Add a retry layer to all providers (#224)

* Add a `ReceiptRetryLayer` for providers

* Fix the retry layer

* Rename the retry layer

* Remove outdated polling function

* Remoe unneeded dependencies
This commit is contained in:
Omar
2026-01-13 22:39:47 +03:00
committed by GitHub
parent d38e6d419d
commit 2d3602aaed
17 changed files with 492 additions and 528 deletions
@@ -1,5 +1,3 @@
use std::{borrow::Cow, fmt::Display};
use alloy::{
eips::BlockNumberOrTag,
network::{Network, TransactionBuilder},
@@ -111,28 +109,23 @@ where
},
state_overrides: Default::default(),
block_overrides: Default::default(),
tx_index: Default::default(),
},
)
.await?
.try_into_call_frame()
.map_err(|err| {
RpcError::LocalUsageError(
FallbackGasFillerError::new(format!(
"Expected a callframe trace, but got: {err:?}"
))
.boxed(),
RpcError::local_usage_str(
format!("Expected a callframe trace, but got: {err:?}").as_str(),
)
})?;
let gas_used = u64::try_from(trace.gas_used).map_err(|_| {
RpcError::LocalUsageError(
FallbackGasFillerError::new(
"Transaction trace returned a value of gas used that exceeds u64",
)
.boxed(),
RpcError::local_usage_str(
"Transaction trace returned a value of gas used that exceeds u64",
)
})?;
let gas_limit = gas_used.saturating_mul(120) / 100;
let gas_limit = gas_used.saturating_mul(2);
if let Some(gas_price) = tx.gas_price() {
return Ok(GasFillable::Legacy {
@@ -174,24 +167,3 @@ impl Default for FallbackGasFiller {
Self::new()
}
}
#[derive(Clone, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)]
struct FallbackGasFillerError(Cow<'static, str>);
impl FallbackGasFillerError {
pub fn new(string: impl Into<Cow<'static, str>>) -> Self {
Self(string.into())
}
pub fn boxed(self) -> Box<Self> {
Box::new(self)
}
}
impl Display for FallbackGasFillerError {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
Display::fmt(&self.0, f)
}
}
impl std::error::Error for FallbackGasFillerError {}
+2
View File
@@ -1,7 +1,9 @@
mod concurrency_limiter;
mod fallback_gas_filler;
mod provider;
mod receipt_retry_layer;
pub use concurrency_limiter::*;
pub use fallback_gas_filler::*;
pub use provider::*;
pub use receipt_retry_layer::*;
+5 -73
View File
@@ -1,18 +1,16 @@
use std::{ops::ControlFlow, sync::LazyLock, time::Duration};
use std::sync::LazyLock;
use alloy::{
network::{Ethereum, Network, NetworkWallet, TransactionBuilder4844},
network::{Network, NetworkWallet, TransactionBuilder4844},
providers::{
Identity, PendingTransactionBuilder, Provider, ProviderBuilder, RootProvider,
Identity, ProviderBuilder, RootProvider,
fillers::{ChainIdFiller, FillProvider, JoinFill, NonceFiller, TxFiller, WalletFiller},
},
rpc::client::ClientBuilder,
};
use anyhow::{Context, Result};
use revive_dt_common::futures::{PollingWaitBehavior, poll};
use tracing::{Instrument, debug, info, info_span};
use crate::provider_utils::{ConcurrencyLimiterLayer, FallbackGasFiller};
use crate::provider_utils::{ConcurrencyLimiterLayer, FallbackGasFiller, RetryLayer};
pub type ConcreteProvider<N, W> = FillProvider<
JoinFill<
@@ -48,6 +46,7 @@ where
let client = ClientBuilder::default()
.layer(GLOBAL_CONCURRENCY_LIMITER_LAYER.clone())
.layer(RetryLayer::default())
.connect(rpc_url)
.await
.context("Failed to construct the RPC client")?;
@@ -63,70 +62,3 @@ where
Ok(provider)
}
pub async fn execute_transaction<N, W>(
provider: ConcreteProvider<N, W>,
transaction: N::TransactionRequest,
) -> Result<N::ReceiptResponse>
where
N: Network<
TransactionRequest: TransactionBuilder4844,
TxEnvelope = <Ethereum as Network>::TxEnvelope,
>,
W: NetworkWallet<N>,
Identity: TxFiller<N>,
FallbackGasFiller: TxFiller<N>,
ChainIdFiller: TxFiller<N>,
NonceFiller: TxFiller<N>,
WalletFiller<W>: TxFiller<N>,
{
let sendable_transaction = provider
.fill(transaction)
.await
.context("Failed to fill transaction")?;
let transaction_envelope = sendable_transaction
.try_into_envelope()
.context("Failed to convert transaction into an envelope")?;
let tx_hash = *transaction_envelope.tx_hash();
let mut pending_transaction = match provider.send_tx_envelope(transaction_envelope).await {
Ok(pending_transaction) => pending_transaction,
Err(error) => {
let error_string = error.to_string();
if error_string.contains("Transaction Already Imported") {
PendingTransactionBuilder::<N>::new(provider.root().clone(), tx_hash)
} else {
return Err(error).context(format!("Failed to submit transaction {tx_hash}"));
}
}
};
debug!(%tx_hash, "Submitted Transaction");
pending_transaction.set_timeout(Some(Duration::from_secs(120)));
let tx_hash = pending_transaction.watch().await.context(format!(
"Transaction inclusion watching timeout for {tx_hash}"
))?;
poll(
Duration::from_secs(60),
PollingWaitBehavior::Constant(Duration::from_secs(3)),
|| {
let provider = provider.clone();
