chore(guard): configure local timeout knobs

Author: NathanFlurry

engine/artifacts/config-schema.json

@@ -14,6 +14,14 @@ pub struct Guard {
 	/// Enables the internal websocket health route for debug and latency testing. This is intended
 	/// for websocket ping/pong verification and should remain disabled in normal deployments.
 	pub enable_websocket_health_route: Option<bool>,
+	/// TTL for cached route lookups in milliseconds.
+	pub route_cache_ttl_ms: Option<u64>,
+	/// Timeout for route resolution in milliseconds.
+	pub route_timeout_ms: Option<u64>,
+	/// Timeout for waiting for an actor to become ready in milliseconds.
+	pub actor_ready_timeout_ms: Option<u64>,
+	/// Timeout sent with actor force-wake requests in milliseconds.
+	pub actor_force_wake_pending_timeout_ms: Option<i64>,
 	/// Enable & configure HTTPS
 	pub https: Option<Https>,
 	/// Max HTTP request body size in bytes (first line of defense).
@@ -37,6 +45,23 @@ impl Guard {
 		self.enable_websocket_health_route.unwrap_or(false)
 	}
 
+	pub fn route_cache_ttl(&self) -> std::time::Duration {
+		std::time::Duration::from_millis(self.route_cache_ttl_ms.unwrap_or(60 * 10 * 1000))
+	}
+
+	pub fn route_timeout(&self) -> std::time::Duration {
+		std::time::Duration::from_millis(self.route_timeout_ms.unwrap_or(15_000))
+	}
+
+	pub fn actor_ready_timeout(&self) -> std::time::Duration {
+		std::time::Duration::from_millis(self.actor_ready_timeout_ms.unwrap_or(10_000))
+	}
+
+	pub fn actor_force_wake_pending_timeout(&self) -> i64 {
+		self.actor_force_wake_pending_timeout_ms
+			.unwrap_or(60 * 1000)
+	}
+
 	pub fn http_max_request_body_size(&self) -> usize {
 		self.http_max_request_body_size.unwrap_or(20 * 1024 * 1024) // 20 MiB
 	}

engine/packages/config/src/config/guard.rs

@@ -31,7 +31,7 @@ use crate::RouteTarget;
 use crate::request_context::RequestContext;
 use crate::response_body::ResponseBody;
 use crate::route::{
-	CacheKeyFn, DEFAULT_ROUTE_TIMEOUT, ResolveRouteOutput, RouteCache, RoutingFn, RoutingOutput,
+	CacheKeyFn, ResolveRouteOutput, RouteCache, RoutingFn, RoutingOutput,
 };
 use crate::utils::{InFlightCounter, RateLimiter};
 use crate::{
@@ -89,13 +89,14 @@ impl ProxyState {
 		let client = Client::builder(TokioExecutor::new())
 			.pool_idle_timeout(Duration::from_secs(30))
 			.build(https_connector);
+		let route_cache_ttl = config.guard().route_cache_ttl();
 
 		Self {
 			config,
 			routing_fn,
 			cache_key_fn,
 			client,
-			route_cache: RouteCache::new(),
+			route_cache: RouteCache::new(route_cache_ttl),
 			rate_limiters: Cache::builder()
 				.max_capacity(10_000)
 				.time_to_live(PROXY_STATE_CACHE_TTL)
@@ -134,21 +135,22 @@ impl ProxyState {
 		let res = if let Some(res) = cache_res {
 			res
 		} else {
-			// Not in cache, call routing function with a default timeout
-			// Default 15 seconds, routing functions should have their own internal timeouts that are shorter
+			// Not in cache, call routing function with a configured timeout.
+			// Routing functions should have their own internal timeouts that are shorter.
+			let route_timeout = self.config.guard().route_timeout();
 			tracing::debug!(
 				hostname = %req_ctx.hostname,
 				path = %req_ctx.path,
 				cache_hit = false,
-				timeout_seconds = DEFAULT_ROUTE_TIMEOUT.as_secs(),
+				timeout_seconds = route_timeout.as_secs(),
 				"Cache miss, calling routing function"
 			);
 
