feat(aggregation mode): add retry logic to batches download

Makefile

@@ -298,7 +298,7 @@ verify_aggregated_proof_risc0:
 		--rpc_url $(RPC_URL)
 
 proof_aggregator_install: ## Install the aggregation mode with proving enabled
-	cargo install --path aggregation_mode --features prove,gpu --bin proof_aggregator --locked
+	cargo install --path aggregation_mode --features prove,gpu --bin proof_aggregator_gpu --locked
 
 proof_aggregator_write_program_ids: ## Write proof aggregator zkvm programs ids
 	@cd aggregation_mode && ./scripts/build_programs.sh

aggregation_mode/Cargo.toml

@@ -18,6 +18,7 @@ reqwest = { version = "0.12" }
 ciborium = "=0.2.2"
 lambdaworks-crypto = { git = "https://github.com/lambdaclass/lambdaworks.git", rev = "5f8f2cfcc8a1a22f77e8dff2d581f1166eefb80b", features = ["serde"]}
 rayon = "1.10.0"
+backon = "1.2.0"
 # Necessary for the VerificationData type
 aligned-sdk = { path = "../crates/sdk/" }
 # zkvms

aggregation_mode/src/backend/mod.rs

@@ -1,6 +1,7 @@
 pub mod config;
 pub mod fetcher;
 mod merkle_tree;
+mod retry;
 mod s3;
 mod types;
 

aggregation_mode/src/backend/retry.rs

@@ -0,0 +1,62 @@
+use backon::ExponentialBuilder;
+use backon::Retryable;
+use std::{future::Future, time::Duration};
+
+#[derive(Debug)]
+pub enum RetryError<E> {
+    Transient(E),
+    Permanent(E),
+}
+
+impl<E: std::fmt::Display> std::fmt::Display for RetryError<E> {
+    fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
+        match self {
+            RetryError::Transient(e) => write!(f, "{}", e),
+            RetryError::Permanent(e) => write!(f, "{}", e),
+        }
+    }
+}
+
+impl<E> RetryError<E> {
+    pub fn inner(self) -> E {
+        match self {
+            RetryError::Transient(e) => e,
+            RetryError::Permanent(e) => e,
+        }
+    }
+}
+
+impl<E: std::fmt::Display> std::error::Error for RetryError<E> where E: std::fmt::Debug {}
+
+/// Supports retries only on async functions. See: https://docs.rs/backon/latest/backon/#retry-an-async-function
+/// Runs with `jitter: false`.
+///
+/// # Parameters
+/// * `function` - The async function to retry
+/// * `min_delay_millis` - Initial delay before first retry attempt (in milliseconds)
+/// * `factor` - Exponential backoff multiplier for retry delays
+/// * `max_times` - Maximum number of retry attempts
+/// * `max_delay_seconds` - Maximum delay between retry attempts (in seconds)
+pub async fn retry_function<FutureFn, Fut, T, E>(
+    function: FutureFn,
+    min_delay_millis: u64,
+    factor: f32,
+    max_times: usize,
+    max_delay_seconds: u64,
+) -> Result<T, RetryError<E>>
+where
+    Fut: Future<Output = Result<T, RetryError<E>>>,
+    FutureFn: FnMut() -> Fut,
+{
+    let backoff = ExponentialBuilder::default()
+        .with_min_delay(Duration::from_millis(min_delay_millis))
+        .with_max_times(max_times)
+        .with_factor(factor)
+        .with_max_delay(Duration::from_secs(max_delay_seconds));
+
+    function
+        .retry(backoff)
+        .sleep(tokio::time::sleep)
+        .when(|e| matches!(e, RetryError::Transient(_)))
+        .await
+}

aggregation_mode/src/backend/s3.rs

@@ -1,4 +1,7 @@
+use crate::backend::retry::{retry_function, RetryError};
 use aligned_sdk::common::types::VerificationData;
+use std::time::Duration;
+use tracing::{info, warn};
 
 #[derive(Debug)]
 #[allow(dead_code)]
@@ -13,38 +16,93 @@ pub enum GetBatchProofsError {
 // needed to make S3 bucket work
 const DEFAULT_USER_AGENT: &str = "proof-aggregator/aligned-layer";
 
