Address tokio contention messages

core/src/subgraph/context/mod.rs

@@ -23,7 +23,9 @@ use graph::{
     slog::Logger,
 };
 use std::collections::HashMap;
-use std::sync::{Arc, RwLock};
+use std::sync::Arc;
+
+use graph::parking_lot::RwLock;
 use tokio::sync::mpsc;
 
 use self::instance::SubgraphInstance;
@@ -44,18 +46,18 @@ impl SubgraphKeepAlive {
     }
 
     pub fn remove(&self, deployment_id: &DeploymentId) {
-        self.alive_map.write().unwrap().remove(deployment_id);
+        self.alive_map.write().remove(deployment_id);
         self.sg_metrics.running_count.dec();
     }
     pub fn insert(&self, deployment_id: DeploymentId, guard: CancelGuard) {
-        let old = self.alive_map.write().unwrap().insert(deployment_id, guard);
+        let old = self.alive_map.write().insert(deployment_id, guard);
         if old.is_none() {
             self.sg_metrics.running_count.inc();
         }
     }
 
     pub fn contains(&self, deployment_id: &DeploymentId) -> bool {
-        self.alive_map.read().unwrap().contains_key(deployment_id)
+        self.alive_map.read().contains_key(deployment_id)
     }
 }
 

graph/Cargo.toml

@@ -79,6 +79,7 @@ futures03 = { version = "0.3.31", package = "futures", features = ["compat"] }
 wasmparser = "0.118.1"
 thiserror = { workspace = true }
 parking_lot = "0.12.5"
+portable-atomic = { version = "1.11", features = ["fallback"] }
 itertools = "0.14.0"
 defer = "0.2"
 

graph/src/components/metrics/registry.rs

@@ -1,5 +1,7 @@
 use std::collections::HashMap;
-use std::sync::{Arc, RwLock};
+use std::sync::Arc;
+
+use crate::parking_lot::RwLock;
 
 use prometheus::{labels, Histogram, IntCounterVec};
 use prometheus::{IntCounter, IntGauge};
@@ -109,14 +111,13 @@ impl MetricsRegistry {
         };
         let counters = CounterVec::new(opts, variable_labels)?;
         let id = counters.desc().first().unwrap().id;
-        let maybe_counter = self.global_counter_vecs.read().unwrap().get(&id).cloned();
+        let maybe_counter = self.global_counter_vecs.read().get(&id).cloned();
         if let Some(counters) = maybe_counter {
             Ok(counters)
         } else {
             self.register(name, Box::new(counters.clone()));
             self.global_counter_vecs
                 .write()
-                .unwrap()
                 .insert(id, counters.clone());
             Ok(counters)
         }
@@ -161,15 +162,12 @@ impl MetricsRegistry {
     ) -> Result<Counter, PrometheusError> {
         let counter = counter_with_labels(name, help, const_labels)?;
         let id = counter.desc().first().unwrap().id;
-        let maybe_counter = self.global_counters.read().unwrap().get(&id).cloned();
+        let maybe_counter = self.global_counters.read().get(&id).cloned();
         if let Some(counter) = maybe_counter {
             Ok(counter)
         } else {
             self.register(name, Box::new(counter.clone()));
-            self.global_counters
-                .write()
-                .unwrap()
-                .insert(id, counter.clone());
+            self.global_counters.write().insert(id, counter.clone());
             Ok(counter)
         }
     }
@@ -210,15 +208,12 @@ impl MetricsRegistry {
     ) -> Result<Gauge, PrometheusError> {
         let gauge = gauge_with_labels(name, help, const_labels)?;
         let id = gauge.desc().first().unwrap().id;
-        let maybe_gauge = self.global_gauges.read().unwrap().get(&id).cloned();
+        let maybe_gauge = self.global_gauges.read().get(&id).cloned();
         if let Some(gauge) = maybe_gauge {
             Ok(gauge)
         } else {
             self.register(name, Box::new(gauge.clone()));
-            self.global_gauges
-                .write()
-                .unwrap()
-                .insert(id, gauge.clone());
+            self.global_gauges.write().insert(id, gauge.clone());
             Ok(gauge)
         }
     }
@@ -232,15 +227,12 @@ impl MetricsRegistry {
         let opts = Opts::new(name, help);
         let gauges = GaugeVec::new(opts, variable_labels)?;
         let id = gauges.desc().first().unwrap().id;
-        let maybe_gauge = self.global_gauge_vecs.read().unwrap().get(&id).cloned();
+        let maybe_gauge = self.global_gauge_vecs.read().get(&id).cloned();
         if let Some(gauges) = maybe_gauge {
             Ok(gauges)
         } else {
             self.register(name, Box::new(gauges.clone()));
-            self.global_gauge_vecs
-                .write()
-                .unwrap()
-                .insert(id, gauges.clone());
+            self.global_gauge_vecs.write().insert(id, gauges.clone());
             Ok(gauges)
         }
     }
@@ -254,14 +246,13 @@ impl MetricsRegistry {
         let opts = HistogramOpts::new(name, help);
         let histograms = HistogramVec::new(opts, variable_labels)?;
         let id = histograms.desc().first().unwrap().id;
-        let maybe_histogram = self.global_histogram_vecs.read().unwrap().get(&id).cloned();
+        let maybe_histogram = self.global_histogram_vecs.read().get(&id).cloned();
         if let Some(histograms) = maybe_histogram {
             Ok(histograms)
         } else {
             self.register(name, Box::new(histograms.clone()));
             self.global_histogram_vecs
                 .write()
-                .unwrap()
                 .insert(id, histograms.clone());
             Ok(histograms)
         }