async move {
match provider.get_transaction_receipt(tx_hash).await {
Ok(Some(receipt)) => {
info!("Found the transaction receipt");
Ok(ControlFlow::Break(receipt))
}
_ => Ok(ControlFlow::Continue(())),
}
}
},
)
.instrument(info_span!("Polling for receipt", %tx_hash))
.await
.context(format!("Polling for receipt failed for {tx_hash}"))
}
@@ -0,0 +1,158 @@
use std::time::Duration;
use alloy::{
network::{AnyNetwork, Network},
rpc::json_rpc::{RequestPacket, ResponsePacket},
transports::{TransportError, TransportErrorKind, TransportFut},
};
use tokio::time::{interval, timeout};
use tower::{Layer, Service};
/// A layer that allows for automatic retries for getting the receipt.
///
/// There are certain cases where getting the receipt of a committed transaction might fail. In Geth
/// this can happen if the transaction has been committed to the ledger but has not been indexed, in
/// the substrate and revive stack it can also happen for other reasons.
///
/// Therefore, just because the first attempt to get the receipt (after transaction confirmation)
/// has failed it doesn't mean that it will continue to fail. This layer can be added to any alloy
/// provider to allow the provider to retry getting the receipt for some period of time before it
/// considers that a timeout. It attempts to poll for the receipt for the `polling_duration` with an
/// interval of `polling_interval` between each poll. If by the end of the `polling_duration` it was
/// not able to get the receipt successfully then this is considered to be a timeout.
///
/// Additionally, this layer allows for retries for other rpc methods such as all tracing methods.
#[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)]
pub struct RetryLayer {
/// The amount of time to keep polling for the receipt before considering it a timeout.
polling_duration: Duration,
/// The interval of time to wait between each poll for the receipt.
polling_interval: Duration,
}
impl RetryLayer {
pub fn new(polling_duration: Duration, polling_interval: Duration) -> Self {
Self {
polling_duration,
polling_interval,
}
}
pub fn with_polling_duration(mut self, polling_duration: Duration) -> Self {
self.polling_duration = polling_duration;
self
}
pub fn with_polling_interval(mut self, polling_interval: Duration) -> Self {
self.polling_interval = polling_interval;
self
}
}
impl Default for RetryLayer {
fn default() -> Self {
Self {
polling_duration: Duration::from_secs(90),
polling_interval: Duration::from_millis(500),
}
}
}
impl<S> Layer<S> for RetryLayer {
type Service = RetryService<S>;
fn layer(&self, inner: S) -> Self::Service {
RetryService {
service: inner,
polling_duration: self.polling_duration,
polling_interval: self.polling_interval,
}
}
}
#[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)]
pub struct RetryService<S> {
/// The internal service.
service: S,
/// The amount of time to keep polling for the receipt before considering it a timeout.
polling_duration: Duration,
/// The interval of time to wait between each poll for the receipt.
polling_interval: Duration,
}
impl<S> Service<RequestPacket> for RetryService<S>
where
S: Service<RequestPacket, Future = TransportFut<'static>, Error = TransportError>
+ Send
+ 'static
+ Clone,
{
type Response = ResponsePacket;
type Error = TransportError;
type Future = TransportFut<'static>;
fn poll_ready(
&mut self,
cx: &mut std::task::Context<'_>,
) -> std::task::Poll<Result<(), Self::Error>> {
self.service.poll_ready(cx)
}
#[allow(clippy::nonminimal_bool)]
fn call(&mut self, req: RequestPacket) -> Self::Future {
type ReceiptOutput = <AnyNetwork as Network>::ReceiptResponse;
let mut service = self.service.clone();
let polling_interval = self.polling_interval;
let polling_duration = self.polling_duration;
Box::pin(async move {
let request = req.as_single().ok_or_else(|| {
TransportErrorKind::custom_str("Retry layer doesn't support batch requests")
})?;
let method = request.method();
let requires_retries = method == "eth_getTransactionReceipt"
|| (method.contains("debug") && method.contains("trace"));
if !requires_retries {
return service.call(req).await;
}
timeout(polling_duration, async {
let mut interval = interval(polling_interval);
loop {
interval.tick().await;
let Ok(resp) = service.call(req.clone()).await else {
continue;
};
let response = resp.as_single().expect("Can't fail");
if response.is_error() {
continue;
}
if method == "eth_getTransactionReceipt"
&& response
.payload()
.clone()
.deserialize_success::<ReceiptOutput>()
.ok()
.and_then(|resp| resp.try_into_success().ok())
.is_some()
|| method != "eth_getTransactionReceipt"
{
return resp;
} else {
continue;
}
}
})
.await
.map_err(|_| TransportErrorKind::custom_str("Timeout when retrying request"))
})
}
}