@rivetkit/engine 1.0.0 → 2.2.1-pr.4600.b74ff3b

package/packages/engine/tests/envoy/actors_lifecycle.rs

@@ -0,0 +1,1277 @@
+use std::sync::{Arc, Mutex};
+
+use super::super::common;
+
+// MARK: Creation and Initialization
+#[test]
+fn envoy_actor_basic_create() {
+	common::run(common::TestOpts::new(1), |ctx| async move {
+		let (namespace, _) = common::setup_test_namespace(ctx.leader_dc()).await;
+
+		let (start_tx, start_rx) = tokio::sync::oneshot::channel();
+		let start_tx = Arc::new(Mutex::new(Some(start_tx)));
+
+		let envoy = common::setup_envoy(ctx.leader_dc(), &namespace, |builder| {
+			builder.with_actor_behavior("test-actor", move |_| {
+				Box::new(common::test_envoy::NotifyOnStartActor::new(
+					start_tx.clone(),
+				))
+			})
+		})
+		.await;
+
+		let res = common::create_actor(
+			ctx.leader_dc().guard_port(),
+			&namespace,
+			"test-actor",
+			envoy.pool_name(),
+			rivet_types::actors::CrashPolicy::Destroy,
+		)
+		.await;
+
+		let actor_id = res.actor.actor_id.to_string();
+
+		// Wait for actor to start (notification from actor)
+		start_rx
+			.await
+			.expect("actor should have sent start notification");
+
+		// Verify actor is allocated to envoy
+		assert!(
+			envoy.has_actor(&actor_id).await,
+			"envoy should have the actor allocated"
+		);
+
+		tracing::info!(?actor_id, envoy_key = ?envoy.envoy_key, "actor allocated to envoy");
+	});
+}
+
+#[test]
+fn envoy_create_actor_with_input() {
+	common::run(common::TestOpts::new(1), |ctx| async move {
+		let (namespace, _) = common::setup_test_namespace(ctx.leader_dc()).await;
+
+		// Generate test input data (base64-encoded String)
+		let input_data = common::generate_test_input_data();
+
+		// Decode the base64 data to get the actual bytes the actor will receive
+		// The API automatically decodes base64 input before sending to the envoy
+		let input_data_bytes =
+			base64::Engine::decode(&base64::engine::general_purpose::STANDARD, &input_data)
+				.expect("failed to decode base64 input");
+
+		// Create envoy with VerifyInputActor that will validate the input
+		let envoy = common::setup_envoy(ctx.leader_dc(), &namespace, |builder| {
+			builder.with_actor_behavior("test-actor", move |_| {
+				Box::new(common::test_envoy::VerifyInputActor::new(
+					input_data_bytes.clone(),
+				))
+			})
+		})
+		.await;
+
+		// Create actor with input data
+		let res = common::api::public::actors_create(
+			ctx.leader_dc().guard_port(),
+			common::api_types::actors::create::CreateQuery {
+				namespace: namespace.clone(),
+			},
+			common::api_types::actors::create::CreateRequest {
+				datacenter: None,
+				name: "test-actor".to_string(),
+				key: None,
+				input: Some(input_data.clone()),
+				runner_name_selector: envoy.pool_name().to_string(),
+				crash_policy: rivet_types::actors::CrashPolicy::Destroy,
+			},
+		)
+		.await
+		.expect("failed to create actor");
+
+		let actor_id = res.actor.actor_id.to_string();
+
+		// Poll for actor to become connectable
+		// If input verification fails, the actor will crash and never become connectable
+		let actor = loop {
+			let actor = common::try_get_actor(ctx.leader_dc().guard_port(), &actor_id, &namespace)
+				.await
+				.expect("failed to get actor")
+				.expect("actor should exist");
+
+			// Check if actor crashed (input verification failed)
+			if actor.destroy_ts.is_some() {
+				panic!(
+					"actor crashed during input verification (input data was not received correctly)"
+				);
+			}
+
+			// Check if actor is connectable (input verification succeeded)
+			if actor.connectable_ts.is_some() {
+				break actor;
+			}
+
+			tokio::time::sleep(tokio::time::Duration::from_millis(50)).await;
+		};
+
+		assert!(
+			actor.connectable_ts.is_some(),
+			"actor should be connectable after successful input verification"
+		);
+
+		tracing::info!(
+			?actor_id,
+			input_size = input_data.len(),
+			"actor successfully verified input data"
+		);
+	});
+}
+
+#[test]
+fn envoy_actor_start_timeout() {
+	// This test takes 35+ seconds
+	common::run(
+		common::TestOpts::new(1).with_timeout(45),
+		|ctx| async move {
+			let (namespace, _) = common::setup_test_namespace(ctx.leader_dc()).await;
+
+			// Create envoy client with timeout actor behavior
+			let envoy = common::setup_envoy(ctx.leader_dc(), &namespace, |builder| {
+				builder.with_actor_behavior("timeout-actor", move |_| {
+					Box::new(common::test_envoy::TimeoutActor::new())
+				})
+			})
+			.await;
+
+			tracing::info!("envoy client ready, creating actor that will timeout");
+
+			// Create actor with destroy crash policy
+			let res = common::create_actor(
