Avoid suspending ephemeral agents

cli/golem-cli/src/command_handler/interactive.rs

@@ -228,6 +228,14 @@ impl InteractiveHandler {
         )
     }
 
+    pub fn confirm_interrupt_ephemeral_agent(&self) -> anyhow::Result<bool> {
+        self.confirm(
+            false,
+            "The target agent is ephemeral. Interrupting it will stop the current invocation and it cannot be resumed. Continue?",
+            None,
+        )
+    }
+
     pub fn confirm_reset_allow_incompatible_component_update(
         &self,
         component_name: &ComponentName,

cli/golem-cli/src/command_handler/worker/mod.rs

@@ -1123,6 +1123,20 @@ impl WorkerCommandHandler {
             .component_by_agent_name_match(&agent_name_match)
             .await?;
 
+        if component
+            .metadata
+            .agent_types()
+            .iter()
+            .find(|agent_type| agent_type.type_name == agent_name_match.agent_type_name)
+            .is_some_and(|agent_type| agent_type.mode == AgentMode::Ephemeral)
+            && !self
+                .ctx
+                .interactive_handler()
+                .confirm_interrupt_ephemeral_agent()?
+        {
+            bail!(NonSuccessfulExit);
+        }
+
         log_action(
             "Interrupting",
             format!("agent {}", format_agent_name_match(&agent_name_match)),

cli/golem-cli/wit/deps/golem-1.x/golem-oplog.wit

@@ -432,6 +432,19 @@ interface oplog {
         resource-name: string
     }
 
+    record ephemeral-sleep-too-long {
+        requested-nanos: u64,
+        max-nanos: u64
+    }
+
+    record ephemeral-fuel-exhausted {
+        overdraft-limit: u64
+    }
+
+    record ephemeral-cannot-suspend {
+        reason: string
+    }
+
     /// Describes the error that occurred in the agent
     variant worker-error {
         unknown(string),
@@ -448,7 +461,10 @@ interface oplog {
         exceeded-rpc-call-limit,
         node-out-of-filesystem-storage,
         agent-exceeded-filesystem-storage-limit,
-        agent-terminated-by-quota(agent-terminated-by-quota-error)
+        agent-terminated-by-quota(agent-terminated-by-quota-error),
+        ephemeral-sleep-too-long(ephemeral-sleep-too-long),
+        ephemeral-fuel-exhausted(ephemeral-fuel-exhausted),
+        ephemeral-cannot-suspend(ephemeral-cannot-suspend)
     }
 
     record raw-create-parameters {

golem-api-grpc/proto/golem/worker/worker_error.proto

@@ -21,6 +21,9 @@ message AgentError {
     ExceededHttpCallLimit exceeded_http_call_limit = 13;
     ExceededRpcCallLimit exceeded_rpc_call_limit = 14;
     AgentTerminatedByQuota agent_terminated_by_quota = 15;
+    EphemeralSleepTooLong ephemeral_sleep_too_long = 16;
+    EphemeralFuelExhausted ephemeral_fuel_exhausted = 17;
+    EphemeralCannotSuspend ephemeral_cannot_suspend = 18;
   }
 }
 
@@ -68,3 +71,16 @@ message AgentTerminatedByQuota {
   golem.common.EnvironmentId environment_id = 1;
   string resource_name = 2;
 }
+
+message EphemeralSleepTooLong {
+  uint64 requested_nanos = 1;
+  uint64 max_nanos = 2;
+}
+
+message EphemeralFuelExhausted {
+  uint64 overdraft_limit = 1;
+}
+
+message EphemeralCannotSuspend {
+  string reason = 1;
+}

