(&new_hash) {
// we only save blocks with a total difficulty
debug_assert!(existing_block.total_difficulty.is_some());
existing_block
} else if new_head_block.total_difficulty.is_some() {
// this block has a total difficulty, it is safe to use
block_map.insert(new_hash, new_head_block).await;
// we get instead of return new_head_block just in case there was a race
// TODO: but how bad is this race? it might be fine
block_map.get(&new_hash).expect("we just inserted")
} else {
// Cache miss and NO TOTAL DIFFICULTY!
2022-08-30 23:01:42 +03:00
// self got the head block first. unfortunately its missing a necessary field
// keep this even after https://github.com/ledgerwatch/erigon/issues/5190 is closed.
// there are other clients and we might have to use a third party without the td fix.
trace!(rpc=?self, ?new_hash, "total_difficulty missing");
2022-09-05 08:53:58 +03:00
// todo: this can wait forever!
2022-08-30 23:01:42 +03:00
let complete_head_block: Block<TxHash> = self
.wait_for_request_handle()
.await?
.request("eth_getBlockByHash", (new_hash, false))
.await?;
2022-09-05 08:53:58 +03:00
debug_assert!(complete_head_block.total_difficulty.is_some());
block_map
.insert(new_hash, Arc::new(complete_head_block))
.await;
// we get instead of return new_head_block just in case there was a race
// TODO: but how bad is this race? it might be fine
block_map.get(&new_hash).expect("we just inserted")
};
2022-08-30 23:01:42 +03:00
2022-08-26 20:26:17 +03:00
let new_num = new_head_block.number.unwrap_or_default();
// save the block so we don't send the same one multiple times
// also save so that archive checks can know how far back to query
{
2022-07-19 04:31:12 +03:00
let mut head_block = self.head_block.write();
2022-08-26 20:26:17 +03:00
head_block.hash = new_hash;
head_block.num = new_num;
2022-07-19 04:31:12 +03:00
}
2022-06-16 05:53:37 +03:00
block_sender
2022-08-26 20:26:17 +03:00
.send_async((new_head_block, self.clone()))
2022-06-16 05:53:37 +03:00
.await
.context("block_sender")?;
}
Err(e) => {
warn!("unable to get block from {}: {}", self, e);
2022-08-26 20:26:17 +03:00
// TODO: do something to rpc_chain?
2022-08-07 09:48:57 +03:00
// send an empty block to take this server out of rotation
block_sender
2022-08-26 20:26:17 +03:00
.send_async((Arc::new(Block::default()), self.clone()))
2022-08-07 09:48:57 +03:00
.await
.context("block_sender")?;
}
}
2022-05-30 07:30:13 +03:00
Ok(())
}
2022-09-02 23:46:39 +03:00
#[instrument(skip_all)]
2022-06-14 08:43:28 +03:00
async fn subscribe(
self: Arc<Self>,
2022-06-29 22:15:05 +03:00
http_interval_sender: Option<Arc<broadcast::Sender<()>>>,
2022-08-28 02:49:41 +03:00
block_map: BlockHashesMap,
2022-07-22 08:11:26 +03:00
block_sender: Option<flume::Sender<BlockAndRpc>>,
2022-06-14 08:43:28 +03:00
tx_id_sender: Option<flume::Sender<(TxHash, Arc<Self>)>>,
reconnect: bool,
) -> anyhow::Result<()> {
2022-06-16 05:53:37 +03:00
loop {
2022-07-19 07:24:16 +03:00
let http_interval_receiver = http_interval_sender.as_ref().map(|x| x.subscribe());
2022-06-29 22:15:05 +03:00
2022-06-16 05:53:37 +03:00
let mut futures = vec![];
if let Some(block_sender) = &block_sender {
2022-08-26 20:26:17 +03:00
let f = self.clone().subscribe_new_heads(
http_interval_receiver,
block_sender.clone(),
block_map.clone(),
);
2022-06-16 05:53:37 +03:00
futures.push(flatten_handle(tokio::spawn(f)));
}
if let Some(tx_id_sender) = &tx_id_sender {
let f = self
.clone()
.subscribe_pending_transactions(tx_id_sender.clone());
futures.push(flatten_handle(tokio::spawn(f)));
}
if futures.is_empty() {
2022-06-14 08:43:28 +03:00
// TODO: is there a better way to make a channel that is never ready?
