@temporalio/core-bridge 0.17.0 → 0.17.1

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@temporalio/core-bridge",
-  "version": "0.17.0",
+  "version": "0.17.1",
   "description": "Temporal.io SDK Core<>Node bridge",
   "main": "index.node",
   "types": "index.d.ts",
@@ -19,7 +19,7 @@
   "license": "MIT",
   "dependencies": {
     "@opentelemetry/api": "^1.0.3",
-    "@temporalio/common": "^0.17.0",
+    "@temporalio/common": "^0.17.1",
     "arg": "^5.0.1",
     "cargo-cp-artifact": "^0.1.4",
     "which": "^2.0.2"
@@ -40,5 +40,5 @@
   "publishConfig": {
     "access": "public"
   },
-  "gitHead": "7706410cb67a1c92b9ae9eea1cac48e0dad4a8d1"
+  "gitHead": "3ee26c92e9c18500f919654a402237a86e69652e"
 }

package/sdk-core/sdk-core-protos/src/lib.rs

@@ -264,6 +264,7 @@ pub mod coresdk {
     }
 
     pub mod workflow_completion {
+        use crate::temporal::api::failure;
         tonic::include_proto!("coresdk.workflow_completion");
 
         impl wf_activation_completion::Status {
@@ -274,6 +275,12 @@ pub mod coresdk {
                 }
             }
         }
+
+        impl From<failure::v1::Failure> for Failure {
+            fn from(f: failure::v1::Failure) -> Self {
+                Failure { failure: Some(f) }
+            }
+        }
     }
 
     pub mod child_workflow {

package/sdk-core/src/core_tests/queries.rs

@@ -290,10 +290,17 @@ async fn legacy_query_failure_on_wft_failure() {
     core.shutdown().await;
 }
 
+#[rstest::rstest]
 #[tokio::test]
-async fn legacy_query_with_full_history_after_complete() {
+async fn legacy_query_after_complete(#[values(false, true)] full_history: bool) {
     let wfid = "fake_wf_id";
-    let t = canned_histories::single_timer_wf_completes("1");
+    let t = if full_history {
+        canned_histories::single_timer_wf_completes("1")
+    } else {
+        let mut t = canned_histories::single_timer("1");
+        t.add_workflow_task_completed();
+        t
+    };
     let query_with_hist_task = {
         let mut pr = hist_to_poll_resp(
             &t,

package/sdk-core/src/core_tests/workflow_tasks.rs

@@ -1619,19 +1619,19 @@ async fn failing_wft_doesnt_eat_permit_forever() {
     t.add_by_type(EventType::WorkflowExecutionStarted);
     t.add_workflow_task_scheduled_and_started();
 
-    let failures = 5;
-    // One extra response for when we stop failing
-    let resps = (1..=(failures + 1)).map(|_| 1);
     let mock = MockServerGatewayApis::new();
-    let mut mock = single_hist_mock_sg("fake_wf_id", t, resps, mock, true);
+    let mut mock = single_hist_mock_sg("fake_wf_id", t, [1, 1, 1], mock, true);
     mock.worker_cfg(TEST_Q, |cfg| {
         cfg.max_cached_workflows = 2;
         cfg.max_outstanding_workflow_tasks = 2;
     });
+    let outstanding_mock_tasks = mock.outstanding_task_map.clone();
     let core = mock_core(mock);
 
-    // Spin failing the WFT to verify that we don't get stuck
-    for _ in 1..=failures {
+    let mut run_id = "".to_string();
+    // Fail twice, verifying a permit is eaten. We cannot fail the same run more than twice in a row
+    // because we purposefully time out rather than spamming.
+    for _ in 1..=2 {
         let activation = core.poll_workflow_activation(TEST_Q).await.unwrap();
         // Issue a nonsense completion that will trigger a WFT failure
         core.complete_workflow_activation(WfActivationCompletion::from_cmd(
@@ -1648,12 +1648,22 @@ async fn failing_wft_doesnt_eat_permit_forever() {
                 variant: Some(wf_activation_job::Variant::RemoveFromCache(_)),
             },]
         );
+        run_id = activation.run_id.clone();
         core.complete_workflow_activation(WfActivationCompletion::empty(TEST_Q, activation.run_id))
             .await
             .unwrap();
         assert_eq!(core.outstanding_wfts(TEST_Q), 0);
         assert_eq!(core.available_wft_permits(TEST_Q), 2);
     }
+    // We should be "out of work" because the mock service thinks we didn't complete the last task,
+    // which we didn't, because we don't spam failures. The real server would eventually time out
+    // the task. Mock doesn't understand that, so the WFT permit is released because eventually a
+    // new one will be generated. We manually clear the mock's outstanding task list so the next
+    // poll will work.
+    outstanding_mock_tasks
+        .unwrap()
+        .write()
+        .remove_by_left(&run_id);
     let activation = core.poll_workflow_activation(TEST_Q).await.unwrap();
     core.complete_workflow_activation(WfActivationCompletion::from_cmd(
         TEST_Q,