graph/src/components/store/mod.rs

@@ -25,7 +25,7 @@ use std::collections::{BTreeMap, BTreeSet, HashSet};
 use std::fmt;
 use std::fmt::Display;
 use std::sync::atomic::{AtomicUsize, Ordering};
-use std::sync::{Arc, RwLock};
+use std::sync::Arc;
 use std::time::Duration;
 
 use async_trait::async_trait;
@@ -42,7 +42,7 @@ use crate::env::ENV_VARS;
 use crate::internal_error;
 use crate::prelude::{s, Attribute, DeploymentHash, ValueType};
 use crate::schema::{ast as sast, EntityKey, EntityType, InputSchema};
-use crate::util::stats::MovingStats;
+use crate::util::stats::AtomicMovingStats;
 
 #[derive(Clone, Debug, Serialize, Deserialize, PartialEq, Eq, PartialOrd, Ord, Hash)]
 pub struct EntityFilterDerivative(bool);
@@ -742,8 +742,8 @@ impl Display for DeploymentLocator {
 }
 
 // The type that the connection pool uses to track wait times for
-// connection checkouts
-pub type PoolWaitStats = Arc<RwLock<MovingStats>>;
+// connection checkouts. Uses lock-free atomic operations internally.
+pub type PoolWaitStats = Arc<AtomicMovingStats>;
 
 /// Determines which columns should be selected in a table.
 #[derive(Clone, Debug, PartialEq, Eq, Hash)]

graph/src/data/graphql/load_manager.rs

@@ -4,9 +4,11 @@ use prometheus::core::GenericCounter;
 use rand::{prelude::Rng, rng};
 use std::collections::{HashMap, HashSet};
 use std::iter::FromIterator;
-use std::sync::{Arc, RwLock};
+use std::sync::Arc;
 use std::time::{Duration, Instant};
 
+use crate::parking_lot::RwLock;
+
 use crate::components::metrics::{Counter, GaugeVec, MetricsRegistry};
 use crate::components::store::{DeploymentId, PoolWaitStats};
 use crate::data::graphql::shape_hash::shape_hash;
@@ -57,7 +59,7 @@ impl ShardEffort {
     }
 
     pub fn add(&self, shard: &str, qref: QueryRef, duration: Duration, gauge: &GaugeVec) {
-        let mut inner = self.inner.write().unwrap();
+        let mut inner = self.inner.write();
         inner.add(qref, duration);
         gauge
             .with_label_values(&[shard])
@@ -70,7 +72,7 @@ impl ShardEffort {
     /// data for the particular query, return `None` as the effort
     /// for the query
     pub fn current_effort(&self, qref: &QueryRef) -> (Option<Duration>, Duration) {
-        let inner = self.inner.read().unwrap();
+        let inner = self.inner.read();
         let total_effort = inner.total.duration();
         let query_effort = inner.effort.get(qref).map(|stats| stats.duration());
         (query_effort, total_effort)
@@ -381,7 +383,7 @@ impl LoadManager {
 
         let qref = QueryRef::new(deployment, shape_hash);
 