-pub async fn get_aligned_batch_from_s3(
+// Retry parameters for S3 requests
+/// Initial delay before first retry attempt (in milliseconds)
+const RETRY_MIN_DELAY_MILLIS: u64 = 500;
+/// Exponential backoff multiplier for retry delays
+const RETRY_FACTOR: f32 = 2.0;
+/// Maximum number of retry attempts
+const RETRY_MAX_TIMES: usize = 5;
+/// Maximum delay between retry attempts (in seconds)
+const RETRY_MAX_DELAY_SECONDS: u64 = 10;
+
+/// Timeout for establishing a connection to S3
+const CONNECT_TIMEOUT_SECONDS: Duration = Duration::from_secs(10);
+/// Timeout for Batch Download Requests
+const BATCH_DOWNLOAD_TIMEOUT_SECONDS: Duration = Duration::from_secs(5 * 60);
+
+async fn get_aligned_batch_from_s3_retryable(
     url: String,
-) -> Result<Vec<VerificationData>, GetBatchProofsError> {
+) -> Result<Vec<VerificationData>, RetryError<GetBatchProofsError>> {
+    info!("Fetching batch from S3 URL: {}", url);
     let client = reqwest::Client::builder()
         .user_agent(DEFAULT_USER_AGENT)
+        .connect_timeout(CONNECT_TIMEOUT_SECONDS)
+        .timeout(BATCH_DOWNLOAD_TIMEOUT_SECONDS)
         .build()
-        .map_err(|e| GetBatchProofsError::ReqwestClientFailed(e.to_string()))?;
+        .map_err(|e| {
+            RetryError::Permanent(GetBatchProofsError::ReqwestClientFailed(e.to_string()))
+        })?;
+
+    let response = client.get(&url).send().await.map_err(|e| {
+        warn!("Failed to send request to {}: {}", url, e);
+        RetryError::Transient(GetBatchProofsError::FetchingS3Batch(e.to_string()))
+    })?;
 
-    let response = client
-        .get(url)
-        .send()
-        .await
-        .map_err(|e| GetBatchProofsError::FetchingS3Batch(e.to_string()))?;
     if !response.status().is_success() {
-        return Err(GetBatchProofsError::StatusFailed((
-            response.status().as_u16(),
-            response
-                .status()
-                .canonical_reason()
-                .unwrap_or("")
-                .to_string(),
-        )));
+        let status_code = response.status().as_u16();
+        let reason = response
+            .status()
+            .canonical_reason()
+            .unwrap_or("")
+            .to_string();
+
+        // Determine if the error is retryable based on status code
+        let error = GetBatchProofsError::StatusFailed((status_code, reason));
+        return match status_code {
+            // Client errors (4xx) are generally permanent, except for specific cases
+            400..=499 => match status_code {
+                408 | 429 => Err(RetryError::Transient(error)), // Request Timeout, Too Many Requests
+                _ => Err(RetryError::Permanent(error)),
+            },
+            // Server errors (5xx) are generally transient
+            500..=599 => Err(RetryError::Transient(error)),
+            _ => Err(RetryError::Permanent(error)),
+        };
     }
 
-    let bytes = response
-        .bytes()
-        .await
-        .map_err(|e| GetBatchProofsError::EmptyBody(e.to_string()))?;
+    let bytes = response.bytes().await.map_err(|e| {
+        warn!("Failed to read response body from {}: {}", url, e);
+        RetryError::Transient(GetBatchProofsError::EmptyBody(e.to_string()))
+    })?;
     let bytes: &[u8] = bytes.iter().as_slice();
 
-    let data: Vec<VerificationData> = ciborium::from_reader(bytes)
-        .map_err(|e| GetBatchProofsError::Deserialization(e.to_string()))?;
+    let data: Vec<VerificationData> = ciborium::from_reader(bytes).map_err(|e| {
+        warn!("Failed to deserialize batch data from {}: {}", url, e);
+        RetryError::Permanent(GetBatchProofsError::Deserialization(e.to_string()))
+    })?;
 
     Ok(data)
 }
+
+/// Download batch from Storage Service using the provided URL.
+///
+/// Retries on recoverable errors using exponential backoff up to `RETRY_MAX_TIMES` times:
+/// (0,5 secs - 1 secs - 2 secs - 4 secs - 8 secs).
+pub async fn get_aligned_batch_from_s3(
+    url: String,
+) -> Result<Vec<VerificationData>, GetBatchProofsError> {
+    let url_clone = url.clone();
+    retry_function(
+        move || {
+            let url = url_clone.clone();
+            get_aligned_batch_from_s3_retryable(url)
+        },
+        RETRY_MIN_DELAY_MILLIS,
+        RETRY_FACTOR,
+        RETRY_MAX_TIMES,
+        RETRY_MAX_DELAY_SECONDS,
+    )
+    .await
+    .map_err(|retry_err| retry_err.inner())
+}