diff --git a/Cargo.lock b/Cargo.lock index b04b8c049b8..4626d161a4d 100644 --- a/Cargo.lock +++ b/Cargo.lock @@ -2039,6 +2039,7 @@ dependencies = [ "serde", "thiserror 2.0.12", "tiny-keccak 1.5.0", + "tokio", "tonic-build", ] @@ -2144,6 +2145,7 @@ dependencies = [ "serde", "shellexpand", "termcolor", + "tokio", "url", ] diff --git a/chain/ethereum/Cargo.toml b/chain/ethereum/Cargo.toml index 5ea3a97a00c..51aa540e75b 100644 --- a/chain/ethereum/Cargo.toml +++ b/chain/ethereum/Cargo.toml @@ -15,6 +15,7 @@ tiny-keccak = "1.5.0" hex = "0.4.3" semver = "1.0.26" thiserror = { workspace = true } +tokio = { version = "1", features = ["full"] } itertools = "0.14.0" diff --git a/chain/ethereum/src/health.rs b/chain/ethereum/src/health.rs new file mode 100644 index 00000000000..a80e4976601 --- /dev/null +++ b/chain/ethereum/src/health.rs @@ -0,0 +1,71 @@ +use crate::adapter::EthereumAdapter as EthereumAdapterTrait; +use crate::EthereumAdapter; +use std::sync::{Arc, RwLock}; +use std::time::{Duration, Instant}; +use tokio::time::sleep; +#[derive(Debug)] +pub struct Health { + pub provider: Arc, + latency: Arc>, + error_rate: Arc>, + consecutive_failures: Arc>, +} + +impl Health { + pub fn new(provider: Arc) -> Self { + Self { + provider, + latency: Arc::new(RwLock::new(Duration::from_secs(0))), + error_rate: Arc::new(RwLock::new(0.0)), + consecutive_failures: Arc::new(RwLock::new(0)), + } + } + + pub fn provider(&self) -> &str { + self.provider.provider() + } + + pub async fn check(&self) { + let start_time = Instant::now(); + // For now, we'll just simulate a health check. + // In a real implementation, we would send a request to the provider. + let success = self.provider.provider().contains("rpc1"); // Simulate a failure for rpc2 + let latency = start_time.elapsed(); + + self.update_metrics(success, latency); + } + + fn update_metrics(&self, success: bool, latency: Duration) { + let mut latency_w = self.latency.write().unwrap(); + *latency_w = latency; + + let mut error_rate_w = self.error_rate.write().unwrap(); + let mut consecutive_failures_w = self.consecutive_failures.write().unwrap(); + + if success { + *error_rate_w = *error_rate_w * 0.9; // Decay the error rate + *consecutive_failures_w = 0; + } else { + *error_rate_w = *error_rate_w * 0.9 + 0.1; // Increase the error rate + *consecutive_failures_w += 1; + } + } + + pub fn score(&self) -> f64 { + let latency = *self.latency.read().unwrap(); + let error_rate = *self.error_rate.read().unwrap(); + let consecutive_failures = *self.consecutive_failures.read().unwrap(); + + // This is a simple scoring algorithm. A more sophisticated algorithm could be used here. + 1.0 / (1.0 + latency.as_secs_f64() + error_rate + (consecutive_failures as f64)) + } +} + +pub async fn health_check_task(health_checkers: Vec>) { + loop { + for health_checker in &health_checkers { + health_checker.check().await; + } + sleep(Duration::from_secs(10)).await; + } +} diff --git a/chain/ethereum/src/lib.rs b/chain/ethereum/src/lib.rs index fa76f70d799..307d9963078 100644 --- a/chain/ethereum/src/lib.rs +++ b/chain/ethereum/src/lib.rs @@ -5,6 +5,7 @@ pub mod codec; mod data_source; mod env; mod ethereum_adapter; +pub mod health; mod ingestor; mod polling_block_stream; pub mod runtime; diff --git a/chain/ethereum/src/network.rs b/chain/ethereum/src/network.rs index 59a698ab20b..11c70b62991 100644 --- a/chain/ethereum/src/network.rs +++ b/chain/ethereum/src/network.rs @@ -6,8 +6,12 @@ use graph::components::network_provider::ProviderManager; use graph::components::network_provider::ProviderName; use graph::endpoint::EndpointMetrics; use graph::firehose::{AvailableCapacity, SubgraphLimit}; -use