2022-06-16 05:53:37 +03:00
info!(?self, "no-op subscription");
return Ok(());
}
2022-06-16 05:53:37 +03:00
match try_join_all(futures).await {
Ok(_) => break,
Err(err) => {
if reconnect {
// TODO: exponential backoff
let retry_in = Duration::from_secs(1);
warn!(
?self,
"subscription exited. Attempting to reconnect in {:?}. {:?}",
retry_in,
err
);
sleep(retry_in).await;
// TODO: loop on reconnecting! do not return with a "?" here
// TODO: this isn't going to work. it will get in a loop with newHeads
self.reconnect(block_sender.clone()).await?;
} else {
error!(?self, ?err, "subscription exited");
return Err(err);
}
2022-06-14 08:43:28 +03:00
}
}
}
Ok(())
}
2022-05-17 05:26:47 +03:00
/// Subscribe to new blocks. If `reconnect` is true, this runs forever.
2022-05-17 03:56:56 +03:00
#[instrument(skip_all)]
async fn subscribe_new_heads(
2022-05-05 22:07:09 +03:00
self: Arc<Self>,
2022-06-29 22:15:05 +03:00
http_interval_receiver: Option<broadcast::Receiver<()>>,
2022-07-22 08:11:26 +03:00
block_sender: flume::Sender<BlockAndRpc>,
2022-08-28 02:49:41 +03:00
block_map: BlockHashesMap,
2022-05-05 22:07:09 +03:00
) -> anyhow::Result<()> {
2022-09-03 05:59:30 +03:00
info!(%self, "watching new heads");
// TODO: is a RwLock of an Option<Arc> the right thing here?
if let Some(provider) = self.provider.read().await.clone() {
match &*provider {
Web3Provider::Http(_provider) => {
// there is a "watch_blocks" function, but a lot of public nodes do not support the necessary rpc endpoints
2022-06-29 22:15:05 +03:00
// TODO: try watch_blocks and fall back to this?
let mut http_interval_receiver = http_interval_receiver.unwrap();
2022-07-19 04:31:12 +03:00
let mut last_hash = H256::zero();
loop {
2022-08-26 20:26:17 +03:00
// TODO: try, or wait_for?
match self.wait_for_request_handle().await {
Ok(active_request_handle) => {
let block: Result<Block<TxHash>, _> = active_request_handle
.request("eth_getBlockByNumber", ("latest", false))
.await;
2022-08-26 20:26:17 +03:00
match block {
Ok(block) => {
// don't send repeat blocks
let new_hash = block
.hash
.expect("blocks here should always have hashes");
if new_hash != last_hash {
// new hash!
last_hash = new_hash;
self.send_head_block_result(
Ok(Arc::new(block)),
&block_sender,
block_map.clone(),
)
2022-08-07 09:48:57 +03:00
.await?;
2022-08-26 20:26:17 +03:00
}
}
Err(err) => {
// we did not get a block back. something is up with the server. take it out of rotation
self.send_head_block_result(
Err(err),
&block_sender,
block_map.clone(),
)
.await?;
2022-07-19 07:31:30 +03:00
}
2022-05-16 22:15:40 +03:00
}
}
2022-07-09 01:14:45 +03:00
Err(err) => {
2022-08-11 00:29:50 +03:00
warn!(?err, "Internal error on latest block from {}", self);
// TODO: what should we do? sleep? extra time?
2022-05-16 22:15:40 +03:00
}
}
2022-07-19 04:31:12 +03:00
2022-08-26 20:26:17 +03:00
// wait for the next interval
2022-07-19 04:31:12 +03:00
// TODO: if error or rate limit, increase interval?
while let Err(err) = http_interval_receiver.recv().await {
match err {
broadcast::error::RecvError::Closed => {
2022-08-26 20:26:17 +03:00
// channel is closed! that's not good. bubble the error up
2022-07-19 04:31:12 +03:00
return Err(err.into());
}
broadcast::error::RecvError::Lagged(lagged) => {
2022-08-26 20:26:17 +03:00
// querying the block was delayed
// this can happen if tokio is very busy or waiting for requests limits took too long
2022-07-19 04:31:12 +03:00
warn!(?err, ?self, "http interval lagging by {}!", lagged);
}
}
}
trace!(?self, "ok http interval");
2022-05-15 22:28:22 +03:00
}
}
Web3Provider::Ws(provider) => {
2022-08-30 23:01:42 +03:00
// todo: move subscribe_blocks onto the request handle?
let active_request_handle = self.wait_for_request_handle().await;
let mut stream = provider.subscribe_blocks().await?;
drop(active_request_handle);
// query the block once since the subscription doesn't send the current block
// there is a very small race condition here where the stream could send us a new block right now
// all it does is print "new block" for the same block as current block
2022-08-30 23:01:42 +03:00
let block: Result<ArcBlock, _> = self
.wait_for_request_handle()
2022-08-06 08:26:43 +03:00
.await?