+				ctx.leader_dc().guard_port(),
+				&namespace,
+				"timeout-actor",
+				envoy.pool_name(),
+				rivet_types::actors::CrashPolicy::Destroy,
+			)
+			.await;
+
+			let actor_id_str = res.actor.actor_id.to_string();
+
+			tracing::info!(?actor_id_str, "actor created, waiting for timeout");
+
+			// Wait for the actor start timeout threshold (30s + buffer)
+			tokio::time::sleep(tokio::time::Duration::from_secs(35)).await;
+
+			// Verify actor was marked as destroyed due to timeout
+			let actor =
+				common::try_get_actor(ctx.leader_dc().guard_port(), &actor_id_str, &namespace)
+					.await
+					.expect("failed to get actor")
+					.expect("actor should exist");
+
+			assert!(
+				actor.destroy_ts.is_some(),
+				"actor should be destroyed after start timeout"
+			);
+
+			tracing::info!(?actor_id_str, "actor correctly destroyed after timeout");
+		},
+	);
+}
+
+// MARK: Running State Management
+#[test]
+fn envoy_actor_starts_and_connectable_via_guard_http() {
+	common::run(common::TestOpts::new(1), |ctx| async move {
+		let (namespace, _) = common::setup_test_namespace(ctx.leader_dc()).await;
+
+		let (start_tx, start_rx) = tokio::sync::oneshot::channel();
+		let start_tx = Arc::new(Mutex::new(Some(start_tx)));
+
+		let envoy = common::setup_envoy(ctx.leader_dc(), &namespace, |builder| {
+			builder.with_actor_behavior("test-actor", move |_| {
+				Box::new(common::test_envoy::NotifyOnStartActor::new(
+					start_tx.clone(),
+				))
+			})
+		})
+		.await;
+
+		let res = common::create_actor(
+			ctx.leader_dc().guard_port(),
+			&namespace,
+			"test-actor",
+			envoy.pool_name(),
+			rivet_types::actors::CrashPolicy::Destroy,
+		)
+		.await;
+
+		let actor_id = res.actor.actor_id.to_string();
+
+		// Wait for actor to start
+		start_rx
+			.await
+			.expect("actor should have sent start notification");
+
+		// Poll for connectable_ts to be set
+		let actor = loop {
+			let actor = common::try_get_actor(ctx.leader_dc().guard_port(), &actor_id, &namespace)
+				.await
+				.expect("failed to get actor")
+				.expect("actor should exist");
+
+			if actor.connectable_ts.is_some() {
+				break actor;
+			}
+
+			tokio::time::sleep(tokio::time::Duration::from_millis(50)).await;
+		};
+
+		assert!(
+			actor.connectable_ts.is_some(),
+			"actor should be connectable"
+		);
+
+		// TODO: HTTP ping test via guard needs to be implemented: the Rust envoy client atm
+		// doesn't implement HTTP tunneling yet. The original test with TypeScript
+		// envoy included: common::ping_actor_via_guard(ctx.leader_dc(), &actor_id).await;
+
+		tracing::info!(?actor_id, "actor is connectable (state verified)");
+	});
+}
+
+#[test]
+fn envoy_actor_connectable_via_guard_websocket() {
+	common::run(common::TestOpts::new(1), |ctx| async move {
+		let (namespace, _) = common::setup_test_namespace(ctx.leader_dc()).await;
+
+		let (start_tx, start_rx) = tokio::sync::oneshot::channel();
+		let start_tx = Arc::new(Mutex::new(Some(start_tx)));
+
+		let envoy = common::setup_envoy(ctx.leader_dc(), &namespace, |builder| {
+			builder.with_actor_behavior("test-actor", move |_| {
+				Box::new(common::test_envoy::NotifyOnStartActor::new(
+					start_tx.clone(),
+				))
+			})
+		})
+		.await;
+
+		let res = common::create_actor(
+			ctx.leader_dc().guard_port(),
+			&namespace,
+			"test-actor",
+			envoy.pool_name(),
+			rivet_types::actors::CrashPolicy::Destroy,
+		)
+		.await;
+
+		let actor_id = res.actor.actor_id.to_string();
+
+		// Wait for actor to start
+		start_rx
+			.await
+			.expect("actor should have sent start notification");
+
+		// Poll for connectable_ts to be set
+		let actor = loop {
+			let actor = common::try_get_actor(ctx.leader_dc().guard_port(), &actor_id, &namespace)
+				.await
+				.expect("failed to get actor")
+				.expect("actor should exist");
+
+			if actor.connectable_ts.is_some() {
+				break actor;
+			}
+
+			tokio::time::sleep(tokio::time::Duration::from_millis(100)).await;
+		};
+
+		assert!(
+			actor.connectable_ts.is_some(),
+			"actor should be connectable"
+		);
+
+		// Note: WebSocket ping test via guard is skipped because the Rust envoy client
+		// doesn't implement HTTP tunneling yet. The original test with TypeScript
+		// envoy included: common::ping_actor_websocket_via_guard(ctx.leader_dc(), &actor_id).await;
+
+		tracing::info!(?actor_id, "actor is connectable (state verified)");
+	});
+}
+
+// MARK: Stopping and Graceful Shutdown
+#[test]
+fn envoy_actor_graceful_stop_with_destroy_policy() {
+	common::run(common::TestOpts::new(1), |ctx| async move {
+		let (namespace, _) = common::setup_test_namespace(ctx.leader_dc()).await;