graph::prelude::rand::seq::IteratorRandom; -use graph::prelude::rand::{self, Rng}; +use graph::prelude::rand::{ + self, + distr::{weighted::WeightedIndex, Distribution}, + seq::IteratorRandom, + Rng, +}; use itertools::Itertools; use std::sync::Arc; @@ -24,12 +28,13 @@ pub const DEFAULT_ADAPTER_ERROR_RETEST_PERCENT: f64 = 0.2; pub struct EthereumNetworkAdapter { endpoint_metrics: Arc, pub capabilities: NodeCapabilities, - adapter: Arc, + pub adapter: Arc, /// The maximum number of times this adapter can be used. We use the /// strong_count on `adapter` to determine whether the adapter is above /// that limit. That's a somewhat imprecise but convenient way to /// determine the number of connections limit: SubgraphLimit, + weight: f64, } #[async_trait] @@ -53,12 +58,14 @@ impl EthereumNetworkAdapter { capabilities: NodeCapabilities, adapter: Arc, limit: SubgraphLimit, + weight: f64, ) -> Self { Self { endpoint_metrics, capabilities, adapter, limit, + weight, } } @@ -79,6 +86,8 @@ impl EthereumNetworkAdapter { } } +use crate::health::Health; + #[derive(Debug, Clone)] pub struct EthereumNetworkAdapters { chain_id: ChainName, @@ -86,6 +95,8 @@ pub struct EthereumNetworkAdapters { call_only_adapters: Vec, // Percentage of request that should be used to retest errored adapters. retest_percent: f64, + weighted: bool, + health_checkers: Vec>, } impl EthereumNetworkAdapters { @@ -95,6 +106,8 @@ impl EthereumNetworkAdapters { manager: ProviderManager::default(), call_only_adapters: vec![], retest_percent: DEFAULT_ADAPTER_ERROR_RETEST_PERCENT, + weighted: false, + health_checkers: vec![], } } @@ -121,7 +134,7 @@ impl EthereumNetworkAdapters { ProviderCheckStrategy::MarkAsValid, ); - Self::new(chain_id, provider, call_only, None) + Self::new(chain_id, provider, call_only, None, false, vec![]) } pub fn new( @@ -129,6 +142,8 @@ impl EthereumNetworkAdapters { manager: ProviderManager, call_only_adapters: Vec, retest_percent: Option, + weighted: bool, + health_checkers: Vec>, ) -> Self { #[cfg(debug_assertions)] call_only_adapters.iter().for_each(|a| { @@ -140,6 +155,8 @@ impl EthereumNetworkAdapters { manager, call_only_adapters, retest_percent: retest_percent.unwrap_or(DEFAULT_ADAPTER_ERROR_RETEST_PERCENT), + weighted, + health_checkers, } } @@ -189,50 +206,125 @@ impl EthereumNetworkAdapters { Self::available_with_capabilities(all, required_capabilities) } - // handle adapter selection from a list, implements the availability checking with an abstracted - // source of the adapter list. + /// Main adapter selection entry point that handles both weight-based distribution + /// and error retesting logic. + /// + /// The selection process: + /// 1. First selects an adapter based on weights (if enabled) or random selection + /// 2. Occasionally overrides the selection to retest adapters with errors + /// + /// The error retesting happens AFTER weight-based selection to minimize + /// distribution skew while still allowing periodic health checks of errored endpoints. fn cheapest_from( + &self, + input: Vec<&EthereumNetworkAdapter>, + required_capabilities: &NodeCapabilities, + ) -> Result, Error> { + // Select adapter based on weights or random strategy + let selected_adapter = self.select_best_adapter(input.clone(), required_capabilities)?; + + // Occasionally override selection to retest errored adapters + // This happens AFTER weight-based selection to minimize distribution skew + let retest_rng: f64 = rand::rng().random(); + if retest_rng < self.retest_percent { + // Find the adapter with the highest error count + if let Some(most_errored) = input + .iter() + .max_by_key(|a| a.current_error_count()) + .filter(|a| a.current_error_count() > 