-        if self.jailed_queries.read().unwrap().contains(&qref) {
+        if self.jailed_queries.read().contains(&qref) {
             return if ENV_VARS.load_simulate {
                 Proceed
             } else {
@@ -426,7 +428,7 @@ impl LoadManager {
                 "query_effort_ms" => query_effort,
                 "total_effort_ms" => total_effort,
                 "ratio" => format!("{:.4}", query_effort/total_effort));
-                self.jailed_queries.write().unwrap().insert(qref);
+                self.jailed_queries.write().insert(qref);
                 return if ENV_VARS.load_simulate {
                     Proceed
                 } else {
@@ -457,15 +459,15 @@ impl LoadManager {
     }
 
     fn overloaded(&self, wait_stats: &PoolWaitStats) -> (bool, Duration) {
-        let store_avg = wait_stats.read().unwrap().average();
+        let store_avg = wait_stats.average();
         let overloaded = store_avg
             .map(|average| average > ENV_VARS.load_threshold)
             .unwrap_or(false);
         (overloaded, store_avg.unwrap_or(Duration::ZERO))
     }
 
     fn kill_state(&self, shard: &str) -> (f64, Instant) {
-        let state = self.kill_state.get(shard).unwrap().read().unwrap();
+        let state = self.kill_state.get(shard).unwrap().read();
         (state.kill_rate, state.last_update)
     }
 
@@ -505,7 +507,7 @@ impl LoadManager {
                 kill_rate = (kill_rate - KILL_RATE_STEP_DOWN).max(0.0);
             }
             let event = {
-                let mut state = self.kill_state.get(shard).unwrap().write().unwrap();
+                let mut state = self.kill_state.get(shard).unwrap().write();
                 state.kill_rate = kill_rate;
                 state.last_update = now;
                 state.log_event(now, kill_rate, overloaded)

graph/src/env/store.rs

@@ -163,6 +163,14 @@ pub struct EnvVarsStore {
     /// Disables storing or reading `eth_call` results from the store call cache.
     /// Set by `GRAPH_STORE_DISABLE_CALL_CACHE`. Defaults to false.
     pub disable_call_cache: bool,
+    /// Set by `GRAPH_STORE_DISABLE_CHAIN_HEAD_PTR_CACHE`. Default is false.
+    /// Set to true to disable chain_head_ptr caching (safety escape hatch).
+    pub disable_chain_head_ptr_cache: bool,
+    /// Minimum idle time before running connection health check (SELECT 67).
+    /// Connections used more recently than this threshold skip validation.
+    /// Set to 0 to always validate (previous behavior).
+    /// Set by `GRAPH_STORE_CONNECTION_VALIDATION_IDLE_SECS`. Default is 30 seconds.
+    pub connection_validation_idle_secs: Duration,
 }
 
 // This does not print any values avoid accidentally leaking any sensitive env vars
@@ -224,6 +232,8 @@ impl TryFrom<InnerStore> for EnvVarsStore {
             account_like_min_versions_count: x.account_like_min_versions_count,
             account_like_max_unique_ratio: x.account_like_max_unique_ratio.map(|r| r.0),
             disable_call_cache: x.disable_call_cache,
+            disable_chain_head_ptr_cache: x.disable_chain_head_ptr_cache,
+            connection_validation_idle_secs: Duration::from_secs(x.connection_validation_idle_secs),
         };
         if let Some(timeout) = vars.batch_timeout {
             if timeout < 2 * vars.batch_target_duration {
@@ -331,6 +341,10 @@ pub struct InnerStore {
     account_like_max_unique_ratio: Option<ZeroToOneF64>,
     #[envconfig(from = "GRAPH_STORE_DISABLE_CALL_CACHE", default = "false")]
     disable_call_cache: bool,
+    #[envconfig(from = "GRAPH_STORE_DISABLE_CHAIN_HEAD_PTR_CACHE", default = "false")]
+    disable_chain_head_ptr_cache: bool,
+    #[envconfig(from = "GRAPH_STORE_CONNECTION_VALIDATION_IDLE_SECS", default = "30")]
+    connection_validation_idle_secs: u64,
 }
 
 #[derive(Clone, Copy, Debug)]

graph/src/lib.rs

@@ -170,7 +170,7 @@ pub mod prelude {
     pub use crate::log::split::split_logger;
     pub use crate::util::cache_weight::CacheWeight;
     pub use crate::util::futures::{retry, TimeoutError};
-    pub use crate::util::stats::MovingStats;
+    pub use crate::util::stats::{AtomicMovingStats, MovingStats};
 
     macro_rules! static_graphql {
         ($m:ident, $m2:ident, {$($n:ident,)*}) => {