+
+		// Create envoy client with stop immediately actor
+		let envoy = common::setup_envoy(ctx.leader_dc(), &namespace, |builder| {
+			builder.with_actor_behavior("stop-actor", move |_| {
+				Box::new(common::test_envoy::StopImmediatelyActor::new())
+			})
+		})
+		.await;
+
+		tracing::info!("envoy client ready, creating actor that will stop gracefully");
+
+		// Create actor with destroy crash policy
+		let res = common::create_actor(
+			ctx.leader_dc().guard_port(),
+			&namespace,
+			"stop-actor",
+			envoy.pool_name(),
+			rivet_types::actors::CrashPolicy::Destroy,
+		)
+		.await;
+
+		let actor_id_str = res.actor.actor_id.to_string();
+
+		tracing::info!(?actor_id_str, "actor created, will send stop intent");
+
+		// Poll for actor to be destroyed after graceful stop
+		let actor = loop {
+			let actor =
+				common::try_get_actor(ctx.leader_dc().guard_port(), &actor_id_str, &namespace)
+					.await
+					.expect("failed to get actor")
+					.expect("actor should exist");
+
+			if actor.destroy_ts.is_some() {
+				break actor;
+			}
+
+			tokio::time::sleep(tokio::time::Duration::from_millis(50)).await;
+		};
+
+		assert!(
+			actor.destroy_ts.is_some(),
+			"actor should be destroyed after graceful stop with destroy policy"
+		);
+
+		// Verify envoy slot freed (actor no longer on envoy)
+		assert!(
+			!envoy.has_actor(&actor_id_str).await,
+			"actor should be removed from envoy after destroy"
+		);
+
+		tracing::info!(?actor_id_str, "actor gracefully stopped and destroyed");
+	});
+}
+
+#[test]
+fn envoy_actor_explicit_destroy() {
+	common::run(common::TestOpts::new(1), |ctx| async move {
+		let (namespace, _) = common::setup_test_namespace(ctx.leader_dc()).await;
+
+		// Create a channel to be notified when the actor starts
+		let (start_tx, start_rx) = tokio::sync::oneshot::channel();
+		let start_tx = Arc::new(Mutex::new(Some(start_tx)));
+
+		// Build a custom envoy with NotifyOnStartActor
+		let envoy = common::setup_envoy(ctx.leader_dc(), &namespace, |builder| {
+			builder.with_actor_behavior("test-actor", move |_| {
+				Box::new(common::test_envoy::NotifyOnStartActor::new(
+					start_tx.clone(),
+				))
+			})
+		})
+		.await;
+
+		let res = common::create_actor(
+			ctx.leader_dc().guard_port(),
+			&namespace,
+			"test-actor",
+			envoy.pool_name(),
+			rivet_types::actors::CrashPolicy::Destroy,
+		)
+		.await;
+
+		let actor_id = res.actor.actor_id.to_string();
+
+		start_rx
+			.await
+			.expect("actor should have sent start notification");
+
+		// Verify actor is running
+		assert!(envoy.has_actor(&actor_id).await, "envoy should have actor");
+
+		// Delete the actor
+		common::api::public::actors_delete(
+			ctx.leader_dc().guard_port(),
+			common::api_types::actors::delete::DeletePath {
+				actor_id: actor_id.parse().expect("failed to parse actor_id"),
+			},
+			common::api_types::actors::delete::DeleteQuery {
+				namespace: namespace.clone(),
+			},
+		)
+		.await
+		.expect("failed to delete actor");
+
+		// Poll for actor to be destroyed or timeout after 5s
+		let start = std::time::Instant::now();
+		let actor = loop {
+			let actor = common::try_get_actor(ctx.leader_dc().guard_port(), &actor_id, &namespace)
+				.await
+				.expect("failed to get actor")
+				.expect("actor should still exist in database");
+
+			if actor.destroy_ts.is_some() {
+				break actor;
+			}
+
+			if start.elapsed() > std::time::Duration::from_secs(5) {
+				panic!("actor deletion timed out after 5 seconds");
+			}
+
+			tokio::time::sleep(tokio::time::Duration::from_millis(50)).await;
+		};
+
+		assert!(
+			actor.destroy_ts.is_some(),
+			"destroy_ts should be set after deletion"
+		);
+
+		tracing::info!(?actor_id, "actor successfully destroyed");
+	});
+}
+
+// MARK: 5. Crash Handling and Policies
+#[test]
+fn envoy_crash_policy_restart() {
+	common::run(common::TestOpts::new(1), |ctx| async move {
+		let (namespace, _) = common::setup_test_namespace(ctx.leader_dc()).await;
+
+		// Create channel to be notified when actor crashes
+		let (crash_tx, crash_rx) = tokio::sync::oneshot::channel();
+		let crash_tx = Arc::new(Mutex::new(Some(crash_tx)));
+
+		// Create envoy client with crashing actor
+		let envoy = common::setup_envoy(ctx.leader_dc(), &namespace, |builder| {
+			builder.with_actor_behavior("crash-actor", move |_| {
+				Box::new(common::test_envoy::CrashOnStartActor::new_with_notify(
+					1,
+					crash_tx.clone(),
+				))
+			})
+		})
+		.await;
+
+		tracing::info!("envoy client ready, creating actor with restart policy");
+
+		// Create actor with restart crash policy
+		let res = common::create_actor(