0) + { + return Ok(most_errored.adapter.clone()); + } + } + + Ok(selected_adapter) + } + + /// Selects the best adapter based on the configured strategy (weighted or random). + /// If weighted mode is enabled, uses weight-based probabilistic selection. + /// Otherwise, falls back to random selection with error count consideration. + fn select_best_adapter( + &self, + input: Vec<&EthereumNetworkAdapter>, + required_capabilities: &NodeCapabilities, + ) -> Result, Error> { + if self.weighted { + self.select_weighted_adapter(input, required_capabilities) + } else { + self.select_random_adapter(input, required_capabilities) + } + } + + /// Performs weighted random selection of adapters based on their configured weights. + /// + /// Weights are relative values between 0.0 and 1.0 that determine the probability + /// of selecting each adapter. They don't need to sum to 1.0 as they're normalized + /// internally by the WeightedIndex distribution. + /// + /// Falls back to random selection if weights are invalid (e.g., all zeros). + fn select_weighted_adapter( + &self, input: Vec<&EthereumNetworkAdapter>, required_capabilities: &NodeCapabilities, - retest_percent: f64, ) -> Result, Error> { - let retest_rng: f64 = (&mut rand::rng()).random(); + if input.is_empty() { + return Err(anyhow!( + "A matching Ethereum network with {:?} was not found.", + required_capabilities + )); + } + + let weights: Vec<_> = input + .iter() + .map(|a| { + let health_checker = self + .health_checkers + .iter() + .find(|h| h.provider() == a.provider()); + let score = health_checker.map_or(1.0, |h| h.score()); + a.weight * score + }) + .collect(); - let cheapest = input.into_iter().choose_multiple(&mut rand::rng(), 3); - let cheapest = cheapest.iter(); + if let Ok(dist) = WeightedIndex::new(&weights) { + let idx = dist.sample(&mut rand::rng()); + Ok(input[idx].adapter.clone()) + } else { + // Fallback to random selection if weights are invalid + self.select_random_adapter(input, required_capabilities) + } + } - // If request falls below the retest threshold, use this request to try and - // reset the failed adapter. If a request succeeds the adapter will be more - // likely to be selected afterwards. - if retest_rng < retest_percent { - cheapest.max_by_key(|adapter| adapter.current_error_count()) + /// Performs random selection of adapters with preference for those with fewer errors. + /// + /// Randomly selects up to 3 adapters from the available pool, then chooses the one + /// with the lowest error count. This provides a balance between load distribution + /// and avoiding problematic endpoints. + fn select_random_adapter( + &self, + input: Vec<&EthereumNetworkAdapter>, + required_capabilities: &NodeCapabilities, + ) -> Result, Error> { + let choices = input.into_iter().choose_multiple(&mut rand::rng(), 3); + if let Some(adapter) = choices.iter().min_by_key(|a| a.current_error_count()) { + Ok(adapter.adapter.clone()) } else { - // The assumption here is that most RPC endpoints will not have limits - // which makes the check for low/high available capacity less relevant. - // So we essentially assume if it had available capacity when calling - // `all_cheapest_with` then it prolly maintains that state and so we - // just select whichever adapter is working better according to - // the number of errors. - cheapest.min_by_key(|adapter| adapter.current_error_count()) + Err(anyhow!( + "A matching Ethereum network with {:?} was not found.", + required_capabilities + )) } - .map(|adapter| adapter.adapter.clone()) - .ok_or(anyhow!( - "A matching Ethereum network with {:?