golem-common/src/model/oplog/protobuf.rs

@@ -50,7 +50,8 @@ use crate::model::oplog::public_oplog_entry::{
     SnapshotParams, StartSpanParams, SuccessfulUpdateParams, SuspendParams,
 };
 use crate::model::oplog::{
-    AgentTerminatedByQuotaError, DurableFunctionType, OplogEntry, PersistenceLevel,
+    AgentTerminatedByQuotaError, DurableFunctionType, EphemeralCannotSuspendError,
+    EphemeralFuelExhaustedError, EphemeralSleepTooLongError, OplogEntry, PersistenceLevel,
 };
 use crate::model::quota::ResourceName;
 use crate::model::regions::OplogRegion;
@@ -129,6 +130,22 @@ impl TryFrom<golem_api_grpc::proto::golem::worker::AgentError> for AgentError {
                     resource_name: ResourceName(inner.resource_name),
                 }))
             }
+            Error::EphemeralSleepTooLong(inner) => {
+                Ok(Self::EphemeralSleepTooLong(EphemeralSleepTooLongError {
+                    requested_nanos: inner.requested_nanos,
+                    max_nanos: inner.max_nanos,
+                }))
+            }
+            Error::EphemeralFuelExhausted(inner) => {
+                Ok(Self::EphemeralFuelExhausted(EphemeralFuelExhaustedError {
+                    overdraft_limit: inner.overdraft_limit,
+                }))
+            }
+            Error::EphemeralCannotSuspend(inner) => {
+                Ok(Self::EphemeralCannotSuspend(EphemeralCannotSuspendError {
+                    reason: inner.reason,
+                }))
+            }
         }
     }
 }
@@ -184,6 +201,21 @@ impl From<AgentError> for golem_api_grpc::proto::golem::worker::AgentError {
                     resource_name: details.resource_name.0,
                 })
             }
+            AgentError::EphemeralSleepTooLong(EphemeralSleepTooLongError {
+                requested_nanos,
+                max_nanos,
+            }) => Error::EphemeralSleepTooLong(grpc_worker::EphemeralSleepTooLong {
+                requested_nanos,
+                max_nanos,
+            }),
+            AgentError::EphemeralFuelExhausted(EphemeralFuelExhaustedError { overdraft_limit }) => {
+                Error::EphemeralFuelExhausted(grpc_worker::EphemeralFuelExhausted {
+                    overdraft_limit,
+                })
+            }
+            AgentError::EphemeralCannotSuspend(EphemeralCannotSuspendError { reason }) => {
+                Error::EphemeralCannotSuspend(grpc_worker::EphemeralCannotSuspend { reason })
+            }
         };
         Self { error: Some(error) }
     }

golem-common/src/model/oplog/raw_types.rs

@@ -355,6 +355,25 @@ pub struct AgentTerminatedByQuotaError {
     pub resource_name: ResourceName,
 }
 
+#[derive(Clone, Debug, PartialEq, Eq, Hash, BinaryCodec, IntoValue, FromValue)]
+#[wit(name = "ephemeral-sleep-too-long", owner = "golem:api@1.5.0/oplog")]
+pub struct EphemeralSleepTooLongError {
+    pub requested_nanos: u64,
+    pub max_nanos: u64,
+}
+
+#[derive(Clone, Debug, PartialEq, Eq, Hash, BinaryCodec, IntoValue, FromValue)]
+#[wit(name = "ephemeral-fuel-exhausted", owner = "golem:api@1.5.0/oplog")]
+pub struct EphemeralFuelExhaustedError {
+    pub overdraft_limit: u64,
+}
+
+#[derive(Clone, Debug, PartialEq, Eq, Hash, BinaryCodec, IntoValue, FromValue)]
+#[wit(name = "ephemeral-cannot-suspend", owner = "golem:api@1.5.0/oplog")]
+pub struct EphemeralCannotSuspendError {
+    pub reason: String,
+}
+
 /// Describes the error that occurred in the worker
 #[derive(Clone, Debug, PartialEq, Eq, Hash, BinaryCodec, IntoValue, FromValue)]
 #[wit(name = "worker-error", owner = "golem:api@1.5.0/oplog")]
@@ -381,6 +400,12 @@ pub enum AgentError {
     AgentExceededFilesystemStorageLimit,
     // The agent was terminated by a quota with the terminae enforcement action (permanent)
     AgentTerminatedByQuota(AgentTerminatedByQuotaError),
+    // Ephemeral agents cannot suspend and the requested sleep exceeded the configured maximum
+    EphemeralSleepTooLong(EphemeralSleepTooLongError),
+    // Ephemeral agent exhausted its per-invocation fuel overdraft allowance
+    EphemeralFuelExhausted(EphemeralFuelExhaustedError),
+    // Ephemeral agent reached a condition that would suspend a durable agent
+    EphemeralCannotSuspend(EphemeralCannotSuspendError),
 }
 
 impl AgentError {
@@ -401,6 +426,9 @@ impl AgentError {
             Self::NodeOutOfFilesystemStorage => "Out of storage space",
             Self::AgentExceededFilesystemStorageLimit => "Exceeded plan storage limit",
             Self::AgentTerminatedByQuota(_) => "Terminated by quota",
+            Self::EphemeralSleepTooLong(_) => "Ephemeral sleep too long",
+            Self::EphemeralFuelExhausted(_) => "Ephemeral fuel exhausted",
+            Self::EphemeralCannotSuspend(_) => "Ephemeral agent cannot suspend",
         }
     }
 