+			ctx.leader_dc().guard_port(),
+			&namespace,
+			"crash-actor",
+			envoy.pool_name(),
+			rivet_types::actors::CrashPolicy::Restart,
+		)
+		.await;
+
+		let actor_id_str = res.actor.actor_id.to_string();
+
+		tracing::info!(?actor_id_str, "actor created, will crash on start");
+
+		// Wait for crash notification
+		crash_rx
+			.await
+			.expect("actor should have sent crash notification");
+
+		// Poll for reschedule_ts to be set (system needs to process the crash)
+		let actor = loop {
+			let actor =
+				common::try_get_actor(ctx.leader_dc().guard_port(), &actor_id_str, &namespace)
+					.await
+					.expect("failed to get actor")
+					.expect("actor should exist");
+
+			if actor.reschedule_ts.is_some() {
+				break actor;
+			}
+
+			tokio::time::sleep(tokio::time::Duration::from_millis(50)).await;
+		};
+
+		assert!(
+			actor.reschedule_ts.is_some(),
+			"actor should have reschedule_ts after crash with restart policy"
+		);
+
+		tracing::info!(?actor_id_str, reschedule_ts = ?actor.reschedule_ts, "actor scheduled for restart");
+	});
+}
+
+#[test]
+fn envoy_crash_policy_restart_resets_on_success() {
+	common::run(common::TestOpts::new(1), |ctx| async move {
+		let (namespace, _) = common::setup_test_namespace(ctx.leader_dc()).await;
+
+		let crash_count = Arc::new(Mutex::new(0));
+
+		// Create envoy client with actor that crashes 2 times then succeeds
+		let envoy = common::setup_envoy(ctx.leader_dc(), &namespace, |builder| {
+			builder.with_actor_behavior("crash-recover-actor", move |_| {
+				Box::new(common::test_envoy::CrashNTimesThenSucceedActor::new(
+					2,
+					crash_count.clone(),
+				))
+			})
+		})
+		.await;
+
+		tracing::info!("envoy client ready, creating actor with restart policy");
+
+		// Create actor with restart crash policy
+		let res = common::create_actor(
+			ctx.leader_dc().guard_port(),
+			&namespace,
+			"crash-recover-actor",
+			envoy.pool_name(),
+			rivet_types::actors::CrashPolicy::Restart,
+		)
+		.await;
+
+		let actor_id_str = res.actor.actor_id.to_string();
+
+		tracing::info!(
+			?actor_id_str,
+			"actor created, will crash twice then succeed"
+		);
+
+		// Poll for actor to eventually become connectable after crashes and restarts
+		// The actor should crash twice, reschedule, and eventually run successfully
+		let actor = loop {
+			let actor =
+				common::try_get_actor(ctx.leader_dc().guard_port(), &actor_id_str, &namespace)
+					.await
+					.expect("failed to get actor")
+					.expect("actor should exist");
+
+			// Actor successfully running after retries
+			if actor.connectable_ts.is_some() {
+				break actor;
+			}
+
+			tokio::time::sleep(tokio::time::Duration::from_millis(100)).await;
+		};
+
+		assert!(
+			actor.connectable_ts.is_some(),
+			"actor should eventually become connectable after crashes"
+		);
+		// actor.reschedule_ts is always Some(), not sure if this is intended
+		assert!(
+			actor.reschedule_ts.is_none()
+				|| (actor.connectable_ts.unwrap() > actor.reschedule_ts.unwrap()),
+			"actor should not be scheduled for retry when running successfully"
+		);
+
+		tracing::info!(?actor_id_str, "actor successfully recovered after crashes");
+	});
+}
+
+#[test]
+fn envoy_crash_policy_sleep() {
+	common::run(common::TestOpts::new(1), |ctx| async move {
+		let (namespace, _) = common::setup_test_namespace(ctx.leader_dc()).await;
+
+		// Create channel to be notified when actor crashes
+		let (crash_tx, crash_rx) = tokio::sync::oneshot::channel();
+		let crash_tx = Arc::new(Mutex::new(Some(crash_tx)));
+
+		// Create envoy client with crashing actor
+		let envoy = common::setup_envoy(ctx.leader_dc(), &namespace, |builder| {
+			builder.with_actor_behavior("crash-actor", move |_| {
+				Box::new(common::test_envoy::CrashOnStartActor::new_with_notify(
+					1,
+					crash_tx.clone(),
+				))
+			})
+		})
+		.await;
+
+		tracing::info!("envoy client ready, creating actor with sleep policy");
+
+		// Create actor with sleep crash policy
+		let res = common::create_actor(
+			ctx.leader_dc().guard_port(),
+			&namespace,
+			"crash-actor",
+			envoy.pool_name(),
+			rivet_types::actors::CrashPolicy::Sleep,
+		)
+		.await;
+
+		let actor_id_str = res.actor.actor_id.to_string();
+
+		tracing::info!(?actor_id_str, "actor created with sleep policy");
+
+		// Wait for crash notification
+		crash_rx
+			.await
+			.expect("actor should have sent crash notification");
+
+		// Poll for sleep_ts to be set (system needs to process the crash)
+		let actor = loop {
+			let actor =
+				common::try_get_actor(ctx.leader_dc().guard_port(), &actor_id_str, &namespace)
+					.await