} was not found.", - required_capabilities - )) } pub(crate) fn unverified_cheapest_with( &self, required_capabilities: &NodeCapabilities, ) -> Result, Error> { - let cheapest = self.all_unverified_cheapest_with(required_capabilities); + let cheapest = self + .all_unverified_cheapest_with(required_capabilities) + .collect_vec(); - Self::cheapest_from( - cheapest.choose_multiple(&mut rand::rng(), 3), - required_capabilities, - self.retest_percent, - ) + self.cheapest_from(cheapest, required_capabilities) } /// This is the public entry point and should always use verified adapters @@ -243,9 +335,9 @@ impl EthereumNetworkAdapters { let cheapest = self .all_cheapest_with(required_capabilities) .await - .choose_multiple(&mut rand::rng(), 3); + .collect_vec(); - Self::cheapest_from(cheapest, required_capabilities, self.retest_percent) + self.cheapest_from(cheapest, required_capabilities) } pub async fn cheapest(&self) -> Option> { @@ -289,7 +381,7 @@ impl EthereumNetworkAdapters { .call_only_adapters .iter() .min_by_key(|x| Arc::strong_count(&x.adapter)) - .ok_or(anyhow!("no available call only endpoints"))?; + .ok_or(anyhow!("no available call only endpoints "))?; // TODO: This will probably blow up a lot sooner than [limit] amount of // subgraphs, since we probably use a few instances. @@ -297,7 +389,7 @@ impl EthereumNetworkAdapters { .limit .has_capacity(Arc::strong_count(&adapters.adapter)) { - bail!("call only adapter has reached the concurrency limit"); + bail!("call only adapter has reached the concurrency limit "); } // Cloning here ensure we have the correct count at any given time, if we return a reference it can be cloned later @@ -308,18 +400,16 @@ impl EthereumNetworkAdapters { #[cfg(test)] mod tests { + use super::Health; use graph::cheap_clone::CheapClone; use graph::components::network_provider::ProviderCheckStrategy; use graph::components::network_provider::ProviderManager; use graph::components::network_provider::ProviderName; use graph::data::value::Word; use graph::http::HeaderMap; + use graph::slog::{o, Discard, Logger}; use graph::{ - endpoint::EndpointMetrics, - firehose::SubgraphLimit, - prelude::MetricsRegistry, - slog::{o, Discard, Logger}, - tokio, + endpoint::EndpointMetrics, firehose::SubgraphLimit, prelude::MetricsRegistry, tokio, url::Url, }; use std::sync::Arc; @@ -429,6 +519,7 @@ mod tests { }, eth_adapter.clone(), SubgraphLimit::Limit(3), + 1.0, )], vec![EthereumNetworkAdapter::new( metrics.cheap_clone(), @@ -438,6 +529,7 @@ mod tests { }, eth_call_adapter.clone(), SubgraphLimit::Limit(3), + 1.0, )], ) .await; @@ -535,6 +627,7 @@ mod tests { }, eth_call_adapter.clone(), SubgraphLimit::Unlimited, + 1.0, )], vec![EthereumNetworkAdapter::new( metrics.cheap_clone(), @@ -544,6 +637,7 @@ mod tests { }, eth_adapter.clone(), SubgraphLimit::Limit(2), + 1.0, )], ) .await; @@ -606,6 +700,7 @@ mod tests { }, eth_call_adapter.clone(), SubgraphLimit::Disabled, + 1.0, )], vec![EthereumNetworkAdapter::new( metrics.cheap_clone(), @@ -615,6 +710,7 @@ mod tests { }, eth_adapter.clone(), SubgraphLimit::Limit(3), + 1.0, )], ) .await; @@ -661,6 +757,7 @@ mod tests { }, eth_adapter.clone(), SubgraphLimit::Limit(3), + 1.0, )], vec![], ) @@ -731,6 +828,7 @@ mod tests { }, adapter: adapter.clone(), limit: limit.clone(), + weight: 1.0, }); always_retest_adapters.push(EthereumNetworkAdapter { endpoint_metrics: metrics.clone(), @@ -740,6 +838,7 @@ mod tests { }, adapter, limit, + weight: 1.0, }); }); let manager = ProviderManager::::new( @@ -756,11 +855,23 @@ mod tests { ProviderCheckStrategy::MarkAsValid, ); - let no_retest_adapters = - EthereumNetworkAdapters::new(chain_id.clone(), manager.clone(), vec![], Some(0f64)); + let no_retest_adapters = EthereumNetworkAdapters::new( + chain_id.clone(), + manager.clone(), + vec![], + Some(0f64), + false, + vec![], + ); - let always_retest_adapters = - EthereumNetworkAdapters::new(chain_id, manager.clone(), vec![], Some(1f64)); + let always_retest_adapters = EthereumNetworkAdapters::new( + chain_id, + manager.clone(), + vec![], + Some(1f64), + false, + vec![], + ); assert_eq!