golem-common/wit/deps/golem-1.x/golem-oplog.wit

@@ -432,6 +432,19 @@ interface oplog {
         resource-name: string
     }
 
+    record ephemeral-sleep-too-long {
+        requested-nanos: u64,
+        max-nanos: u64
+    }
+
+    record ephemeral-fuel-exhausted {
+        overdraft-limit: u64
+    }
+
+    record ephemeral-cannot-suspend {
+        reason: string
+    }
+
     /// Describes the error that occurred in the agent
     variant worker-error {
         unknown(string),
@@ -448,7 +461,10 @@ interface oplog {
         exceeded-rpc-call-limit,
         node-out-of-filesystem-storage,
         agent-exceeded-filesystem-storage-limit,
-        agent-terminated-by-quota(agent-terminated-by-quota-error)
+        agent-terminated-by-quota(agent-terminated-by-quota-error),
+        ephemeral-sleep-too-long(ephemeral-sleep-too-long),
+        ephemeral-fuel-exhausted(ephemeral-fuel-exhausted),
+        ephemeral-cannot-suspend(ephemeral-cannot-suspend)
     }
 
     record raw-create-parameters {

golem-debugging-service/config/debug-worker-executor.sample.env

@@ -38,6 +38,7 @@ GOLEM__KEY_VALUE_STORAGE__CONFIG__RETRIES__MAX_DELAY="2s"
 GOLEM__KEY_VALUE_STORAGE__CONFIG__RETRIES__MAX_JITTER_FACTOR=0.15
 GOLEM__KEY_VALUE_STORAGE__CONFIG__RETRIES__MIN_DELAY="100ms"
 GOLEM__KEY_VALUE_STORAGE__CONFIG__RETRIES__MULTIPLIER=2.0
+GOLEM__LIMITS__EPHEMERAL_FUEL_OVERDRAFT_MULTIPLIER=100
 GOLEM__LIMITS__EPOCH_INTERVAL="10ms"
 GOLEM__LIMITS__EPOCH_TICKS=1
 GOLEM__LIMITS__EVENT_BROADCAST_CAPACITY=1024
@@ -112,6 +113,7 @@ GOLEM__RETRY__MAX_JITTER_FACTOR=0.15
 GOLEM__RETRY__MIN_DELAY="100ms"
 GOLEM__RETRY__MULTIPLIER=3.0
 GOLEM__SCHEDULER__REFRESH_INTERVAL="2s"
+GOLEM__SUSPEND__EPHEMERAL_MAX_SLEEP="1m"
 GOLEM__SUSPEND__SUSPEND_AFTER="10s"
 GOLEM__TRACING__CONSOLE=false
 GOLEM__TRACING__DTOR_FRIENDLY=false
@@ -196,6 +198,7 @@ GOLEM__KEY_VALUE_STORAGE__CONFIG__RETRIES__MAX_DELAY="2s"
 GOLEM__KEY_VALUE_STORAGE__CONFIG__RETRIES__MAX_JITTER_FACTOR=0.15
 GOLEM__KEY_VALUE_STORAGE__CONFIG__RETRIES__MIN_DELAY="100ms"
 GOLEM__KEY_VALUE_STORAGE__CONFIG__RETRIES__MULTIPLIER=2.0
+GOLEM__LIMITS__EPHEMERAL_FUEL_OVERDRAFT_MULTIPLIER=100
 GOLEM__LIMITS__EPOCH_INTERVAL="10ms"
 GOLEM__LIMITS__EPOCH_TICKS=1
 GOLEM__LIMITS__EVENT_BROADCAST_CAPACITY=1024
@@ -270,6 +273,7 @@ GOLEM__RETRY__MAX_JITTER_FACTOR=0.15
 GOLEM__RETRY__MIN_DELAY="100ms"
 GOLEM__RETRY__MULTIPLIER=3.0
 GOLEM__SCHEDULER__REFRESH_INTERVAL="2s"
+GOLEM__SUSPEND__EPHEMERAL_MAX_SLEEP="1m"
 GOLEM__SUSPEND__SUSPEND_AFTER="10s"
 GOLEM__TRACING__CONSOLE=false
 GOLEM__TRACING__DTOR_FRIENDLY=false

golem-debugging-service/config/debug-worker-executor.toml

@@ -71,6 +71,7 @@ min_delay = "100ms"
 multiplier = 2.0
 
 [limits]
+ephemeral_fuel_overdraft_multiplier = 100
 epoch_interval = "10ms"
 epoch_ticks = 1
 event_broadcast_capacity = 1024
@@ -187,6 +188,7 @@ multiplier = 3.0
 refresh_interval = "2s"
 