+					.expect("failed to get actor")
+					.expect("actor should exist");
+
+			if actor.sleep_ts.is_some() {
+				break actor;
+			}
+
+			tokio::time::sleep(tokio::time::Duration::from_millis(50)).await;
+		};
+
+		assert!(
+			actor.sleep_ts.is_some(),
+			"actor should be sleeping after crash with sleep policy"
+		);
+		assert!(
+			actor.connectable_ts.is_none(),
+			"actor should not be connectable while sleeping"
+		);
+
+		tracing::info!(
+			?actor_id_str,
+			"actor correctly entered sleep state after crash"
+		);
+	});
+}
+
+#[test]
+fn envoy_crash_policy_destroy() {
+	common::run(common::TestOpts::new(1), |ctx| async move {
+		let (namespace, _) = common::setup_test_namespace(ctx.leader_dc()).await;
+
+		// Create channel to be notified when actor crashes
+		let (crash_tx, crash_rx) = tokio::sync::oneshot::channel();
+		let crash_tx = Arc::new(Mutex::new(Some(crash_tx)));
+
+		// Create envoy client with crashing actor
+		let envoy = common::setup_envoy(ctx.leader_dc(), &namespace, |builder| {
+			builder.with_actor_behavior("crash-actor", move |_| {
+				Box::new(common::test_envoy::CrashOnStartActor::new_with_notify(
+					1,
+					crash_tx.clone(),
+				))
+			})
+		})
+		.await;
+
+		tracing::info!("envoy client ready, creating actor with destroy policy");
+
+		// Create actor with destroy crash policy
+		let res = common::create_actor(
+			ctx.leader_dc().guard_port(),
+			&namespace,
+			"crash-actor",
+			envoy.pool_name(),
+			rivet_types::actors::CrashPolicy::Destroy,
+		)
+		.await;
+
+		let actor_id_str = res.actor.actor_id.to_string();
+
+		tracing::info!(?actor_id_str, "actor created with destroy policy");
+
+		// Wait for crash notification
+		crash_rx
+			.await
+			.expect("actor should have sent crash notification");
+
+		// Poll for destroy_ts to be set (system needs to process the crash)
+		let actor = loop {
+			let actor =
+				common::try_get_actor(ctx.leader_dc().guard_port(), &actor_id_str, &namespace)
+					.await
+					.expect("failed to get actor")
+					.expect("actor should exist");
+
+			if actor.destroy_ts.is_some() {
+				break actor;
+			}
+
+			tokio::time::sleep(tokio::time::Duration::from_millis(50)).await;
+		};
+
+		assert!(
+			actor.destroy_ts.is_some(),
+			"actor should be destroyed after crash with destroy policy"
+		);
+
+		tracing::info!(?actor_id_str, "actor correctly destroyed after crash");
+	});
+}
+
+// MARK: 6. Sleep and Wake
+#[test]
+fn envoy_actor_sleep_intent() {
+	common::run(common::TestOpts::new(1), |ctx| async move {
+		let (namespace, _) = common::setup_test_namespace(ctx.leader_dc()).await;
+
+		// Create channel to be notified when actor sends sleep intent
+		let (sleep_tx, sleep_rx) = tokio::sync::oneshot::channel();
+		let sleep_tx = Arc::new(Mutex::new(Some(sleep_tx)));
+
+		// Create envoy client with sleep actor behavior
+		let envoy = common::setup_envoy(ctx.leader_dc(), &namespace, |builder| {
+			builder.with_actor_behavior("sleep-actor", move |_| {
+				Box::new(common::test_envoy::SleepImmediatelyActor::new_with_notify(
+					sleep_tx.clone(),
+				))
+			})
+		})
+		.await;
+
+		tracing::info!("envoy client ready, creating actor that will sleep");
+
+		// Create actor
+		let res = common::create_actor(
+			ctx.leader_dc().guard_port(),
+			&namespace,
+			"sleep-actor",
+			envoy.pool_name(),
+			rivet_types::actors::CrashPolicy::Destroy,
+		)
+		.await;
+
+		let actor_id_str = res.actor.actor_id.to_string();
+
+		tracing::info!(?actor_id_str, "actor created, will send sleep intent");
+
+		// Wait for sleep intent notification
+		sleep_rx
+			.await
+			.expect("actor should have sent sleep intent notification");
+
+		// Poll for sleep_ts to be set (system needs to process the sleep intent)
+		let actor = loop {
+			let actor =
+				common::try_get_actor(ctx.leader_dc().guard_port(), &actor_id_str, &namespace)
+					.await
+					.expect("failed to get actor")
+					.expect("actor should exist");
+
+			if actor.sleep_ts.is_some() {
+				break actor;
+			}
+
+			tokio::time::sleep(tokio::time::Duration::from_millis(50)).await;
+		};
+
+		assert!(
+			actor.sleep_ts.is_some(),
+			"actor should have sleep_ts after sending sleep intent"
+		);
+		assert!(
+			actor.connectable_ts.is_none(),
+			"actor should not be connectable while sleeping"
+		);
+
+		tracing::info!(?actor_id_str, "actor correctly entered sleep state");
+	});
+}
+
+// MARK: Pending Allocation Queue
+#[test]
+fn envoy_actor_pending_allocation_no_envoys() {
+	common::run(common::TestOpts::new(1), |ctx| async move {
+		// Create namespace and start a envoy with 1 slot