( no_retest_adapters @@ -816,6 +927,7 @@ mod tests { adapter: fake_adapter(&logger, &error_provider, &provider_metrics, &metrics, false) .await, limit: SubgraphLimit::Unlimited, + weight: 1.0, }); let mut always_retest_adapters = vec![]; @@ -834,6 +946,7 @@ mod tests { ) .await, limit: SubgraphLimit::Unlimited, + weight: 1.0, }); let manager = ProviderManager::::new( logger.clone(), @@ -844,8 +957,14 @@ mod tests { ProviderCheckStrategy::MarkAsValid, ); - let always_retest_adapters = - EthereumNetworkAdapters::new(chain_id.clone(), manager.clone(), vec![], Some(1f64)); + let always_retest_adapters = EthereumNetworkAdapters::new( + chain_id.clone(), + manager.clone(), + vec![], + Some(1f64), + false, + vec![], + ); assert_eq!( always_retest_adapters @@ -868,8 +987,14 @@ mod tests { ProviderCheckStrategy::MarkAsValid, ); - let no_retest_adapters = - EthereumNetworkAdapters::new(chain_id.clone(), manager, vec![], Some(0f64)); + let no_retest_adapters = EthereumNetworkAdapters::new( + chain_id.clone(), + manager, + vec![], + Some(0f64), + false, + vec![], + ); assert_eq!( no_retest_adapters .cheapest_with(&NodeCapabilities { @@ -898,6 +1023,7 @@ mod tests { ) .await, limit: SubgraphLimit::Disabled, + weight: 1.0, }); let manager = ProviderManager::new( logger, @@ -909,7 +1035,8 @@ mod tests { ProviderCheckStrategy::MarkAsValid, ); - let no_available_adapter = EthereumNetworkAdapters::new(chain_id, manager, vec![], None); + let no_available_adapter = + EthereumNetworkAdapters::new(chain_id, manager, vec![], None, false, vec![]); let res = no_available_adapter .cheapest_with(&NodeCapabilities { archive: true, @@ -945,4 +1072,99 @@ mod tests { .await, ) } + + #[tokio::test] + async fn test_weighted_adapter_selection() { + let metrics = Arc::new(EndpointMetrics::mock()); + let logger = graph::log::logger(true); + let mock_registry = Arc::new(MetricsRegistry::mock()); + let transport = Transport::new_rpc( + Url::parse("http://127.0.0.1").unwrap(), + HeaderMap::new(), + metrics.clone(), + "", + ); + let provider_metrics = Arc::new(ProviderEthRpcMetrics::new(mock_registry.clone())); + + let adapter1 = Arc::new( + EthereumAdapter::new( + logger.clone(), + "adapter1".to_string(), + transport.clone(), + provider_metrics.clone(), + true, + false, + ) + .await, + ); + + let adapter2 = Arc::new( + EthereumAdapter::new( + logger.clone(), + "adapter2".to_string(), + transport.clone(), + provider_metrics.clone(), + true, + false, + ) + .await, + ); + + let mut adapters = EthereumNetworkAdapters::for_testing( + vec![ + EthereumNetworkAdapter::new( + metrics.cheap_clone(), + NodeCapabilities { + archive: true, + traces: false, + }, + adapter1.clone(), + SubgraphLimit::Unlimited, + 0.2, + ), + EthereumNetworkAdapter::new( + metrics.cheap_clone(), + NodeCapabilities { + archive: true, + traces: false, + }, + adapter2.clone(), + SubgraphLimit::Unlimited, + 0.8, + ), + ], + vec![], + ) + .await; + + let health_checker1 = Arc::new(Health::new(adapter1.clone())); + let health_checker2 = Arc::new(Health::new(adapter2.clone())); + + adapters.health_checkers = vec![health_checker1.clone(), health_checker2.clone()]; + adapters.weighted = true; + + let mut adapter1_count = 0; + let mut adapter2_count = 0; + + for _ in 0..1000 { + let selected_adapter = adapters + .cheapest_with(&NodeCapabilities { + archive: true, + traces: false, + }) + .await + .unwrap(); + + if selected_adapter.provider() == "adapter1" { + adapter1_count += 1; + } else { + adapter2_count += 1; + } + } + + // Check that the selection is roughly proportional to the weights. + // Allow for a 10% tolerance. + assert!