.request("eth_getBlockByNumber", ("latest", false))
2022-08-26 20:26:17 +03:00
.await
.map(Arc::new);
2022-08-26 20:26:17 +03:00
self.send_head_block_result(block, &block_sender, block_map.clone())
.await?;
2022-07-09 01:14:45 +03:00
while let Some(new_block) = stream.next().await {
2022-08-26 20:26:17 +03:00
self.send_head_block_result(
Ok(Arc::new(new_block)),
&block_sender,
block_map.clone(),
)
.await?;
2022-05-17 05:26:47 +03:00
}
2022-07-09 01:14:45 +03:00
warn!(?self, "subscription ended");
2022-05-05 22:07:09 +03:00
}
2022-05-17 05:26:47 +03:00
}
}
2022-05-17 03:56:56 +03:00
Ok(())
}
2022-05-17 05:26:47 +03:00
#[instrument(skip_all)]
async fn subscribe_pending_transactions(
self: Arc<Self>,
tx_id_sender: flume::Sender<(TxHash, Arc<Self>)>,
) -> anyhow::Result<()> {
2022-09-03 05:59:30 +03:00
info!(%self, "watching pending transactions");
// TODO: is a RwLock of an Option<Arc> the right thing here?
if let Some(provider) = self.provider.read().await.clone() {
match &*provider {
2022-06-18 10:06:54 +03:00
Web3Provider::Http(provider) => {
// there is a "watch_pending_transactions" function, but a lot of public nodes do not support the necessary rpc endpoints
// TODO: what should this interval be? probably automatically set to some fraction of block time
// TODO: maybe it would be better to have one interval for all of the http providers, but this works for now
// TODO: if there are some websocket providers, maybe have a longer interval and a channel that tells the https to update when a websocket gets a new head? if they are slow this wouldn't work well though
2022-06-14 08:43:28 +03:00
let mut interval = interval(Duration::from_secs(60));
interval.set_missed_tick_behavior(MissedTickBehavior::Delay);
loop {
2022-06-14 08:43:28 +03:00
// TODO: actually do something here
/*
match self.try_request_handle().await {
Ok(active_request_handle) => {
// TODO: check the filter
2022-07-22 22:30:39 +03:00
todo!("actually send a request");
}
Err(e) => {
warn!("Failed getting latest block from {}: {:?}", self, e);
}
}
2022-06-14 08:43:28 +03:00
*/
2022-07-07 06:22:09 +03:00
// wait for the interval
// TODO: if error or rate limit, increase interval?
interval.tick().await;
}
}
Web3Provider::Ws(provider) => {
2022-07-09 02:02:32 +03:00
// TODO: maybe the subscribe_pending_txs function should be on the active_request_handle
let active_request_handle = self.wait_for_request_handle().await;
let mut stream = provider.subscribe_pending_txs().await?;
drop(active_request_handle);
2022-07-08 21:27:06 +03:00
while let Some(pending_tx_id) = stream.next().await {
tx_id_sender
.send_async((pending_tx_id, self.clone()))
.await
.context("tx_id_sender")?;
2022-08-11 00:29:50 +03:00
// TODO: periodically check for listeners. if no one is subscribed, unsubscribe and wait for a subscription
}
2022-07-08 21:27:06 +03:00
2022-08-11 05:57:01 +03:00
warn!(?self, "subscription ended");
}
2022-05-05 22:07:09 +03:00
}
}
Ok(())
}
2022-08-30 23:01:42 +03:00
/// be careful with this; it might wait forever!
2022-08-24 03:59:05 +03:00
// TODO: maximum wait time?
2022-08-30 23:01:42 +03:00
#[instrument]
2022-08-24 03:14:49 +03:00
pub async fn wait_for_request_handle(self: &Arc<Self>) -> anyhow::Result<OpenRequestHandle> {
2022-07-09 02:02:32 +03:00
// TODO: maximum wait time? i think timeouts in other parts of the code are probably best
2022-05-16 22:15:40 +03:00
2022-06-17 01:23:41 +03:00
loop {
2022-08-30 23:01:42 +03:00
let x = self.try_request_handle().await;
trace!(?x, "try_request_handle");
match x {
2022-08-24 03:59:05 +03:00
Ok(OpenRequestResult::Handle(handle)) => return Ok(handle),
2022-08-24 03:14:49 +03:00
Ok(OpenRequestResult::RetryAt(retry_at)) => {
2022-08-07 09:48:57 +03:00
// TODO: emit a stat?
2022-08-30 23:01:42 +03:00
trace!(?retry_at);
2022-08-07 09:48:57 +03:00
sleep_until(retry_at).await;
}
2022-08-30 23:01:42 +03:00
Ok(OpenRequestResult::RetryNever) => {
2022-08-24 03:59:05 +03:00
// TODO: when can this happen? log? emit a stat?
// TODO: subscribe to the head block on this
2022-08-07 09:48:57 +03:00
// TODO: sleep how long? maybe just error?