 [suspend]
+ephemeral_max_sleep = "1m"
 suspend_after = "10s"
 
 [tracing]
@@ -314,6 +316,7 @@ without_time = false
 # multiplier = 2.0
 # 
 # [limits]
+# ephemeral_fuel_overdraft_multiplier = 100
 # epoch_interval = "10ms"
 # epoch_ticks = 1
 # event_broadcast_capacity = 1024
@@ -430,6 +433,7 @@ without_time = false
 # refresh_interval = "2s"
 # 
 # [suspend]
+# ephemeral_max_sleep = "1m"
 # suspend_after = "10s"
 # 
 # [tracing]

golem-debugging-service/src/debug_context.rs

@@ -25,7 +25,7 @@ use golem_common::model::component::ComponentRevision;
 use golem_common::model::invocation_context::{
     self, AttributeValue, InvocationContextStack, SpanId,
 };
-use golem_common::model::oplog::TimestampedUpdateDescription;
+use golem_common::model::oplog::{AgentError, TimestampedUpdateDescription};
 use golem_common::model::{
     AgentId, AgentInvocation, AgentInvocationOutput, AgentStatusRecord, IdempotencyKey,
     OwnedAgentId, Timestamp,
@@ -98,8 +98,8 @@ impl DurableWorkerCtxView<DebugContext> for DebugContext {
 
 #[async_trait]
 impl FuelManagement for DebugContext {
-    fn ensure_fuel(&mut self, _current_level: u64) -> bool {
-        true
+    fn ensure_fuel(&mut self, _current_level: u64) -> Result<(), AgentError> {
+        Ok(())
     }
 
     fn return_fuel(&mut self, _current_level: u64) -> u64 {

golem-wasm/wit/deps/golem-1.x/golem-oplog.wit

@@ -432,6 +432,19 @@ interface oplog {
         resource-name: string
     }
 
+    record ephemeral-sleep-too-long {
+        requested-nanos: u64,
+        max-nanos: u64
+    }
+
+    record ephemeral-fuel-exhausted {
+        overdraft-limit: u64
+    }
+
+    record ephemeral-cannot-suspend {
+        reason: string
+    }
+
     /// Describes the error that occurred in the agent
     variant worker-error {
         unknown(string),
@@ -448,7 +461,10 @@ interface oplog {
         exceeded-rpc-call-limit,
         node-out-of-filesystem-storage,
         agent-exceeded-filesystem-storage-limit,
-        agent-terminated-by-quota(agent-terminated-by-quota-error)
+        agent-terminated-by-quota(agent-terminated-by-quota-error),
+        ephemeral-sleep-too-long(ephemeral-sleep-too-long),
+        ephemeral-fuel-exhausted(ephemeral-fuel-exhausted),
+        ephemeral-cannot-suspend(ephemeral-cannot-suspend)
     }
 
     record raw-create-parameters {

golem-worker-executor-test-utils/src/lib.rs

@@ -39,8 +39,8 @@ use golem_common::model::invocation_context::{
     AttributeValue, InvocationContextSpan, InvocationContextStack, SpanId,
 };
 use golem_common::model::oplog::{
-    OplogEntry, PayloadId, PersistenceLevel, RawOplogPayload, TimestampedUpdateDescription,
-    types::ObjectMetadata,
+    AgentError, OplogEntry, PayloadId, PersistenceLevel, RawOplogPayload,
+    TimestampedUpdateDescription, types::ObjectMetadata,
 };
 use golem_common::model::plan::PlanId;
 use golem_common::model::worker::{AgentConfigEntryDto, AgentMetadataDto};
@@ -733,8 +733,8 @@ impl wasmtime_wasi::p2::bindings::cli::environment::Host for TestWorkerCtx {
 
 #[async_trait]
 impl FuelManagement for TestWorkerCtx {
-    fn ensure_fuel(&mut self, _current_level: u64) -> bool {
-        true
+    fn ensure_fuel(&mut self, _current_level: u64) -> Result<(), AgentError> {
+        Ok(())
     }
 
     fn return_fuel(&mut self, _current_level: u64) -> u64 {