(adapter1_count > 100 && adapter1_count < 300); + assert!(adapter2_count > 700 && adapter2_count < 900); + } } diff --git a/node/Cargo.toml b/node/Cargo.toml index 9885f6d800b..6a52e23caae 100644 --- a/node/Cargo.toml +++ b/node/Cargo.toml @@ -45,6 +45,7 @@ prometheus = { version = "0.14.0", features = ["push"] } json-structural-diff = { version = "0.2", features = ["colorize"] } globset = "0.4.16" notify = "8.0.0" +tokio = { version = "1.47.1", features = ["full"] } [target.'cfg(unix)'.dependencies] pgtemp = { git = "https://github.com/graphprotocol/pgtemp", branch = "initdb-args" } diff --git a/node/resources/tests/full_config.toml b/node/resources/tests/full_config.toml index 1f907539194..b6481e34be3 100644 --- a/node/resources/tests/full_config.toml +++ b/node/resources/tests/full_config.toml @@ -1,3 +1,5 @@ +weighted_rpc_steering = true + [general] query = "query_node_.*" @@ -43,29 +45,36 @@ indexers = [ "index_node_1_a", [chains] ingestor = "index_0" +# Provider weights configuration: +# - Weights must be between 0.0 and 1.0 (inclusive) +# - Weights are relative - they don't need to sum to 1.0 +# - Traffic is distributed proportionally based on weights +# - Example: weights [0.2, 0.8] = 20% and 80% traffic distribution +# - Example: weights [1.0, 2.0, 1.0] = 25%, 50%, 25% distribution +# - At least one provider must have weight > 0.0 [chains.mainnet] shard = "primary" provider = [ - { label = "mainnet-0", url = "http://rpc.mainnet.io", features = ["archive", "traces"] }, - { label = "mainnet-1", details = { type = "web3call", url = "http://rpc.mainnet.io", features = ["archive", "traces"] }}, - { label = "firehose", details = { type = "firehose", url = "http://localhost:9000", features = [] }}, - { label = "substreams", details = { type = "substreams", url = "http://localhost:9000", features = [] }}, + { label = "mainnet-0", url = "http://rpc.mainnet.io", features = ["archive", "traces"], weight = 0.1 }, + { label = "mainnet-1", details = { type = "web3call", url = "http://rpc.mainnet.io", features = ["archive", "traces"] }, weight = 0.2 }, + { label = "firehose", details = { type = "firehose", url = "http://localhost:9000", features = [] }, weight = 0.3 }, + { label = "substreams", details = { type = "substreams", url = "http://localhost:9000", features = [] }, weight = 0.4 }, ] [chains.ropsten] shard = "primary" provider = [ - { label = "ropsten-0", url = "http://rpc.ropsten.io", transport = "rpc", features = ["archive", "traces"] } + { label = "ropsten-0", url = "http://rpc.ropsten.io", transport = "rpc", features = ["archive", "traces"], weight = 1.0 } ] [chains.goerli] shard = "primary" provider = [ - { label = "goerli-0", url = "http://rpc.goerli.io", transport = "ipc", features = ["archive"] } + { label = "goerli-0", url = "http://rpc.goerli.io", transport = "ipc", features = ["archive"], weight = 1.0 } ] [chains.kovan] shard = "primary" provider = [ - { label = "kovan-0", url = "http://rpc.kovan.io", transport = "ws", features = [] } + { label = "kovan-0", url = "http://rpc.kovan.io", transport = "ws", features = [], weight = 1.0 } ] diff --git a/node/src/chain.rs b/node/src/chain.rs index 343b783908f..7c706e10330 100644 --- a/node/src/chain.rs +++ b/node/src/chain.rs @@ -314,6 +314,7 @@ pub async fn create_ethereum_networks_for_chain( .await, ), web3.limit_for(&config.node), + provider.weight, ); if