-			let routing_res = timeout(DEFAULT_ROUTE_TIMEOUT, (self.routing_fn)(req_ctx))
+			let routing_res = timeout(route_timeout, (self.routing_fn)(req_ctx))
 				.await
 				.map_err(|_| {
 					errors::RequestTimeout {
-						timeout_seconds: DEFAULT_ROUTE_TIMEOUT.as_secs(),
+						timeout_seconds: route_timeout.as_secs(),
 					}
 					.build()
 				})??;

engine/packages/guard-core/src/proxy_service.rs

@@ -7,9 +7,6 @@ use crate::custom_serve::CustomServeTrait;
 use crate::metrics;
 use crate::request_context::RequestContext;
 
-const ROUTE_CACHE_TTL: Duration = Duration::from_secs(60 * 10); // 10 minutes
-pub(crate) const DEFAULT_ROUTE_TIMEOUT: Duration = Duration::from_secs(15);
-
 // Routing types
 #[derive(Clone, Debug)]
 pub struct RouteTarget {
@@ -51,11 +48,11 @@ pub(crate) struct RouteCache {
 }
 
 impl RouteCache {
-	pub(crate) fn new() -> Self {
+	pub(crate) fn new(ttl: Duration) -> Self {
 		Self {
 			cache: Cache::builder()
 				.max_capacity(10_000)
-				.time_to_live(ROUTE_CACHE_TTL)
+				.time_to_live(ttl)
 				.build(),
 		}
 	}

engine/packages/guard-core/src/route.rs

@@ -23,9 +23,6 @@ use crate::{
 use cors::{CorsPreflight, set_non_preflight_cors};
 use resolve_actor_query::resolve_query;
 
-const ACTOR_FORCE_WAKE_PENDING_TIMEOUT: i64 = util::duration::seconds(60);
-const ACTOR_READY_TIMEOUT: Duration = Duration::from_secs(10);
-
 /// Time to wait before starting pool error checks
 const RUNNER_POOL_ERROR_CHECK_DELAY: Duration = Duration::from_secs(1);
 /// Interval between pool error checks
@@ -436,7 +433,7 @@ async fn handle_actor_v2(
 					}
 				}
 				// Ready timeout
-				_ = tokio::time::sleep(ACTOR_READY_TIMEOUT) => {
+				_ = tokio::time::sleep(ctx.config().guard().actor_ready_timeout()) => {
 					return Err(errors::ActorReadyTimeout { actor_id }.build());
 				}
 			}
@@ -480,7 +477,7 @@ async fn handle_actor_v1(
 
 		ctx.signal(pegboard::workflows::actor::Wake {
 			allocation_override: pegboard::workflows::actor::AllocationOverride::DontSleep {
-				pending_timeout: Some(ACTOR_FORCE_WAKE_PENDING_TIMEOUT),
+				pending_timeout: Some(ctx.config().guard().actor_force_wake_pending_timeout()),
 			},
 		})
 		.to_workflow_id(actor.workflow_id)
@@ -516,7 +513,9 @@ async fn handle_actor_v1(
 
 						let res = ctx.signal(pegboard::workflows::actor::Wake {
 							allocation_override: pegboard::workflows::actor::AllocationOverride::DontSleep {
-								pending_timeout: Some(ACTOR_FORCE_WAKE_PENDING_TIMEOUT),
+								pending_timeout: Some(
+									ctx.config().guard().actor_force_wake_pending_timeout(),
+								),
 							},
 						})
 						.to_workflow_id(actor.workflow_id)
@@ -565,7 +564,7 @@ async fn handle_actor_v1(
 					}
 				}
 				// Ready timeout
-				_ = tokio::time::sleep(ACTOR_READY_TIMEOUT) => {
+				_ = tokio::time::sleep(ctx.config().guard().actor_ready_timeout()) => {
 					return Err(errors::ActorReadyTimeout { actor_id }.build());
 				}
 			}