2022-08-30 23:01:42 +03:00
return Err(anyhow::anyhow!("unable to retry"));
}
2022-08-07 09:48:57 +03:00
Err(err) => return Err(err),
}
2022-05-05 22:07:09 +03:00
}
}
2022-08-30 23:01:42 +03:00
#[instrument]
2022-08-26 20:26:17 +03:00
pub async fn try_request_handle(self: &Arc<Self>) -> anyhow::Result<OpenRequestResult> {
2022-06-04 01:22:55 +03:00
// check that we are connected
if !self.has_provider().await {
2022-08-07 09:48:57 +03:00
// TODO: emit a stat?
2022-08-30 23:01:42 +03:00
return Ok(OpenRequestResult::RetryNever);
2022-06-04 01:22:55 +03:00
}
2022-05-05 22:07:09 +03:00
// check rate limits
2022-05-22 02:34:05 +03:00
if let Some(ratelimiter) = self.hard_limit.as_ref() {
match ratelimiter.throttle().await {
2022-08-07 09:48:57 +03:00
Ok(ThrottleResult::Allowed) => {
2022-08-30 23:01:42 +03:00
trace!("rate limit succeeded")
2022-05-05 22:07:09 +03:00
}
2022-08-07 09:48:57 +03:00
Ok(ThrottleResult::RetryAt(retry_at)) => {
2022-05-05 22:07:09 +03:00
// rate limit failed
2022-05-22 02:34:05 +03:00
// save the smallest retry_after. if nothing succeeds, return an Err with retry_after in it
2022-05-05 22:07:09 +03:00
// TODO: use tracing better
2022-06-25 06:33:49 +03:00
// TODO: i'm seeing "Exhausted rate limit on moralis: 0ns". How is it getting 0?
2022-08-07 09:48:57 +03:00
warn!(?retry_at, ?self, "Exhausted rate limit");
2022-05-05 22:07:09 +03:00
2022-08-24 03:14:49 +03:00
return Ok(OpenRequestResult::RetryAt(retry_at.into()));
2022-08-07 09:48:57 +03:00
}
Ok(ThrottleResult::RetryNever) => {
return Err(anyhow::anyhow!("Rate limit failed."));
}
2022-08-07 09:48:57 +03:00
Err(err) => {
return Err(err);
2022-05-05 22:07:09 +03:00
}
}
};
2022-08-24 03:14:49 +03:00
let handle = OpenRequestHandle::new(self.clone());
2022-08-24 03:11:49 +03:00
2022-08-24 03:59:05 +03:00
Ok(OpenRequestResult::Handle(handle))
2022-08-24 03:11:49 +03:00
}
}
2022-08-07 09:48:57 +03:00
2022-08-24 03:11:49 +03:00
impl fmt::Debug for Web3Provider {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
// TODO: the default Debug takes forever to write. this is too quiet though. we at least need the url
f.debug_struct("Web3Provider").finish_non_exhaustive()
}
}
impl Hash for Web3Connection {
fn hash<H: Hasher>(&self, state: &mut H) {
// TODO: is this enough?
self.name.hash(state);
}
}
2022-05-05 22:07:09 +03:00
impl Eq for Web3Connection {}
impl Ord for Web3Connection {
fn cmp(&self, other: &Self) -> std::cmp::Ordering {
self.name.cmp(&other.name)
2022-05-05 22:07:09 +03:00
}
}
impl PartialOrd for Web3Connection {
fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
Some(self.cmp(other))
}
}
impl PartialEq for Web3Connection {
fn eq(&self, other: &Self) -> bool {
self.name == other.name
2022-05-05 22:07:09 +03:00
}
}
2022-08-10 08:56:09 +03:00
impl Serialize for Web3Connection {
fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
where
S: Serializer,
{
// 3 is the number of fields in the struct.
let mut state = serializer.serialize_struct("Web3Connection", 4)?;
// the url is excluded because it likely includes private information. just show the name
state.serialize_field("name", &self.name)?;
let block_data_limit = self.block_data_limit.load(atomic::Ordering::Relaxed);
state.serialize_field("block_data_limit", &block_data_limit)?;
state.serialize_field("soft_limit", &self.soft_limit)?;
state.serialize_field(
"active_requests",
&self.active_requests.load(atomic::Ordering::Relaxed),
)?;
state.end()
}
}
impl fmt::Debug for Web3Connection {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
let mut f = f.debug_struct("Web3Connection");
f.field("name", &self.name);
2022-08-10 08:56:09 +03:00
let block_data_limit = self.block_data_limit.load(atomic::Ordering::Relaxed);
if block_data_limit == u64::MAX {
f.field("data", &"archive");
} else {
f.field("data", &block_data_limit);
}
f.finish_non_exhaustive()
}
}
impl fmt::Display for Web3Connection {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
// TODO: filter basic auth and api keys
write!(f, "{}", &self.name)
2022-08-10 08:56:09 +03:00
}
}