+		let (namespace, _) = common::setup_test_namespace(ctx.leader_dc()).await;
+
+		let envoy_full = common::setup_envoy(ctx.leader_dc(), &namespace, |builder| {
+			builder
+				.with_actor_behavior("filler-actor", |_| {
+					Box::new(common::test_envoy::EchoActor::new())
+				})
+				.with_actor_behavior("test-actor", |_| {
+					Box::new(common::test_envoy::EchoActor::new())
+				})
+		})
+		.await;
+
+		tracing::info!("envoy with 1 slot started");
+
+		// Fill the slot with a filler actor
+		let filler_res = common::create_actor(
+			ctx.leader_dc().guard_port(),
+			&namespace,
+			"filler-actor",
+			envoy_full.pool_name(),
+			rivet_types::actors::CrashPolicy::Destroy,
+		)
+		.await;
+
+		let filler_actor_id = filler_res.actor.actor_id.to_string();
+
+		// Wait for filler actor to be allocated
+		loop {
+			if envoy_full.has_actor(&filler_actor_id).await {
+				break;
+			}
+			tokio::time::sleep(tokio::time::Duration::from_millis(50)).await;
+		}
+
+		tracing::info!(
+			?filler_actor_id,
+			"filler actor allocated, envoy is now full"
+		);
+
+		// Create test actor (should be pending because envoy is full)
+		let res = common::create_actor(
+			ctx.leader_dc().guard_port(),
+			&namespace,
+			"test-actor",
+			envoy_full.pool_name(),
+			rivet_types::actors::CrashPolicy::Destroy,
+		)
+		.await;
+
+		let actor_id = res.actor.actor_id.to_string();
+
+		// Verify actor is in pending state
+		let actor = common::try_get_actor(ctx.leader_dc().guard_port(), &actor_id, &namespace)
+			.await
+			.expect("failed to get actor")
+			.expect("actor should exist");
+
+		assert!(
+			actor.pending_allocation_ts.is_some(),
+			"pending_allocation_ts should be set when no envoys available"
+		);
+		assert!(
+			actor.connectable_ts.is_none(),
+			"actor should not be connectable yet"
+		);
+
+		tracing::info!(?actor_id, "actor is pending allocation");
+
+		// Now start a envoy with available slots
+		let envoy = common::setup_envoy(ctx.leader_dc(), &namespace, |builder| {
+			builder.with_actor_behavior("test-actor", |_| {
+				Box::new(common::test_envoy::EchoActor::new())
+			})
+		})
+		.await;
+
+		tracing::info!("envoy with 20 slots started");
+
+		// Poll for allocation
+		loop {
+			if envoy.has_actor(&actor_id).await {
+				break;
+			}
+			tokio::time::sleep(tokio::time::Duration::from_millis(50)).await;
+		}
+
+		// Verify actor is now allocated
+		assert!(
+			envoy.has_actor(&actor_id).await,
+			"actor should now be allocated to envoy"
+		);
+
+		tracing::info!(
+			?actor_id,
+			"actor successfully allocated after envoy with slots started"
+		);
+	});
+}
+
+#[test]
+fn envoy_pending_allocation_queue_ordering() {
+	common::run(common::TestOpts::new(1), |ctx| async move {
+		// Create namespace and start envoy with only 2 slots
+		let (namespace, _) = common::setup_test_namespace(ctx.leader_dc()).await;
+
+		let envoy = common::setup_envoy(ctx.leader_dc(), &namespace, |builder| {
+			builder
+				.with_actor_behavior("test-actor-0", |_| {
+					Box::new(common::test_envoy::EchoActor::new())
+				})
+				.with_actor_behavior("test-actor-1", |_| {
+					Box::new(common::test_envoy::EchoActor::new())
+				})
+				.with_actor_behavior("test-actor-2", |_| {
+					Box::new(common::test_envoy::EchoActor::new())
+				})
+		})
+		.await;
+
+		tracing::info!("envoy with 2 slots started");
+
+		// Create 3 actors in sequence
+		// First 2 should be allocated immediately, 3rd should be pending
+		let mut actor_ids = Vec::new();
+		for i in 0..3 {
+			let name = format!("test-actor-{}", i);
+			let res = common::create_actor(
+				ctx.leader_dc().guard_port(),
+				&namespace,
+				&name,
+				envoy.pool_name(),
+				rivet_types::actors::CrashPolicy::Destroy,
+			)
+			.await;
+
+			actor_ids.push(res.actor.actor_id.to_string());
+
+			// Small delay to ensure ordering
+			tokio::time::sleep(tokio::time::Duration::from_millis(50)).await;
+		}
+
+		// Poll for first 2 actors to be allocated
+		loop {
+			let has_0 = envoy.has_actor(&actor_ids[0]).await;
+			let has_1 = envoy.has_actor(&actor_ids[1]).await;
+
+			if has_0 && has_1 {
+				break;
+			}
+
+			tokio::time::sleep(tokio::time::Duration::from_millis(50)).await;
+		}
+
+		// Verify first 2 actors are allocated (FIFO)
+		assert!(
+			envoy.has_actor(&actor_ids[0]).await,
+			"first actor should be allocated"
+		);
+		assert!(
+			envoy.has_actor(&actor_ids[1]).await,
+			"second actor should be allocated"
+		);
+
+		// Third actor should still be pending
+		let actor_c =
+			common::try_get_actor(ctx.leader_dc().guard_port(), &actor_ids[2], &namespace)
+				.await
+				.expect("failed to get actor")