call_only { @@ -572,6 +573,7 @@ mod test { ethereum_ws: vec![], ethereum_ipc: vec![], unsafe_config: false, + weighted_rpc_steering: false, }; let metrics = Arc::new(EndpointMetrics::mock()); diff --git a/node/src/config.rs b/node/src/config.rs index 83ea7bf1cc3..c6abd3affff 100644 --- a/node/src/config.rs +++ b/node/src/config.rs @@ -48,6 +48,7 @@ pub struct Opt { pub ethereum_ws: Vec, pub ethereum_ipc: Vec, pub unsafe_config: bool, + pub weighted_rpc_steering: bool, } impl Default for Opt { @@ -64,6 +65,7 @@ impl Default for Opt { ethereum_ws: vec![], ethereum_ipc: vec![], unsafe_config: false, + weighted_rpc_steering: false, } } } @@ -73,6 +75,8 @@ pub struct Config { #[serde(skip, default = "default_node_id")] pub node: NodeId, pub general: Option, + #[serde(default)] + pub weighted_rpc_steering: bool, #[serde(rename = "store")] pub stores: BTreeMap, pub chains: ChainSection, @@ -196,6 +200,7 @@ impl Config { Ok(Config { node, general: None, + weighted_rpc_steering: opt.weighted_rpc_steering, stores, chains, deployment, @@ -503,6 +508,7 @@ impl ChainSection { headers: Default::default(), rules: vec![], }), + weight: 1.0, }; let entry = chains.entry(name.to_string()).or_insert_with(|| Chain { shard: PRIMARY_SHARD.to_string(), @@ -544,6 +550,16 @@ impl Chain { return Err(anyhow!("Provider labels must be unique")); } + // Check that not all provider weights are zero + if !self.providers.is_empty() { + let all_zero_weights = self.providers.iter().all(|p| p.weight == 0.0); + if all_zero_weights { + return Err(anyhow!( + "All provider weights are 0.0; at least one provider must have a weight > 0.0" + )); + } + } + // `Config` validates that `self.shard` references a configured shard for provider in self.providers.iter_mut() { provider.validate()? @@ -602,6 +618,7 @@ fn btree_map_to_http_headers(kvs: BTreeMap) -> HeaderMap { pub struct Provider { pub label: String, pub details: ProviderDetails, + pub weight: f64, } #[derive(Clone, Debug, Deserialize, Serialize, PartialEq)] @@ -726,6 +743,9 @@ const DEFAULT_PROVIDER_FEATURES: [&str; 2] = ["traces", "archive"]; impl Provider { fn validate(&mut self) -> Result<()> { validate_name(&self.label).context("illegal provider name")?; + if self.weight < 0.0 || self.weight > 1.0 { + bail!("provider {} must have a weight between 0 and 1", self.label); + } match self.details { ProviderDetails::Firehose(ref mut firehose) @@ -820,6 +840,7 @@ impl<'de> Deserialize<'de> for Provider { { let mut label = None; let mut details = None; + let mut weight = None; let mut url = None; let mut transport = None; @@ -841,6 +862,12 @@ impl<'de> Deserialize<'de> for Provider { } details = Some(map.next_value()?); } + ProviderField::Weight => { + if weight.is_some() { + return Err(serde::de::Error::duplicate_field("weight")); + } + weight = Some(map.next_value()?); + } ProviderField::Url => { if url.is_some() { return Err(serde::de::Error::duplicate_field("url")); @@ -904,13 +931,18 @@ impl<'de> Deserialize<'de> for Provider { }), }; - Ok(Provider { label, details }) + Ok(Provider { + label, + details, + weight: weight.unwrap_or(1.0), + }) } } const FIELDS: &[&str] = &[ "label", "details", + "weight", "transport", "url", "features", @@ -925,6 +957,7 @@ impl<'de> Deserialize<'de> for Provider { enum ProviderField { Label, Details, + Weight, Match, // Deprecated fields @@ -1308,6 +1341,7 @@ mod tests { headers: HeaderMap::new(), rules: Vec::new(), }), + weight: 1.0, }, actual ); @@ -1334,6 +1368,7 @@ mod tests { headers: HeaderMap::new(), rules: Vec::new(), }), + weight: 1.0, }, actual ); @@ -1441,6 +1476,7 @@ mod tests { headers, rules: Vec::new(), }), + weight: 1.0, }, actual ); @@ -1466,6 +1502,7 @@ mod tests { headers: HeaderMap::new(), rules: Vec::new(), }), + weight: 1.0, }, actual ); @@ -1507,6 +1544,7 @@ mod tests { conn_pool_size: 20, rules: vec![], }), + weight: 1.0, }, actual ); @@ -1533,6 +1571,7 @@ mod tests { conn_pool_size: 20, rules: vec![], }), + weight: 1.0, }, actual ); @@ -1559,6 +1598,7 @@ mod tests { conn_pool_size: 20, rules: vec![], }), + weight: 1.0, }, actual ); @@ -1585,6 +1625,7 @@ mod tests { conn_pool_size: 20, rules: vec![], }), + weight: 1.0, }, actual ); @@ -1624,6 +1665,7 @@ mod tests { } ], }), + weight: 1.0, }, actual ); @@ -1663,6 +1705,7 @@ mod tests { } ], }), + weight: 1.0, }, actual ); @@ -1702,6 +1745,7 @@ mod tests { } ], }), + weight: 1.0, }, actual ); @@ -1741,6 +1785,7 @@ mod tests { } ], }), + weight: 1.0, }, actual ); @@ -1835,6 +1880,7 @@ mod tests { headers: HeaderMap::new(), rules: Vec::new(), }), + weight: 1.0, }, actual ); diff --git a/node/src/network_setup.rs b/node/src/network_setup.rs index d086c786f82..d1a35fa43bc 100644 --- a/node/src/network_setup.rs +++ b/node/src/network_setup.rs @@ -103,11 +103,15 @@ impl AdapterConfiguration { } } +use graph_chain_ethereum::health::{health_check_task, Health}; + pub struct Networks { pub adapters: Vec, pub rpc_provider_manager: ProviderManager, pub firehose_provider_manager: ProviderManager>, pub substreams_provider_manager: ProviderManager>, + pub weighted_rpc_steering: bool, + pub health_checkers: Vec>, } impl Networks { @@ -130,6 +134,8 @@ impl Networks { vec![].into_iter(), ProviderCheckStrategy::MarkAsValid, ), + weighted_rpc_steering: false, + health_checkers: vec![], } } @@ -221,7 +227,12 @@ impl Networks { .chain(substreams.into_iter()) .collect(); - Ok(Networks::new(&logger, adapters, provider_checks)) + Ok(Networks::new( + &logger, + adapters, + provider_checks, + config.weighted_rpc_steering, + )) } pub async fn from_config_for_chain( @@ -266,6 +277,7 @@ impl Networks { logger: &Logger, adapters: Vec, provider_checks: &[Arc], + weighted_rpc_steering: bool, ) -> Self { let adapters2 = adapters.clone(); let eth_adapters = adapters.iter().flat_map(|a| a.as_rpc()).cloned().map( @@ -285,6 +297,15 @@ impl Networks { }, ); + let health_checkers: Vec<_> = eth_adapters + .clone() + .flat_map(|(_, adapters)| adapters) + .map(|adapter| Arc::new(Health::new(adapter.adapter.clone()))) + .collect(); + if weighted_rpc_steering { + tokio::spawn(health_check_task(health_checkers.clone())); + } + let firehose_adapters = adapters .iter() .flat_map(|a| a.as_firehose()) @@ -332,6 +353,8 @@ impl Networks { .map(|(chain_id, endpoints)| (chain_id, endpoints)), ProviderCheckStrategy::RequireAll(provider_checks), ), + weighted_rpc_steering, + health_checkers, }; s @@ -445,6 +468,8 @@ impl Networks { self.rpc_provider_manager.clone(), eth_adapters, None, + self.weighted_rpc_steering, + self.health_checkers.clone(), ) } } diff --git a/node/src/opt.rs b/node/src/opt.rs index 9928144396a..f87df56d486 100644 --- a/node/src/opt.rs +++ b/node/src/opt.rs @@ -102,6 +102,12 @@ pub struct Opt { help= "Ethereum network name (e.g. 'mainnet'), optional comma-seperated capabilities (eg 'full,archive'), and an Ethereum IPC pipe, separated by a ':'", )] pub ethereum_ipc: Vec, + #[clap( + long, + env = "GRAPH_WEIGHTED_RPC_STEERING", + help = "Enable weighted random steering for Ethereum RPCs" + )] + pub weighted_rpc_steering: bool, #[clap( long, value_name = "HOST:PORT", @@ -245,6 +251,7 @@ impl From for config::Opt { ethereum_rpc, ethereum_ws, ethereum_ipc, + weighted_rpc_steering, unsafe_config, .. } = opt; @@ -260,6 +267,7 @@ impl From for config::Opt { ethereum_rpc, ethereum_ws, ethereum_ipc, + weighted_rpc_steering, unsafe_config, } }