+				.expect("actor should exist");
+
+		assert!(
+			actor_c.pending_allocation_ts.is_some(),
+			"third actor should still be pending"
+		);
+
+		tracing::info!("FIFO allocation ordering verified");
+	});
+}
+
+// MARK: envoy Failures
+#[test]
+fn envoy_actor_survives_envoy_disconnect() {
+	common::run(
+		common::TestOpts::new(1).with_timeout(60),
+		|ctx| async move {
+			let (namespace, _) = common::setup_test_namespace(ctx.leader_dc()).await;
+
+			// Create envoy and start actor
+			let (start_tx, start_rx) = tokio::sync::oneshot::channel();
+			let start_tx = Arc::new(Mutex::new(Some(start_tx)));
+
+			let envoy = common::setup_envoy(ctx.leader_dc(), &namespace, |builder| {
+				builder.with_actor_behavior("test-actor", move |_| {
+					Box::new(common::test_envoy::NotifyOnStartActor::new(
+						start_tx.clone(),
+					))
+				})
+			})
+			.await;
+
+			let res = common::create_actor(
+				ctx.leader_dc().guard_port(),
+				&namespace,
+				"test-actor",
+				envoy.pool_name(),
+				rivet_types::actors::CrashPolicy::Restart,
+			)
+			.await;
+
+			let actor_id_str = res.actor.actor_id.to_string();
+
+			// Wait for actor to start
+			start_rx
+				.await
+				.expect("actor should have sent start notification");
+
+			tracing::info!(?actor_id_str, "actor started, simulating envoy disconnect");
+
+			// Simulate envoy disconnect by shutting down
+			envoy.shutdown().await;
+
+			tracing::info!(
+				"envoy disconnected, waiting for system to detect and apply crash policy"
+			);
+
+			// Now we wait for envoy_lost_threshold so that actor state updates
+			tokio::time::sleep(tokio::time::Duration::from_millis(
+				ctx.leader_dc()
+					.config
+					.pegboard()
+					.envoy_lost_threshold()
+					.try_into()
+					.unwrap(),
+			))
+			.await;
+
+			// Poll for actor to be rescheduled (crash policy is Restart)
+			// The system should detect envoy loss and apply the crash policy
+			let start = std::time::Instant::now();
+			let actor = loop {
+				let actor =
+					common::try_get_actor(ctx.leader_dc().guard_port(), &actor_id_str, &namespace)
+						.await
+						.expect("failed to get actor")
+						.expect("actor should exist");
+				tracing::warn!(?actor);
+				// Actor should be waiting for an allocation after envoy loss
+				if actor.pending_allocation_ts.is_some() {
+					break actor;
+				}
+
+				if start.elapsed() > std::time::Duration::from_secs(50) {
+					// TODO: Always times out here
+					tracing::info!(?actor);
+					break actor;
+				}
+
+				tokio::time::sleep(tokio::time::Duration::from_millis(500)).await;
+			};
+
+			assert!(
+				actor.pending_allocation_ts.is_some(),
+				"actor should be pending allocation after envoy disconnected and threshold hit with restart policy"
+			);
+			assert!(
+				actor.connectable_ts.is_none(),
+				"actor should not be connectable after envoy disconnect"
+			);
+		},
+	);
+}
+
+// MARK: Resource Limits
+#[test]
+#[ignore]
+fn envoy_at_max_capacity() {
+	common::run(
+		common::TestOpts::new(1).with_timeout(30),
+		|ctx| async move {
+			let (namespace, _) = common::setup_test_namespace(ctx.leader_dc()).await;
+
+			// Start envoy with only 2 slots
+
+			let envoy = common::setup_envoy(ctx.leader_dc(), &namespace, |builder| {
+				builder.with_actor_behavior("test-actor", move |_| {
+					Box::new(common::test_envoy::EchoActor::new())
+				})
+			})
+			.await;
+
+			// Create first two actors to fill capacity
+			let mut actor_ids = Vec::new();
+			for _i in 0..2 {
+				let res = common::create_actor(
+					ctx.leader_dc().guard_port(),
+					&namespace,
+					"test-actor",
+					envoy.pool_name(),
+					rivet_types::actors::CrashPolicy::Destroy,
+				)
+				.await;
+
+				actor_ids.push(res.actor.actor_id.to_string());
+			}
+
+			// Poll for both actors to be allocated
+			loop {
+				let has_0 = envoy.has_actor(&actor_ids[0]).await;
+				let has_1 = envoy.has_actor(&actor_ids[1]).await;
+
+				if has_0 && has_1 {
+					break;
+				}
+
+				tokio::time::sleep(tokio::time::Duration::from_millis(50)).await;
+			}
+
+			// Verify both actors are allocated
+			assert!(envoy.has_actor(&actor_ids[0]).await);
+			assert!(envoy.has_actor(&actor_ids[1]).await);
+
+			// Create third actor (should be pending)
+			let res3 = common::create_actor(
+				ctx.leader_dc().guard_port(),
+				&namespace,
+				"test-actor",
+				envoy.pool_name(),
+				rivet_types::actors::CrashPolicy::Destroy,
+			)
+			.await;
+
+			let actor_id3 = res3.actor.actor_id.to_string();
+
+			// Verify third actor is pending
+			let actor3 =
+				common::try_get_actor(ctx.leader_dc().guard_port(), &actor_id3, &namespace)
+					.await
+					.expect("failed to get actor")
+					.expect("actor should exist");
+
+			assert!(
+				actor3.pending_allocation_ts.is_some(),
+				"third actor should be pending when envoy at capacity"
+			);
+
+			// Destroy first actor to free a slot
+			common::api::public::actors_delete(
+				ctx.leader_dc().guard_port(),
+				common::api_types::actors::delete::DeletePath {
+					actor_id: actor_ids[0].parse().unwrap(),
+				},
+				common::api_types::actors::delete::DeleteQuery {
+					namespace: namespace.clone(),
+				},
+			)
+			.await
+			.expect("failed to delete actor");
+
+			// Poll for third actor to be allocated (wait for slot to free and pending actor to be allocated)
+			loop {
+				tracing::warn!(
+					"polling envoy: current actors: {:?}",
+					envoy.get_actor_ids().await
+				);
+				if envoy.has_actor(&actor_id3).await {
+					break;
+				}
+				tokio::time::sleep(tokio::time::Duration::from_millis(100)).await;
+			}
+
+			// Verify third actor is now allocated
+			assert!(
+				envoy.has_actor(&actor_id3).await,
+				"pending actor should be allocated after slot freed"
+			);
+		},
+	);
+}
+
+// MARK: Timeout and Retry Scenarios
+#[test]
+fn envoy_exponential_backoff_max_retries() {
+	common::run(common::TestOpts::new(1), |ctx| async move {
+		let (namespace, _) = common::setup_test_namespace(ctx.leader_dc()).await;
+
+		// Create envoy client with always-crashing actor
+
+		let envoy = common::setup_envoy(ctx.leader_dc(), &namespace, |builder| {
+			builder.with_actor_behavior("crash-always-actor", move |_| {
+				Box::new(common::test_envoy::CrashOnStartActor::new(1))
+			})
+		})
+		.await;
+
+		tracing::info!("envoy client ready, creating actor that will always crash");
+
+		// Create actor with restart crash policy
+		let res = common::create_actor(
+			ctx.leader_dc().guard_port(),
+			&namespace,
+			"crash-always-actor",
+			envoy.pool_name(),
+			rivet_types::actors::CrashPolicy::Restart,
+		)
+		.await;
+
+		let actor_id_str = res.actor.actor_id.to_string();
+
+		tracing::info!(?actor_id_str, "actor created, will crash repeatedly");
+
+		// Track reschedule timestamps to verify backoff increases
+		let mut previous_reschedule_ts: Option<i64> = None;
+		let mut backoff_deltas = Vec::new();
+
+		// Poll for multiple crashes and verify backoff increases
+		for iteration in 0..5 {
+			let actor = loop {
+				let actor =
+					common::try_get_actor(ctx.leader_dc().guard_port(), &actor_id_str, &namespace)
+						.await
+						.expect("failed to get actor")
+						.expect("actor should exist");
+
+				if actor.reschedule_ts.is_some() {
+					break actor;
+				}
+
+				tokio::time::sleep(tokio::time::Duration::from_millis(50)).await;
+			};
+
+			let current_reschedule_ts = actor.reschedule_ts.expect("reschedule_ts should be set");
+
+			tracing::info!(
+				iteration,
+				reschedule_ts = current_reschedule_ts,
+				"actor has reschedule_ts after crash"
+			);
+
+			// Calculate backoff delta if we have a previous timestamp
+			if let Some(prev_ts) = previous_reschedule_ts {
+				let delta = current_reschedule_ts - prev_ts;
+				backoff_deltas.push(delta);
+				tracing::info!(
+					iteration,
+					delta_ms = delta,
+					"backoff delta from previous reschedule"
+				);
+			}
+
+			previous_reschedule_ts = Some(current_reschedule_ts);
+
+			// Wait for the reschedule time to pass so next crash can happen
+			let now = rivet_util::timestamp::now();
+			if current_reschedule_ts > now {
+				let wait_duration = (current_reschedule_ts - now) as u64;
+				tracing::info!(
+					wait_duration_ms = wait_duration,
+					"waiting for reschedule time"
+				);
+				tokio::time::sleep(tokio::time::Duration::from_millis(wait_duration + 100)).await;
+			}
+		}
+
+		// Verify that backoff intervals generally increase (exponential backoff)
+		// We expect each delta to be larger than or equal to the previous
+		// (allowing some tolerance for system timing)
+		for i in 1..backoff_deltas.len() {
+			tracing::info!(
+				iteration = i,
+				current_delta = backoff_deltas[i],
+				previous_delta = backoff_deltas[i - 1],
+				"comparing backoff deltas"
+			);
+
+			// Allow some tolerance: current should be >= 80% of expected growth
+			// (exponential backoff typically doubles, but we allow for some variance)
+			assert!(
+				backoff_deltas[i] >= backoff_deltas[i - 1] / 2,
+				"backoff should not decrease significantly: iteration {}, prev={}, curr={}",
+				i,
+				backoff_deltas[i - 1],
+				backoff_deltas[i]
+			);
+		}
+
+		tracing::info!(
+			?backoff_deltas,
+			"exponential backoff verified across multiple crashes"
+		);
+	});
+}