package/sdk-core/src/errors.rs

@@ -1,6 +1,6 @@
 //! Error types exposed by public APIs
 
-use crate::{machines::WFMachinesError, task_token::TaskToken, WorkerLookupErr};
+use crate::{machines::WFMachinesError, WorkerLookupErr};
 use temporal_sdk_core_protos::coresdk::{
     activity_result::ActivityResult, workflow_completion::WfActivationCompletion,
 };
@@ -11,9 +11,8 @@ pub(crate) struct WorkflowUpdateError {
     /// Underlying workflow error
     pub source: WFMachinesError,
     /// The run id of the erring workflow
+    #[allow(dead_code)] // Useful in debug output
     pub run_id: String,
-    /// The task token associated with this update, if one existed yet.
-    pub task_token: Option<TaskToken>,
 }
 
 impl From<WorkflowMissingError> for WorkflowUpdateError {
@@ -21,7 +20,6 @@ impl From<WorkflowMissingError> for WorkflowUpdateError {
         Self {
             source: WFMachinesError::Fatal("Workflow machines missing".to_string()),
             run_id: wme.run_id,
-            task_token: None,
         }
     }
 }

package/sdk-core/src/machines/child_workflow_state_machine.rs

@@ -7,8 +7,7 @@ use std::convert::{TryFrom, TryInto};
 use temporal_sdk_core_protos::{
     coresdk::{
         child_workflow::{
-            self as wfr, child_workflow_result::Status as ChildWorkflowStatus,
-            ChildWorkflowCancellationType, ChildWorkflowResult,
+            self as wfr, child_workflow_result::Status as ChildWorkflowStatus, ChildWorkflowResult,
         },
         common::Payload,
         workflow_activation::{
@@ -115,7 +114,6 @@ impl StartCommandCreated {
             StartEventRecorded::default(),
             SharedState {
                 initiated_event_id,
-                attrs: None, // Drop the attributes to avoid holding large payloads in memory
                 ..state
             },
         )
@@ -303,9 +301,7 @@ pub(super) struct SharedState {
     workflow_id: String,
     run_id: String,
     workflow_type: String,
-    cancellation_type: ChildWorkflowCancellationType,
     cancelled_before_sent: bool,
-    attrs: Option<StartChildWorkflowExecution>,
 }
 
 /// Creates a new child workflow state machine and a command to start it on the server.
@@ -329,11 +325,6 @@ impl ChildWorkflowMachine {
                 workflow_id: attribs.workflow_id.clone(),
                 workflow_type: attribs.workflow_type.clone(),
                 namespace: attribs.namespace.clone(),
-                cancellation_type: ChildWorkflowCancellationType::from_i32(
-                    attribs.cancellation_type,
-                )
-                .unwrap(),
-                attrs: Some(attribs.clone()),
                 ..Default::default()
             },
         };

package/sdk-core/src/machines/workflow_machines.rs

@@ -61,6 +61,10 @@ pub(crate) struct WorkflowMachines {
     /// Eventually, this number should reach the started id in the latest history update, but
     /// we must incrementally apply the history while communicating with lang.
     next_started_event_id: i64,
+    /// The event id of the most recent event processed. It's possible in some situations (ex legacy
+    /// queries) to receive a history with no new workflow tasks. If the last history we processed
+    /// also had no new tasks, we need a way to know not to apply the same events over again.
+    last_processed_event: i64,
     /// True if the workflow is replaying from history
     pub replaying: bool,
     /// Namespace this workflow exists in
@@ -120,7 +124,6 @@ struct CommandAndMachine {
 
 #[derive(Debug, Clone, Copy)]
 struct ChangeInfo {
-    deprecated: bool,
     created_command: bool,
 }
 
@@ -196,6 +199,7 @@ impl WorkflowMachines {
             // In an ideal world one could say ..Default::default() here and it'd still work.
             current_started_event_id: 0,
             next_started_event_id: 0,
+            last_processed_event: 0,
             workflow_start_time: None,
             workflow_end_time: None,
             current_wf_time: None,
@@ -529,11 +533,16 @@ impl WorkflowMachines {
         }
 
         let last_handled_wft_started_id = self.current_started_event_id;
-        let events = self
-            .last_history_from_server
-            .take_next_wft_sequence(last_handled_wft_started_id)
-            .await
-            .map_err(WFMachinesError::HistoryFetchingError)?;
+        let events = {
+            let mut evts = self
+                .last_history_from_server
+                .take_next_wft_sequence(last_handled_wft_started_id)
+                .await
+                .map_err(WFMachinesError::HistoryFetchingError)?;
+            // Do not re-process events we have already processed
+            evts.retain(|e| e.event_id > self.last_processed_event);
+            evts
+        };
 
         // We're caught up on reply if there are no new events to process
         // TODO: Probably this is unneeded if we evict whenever history is from non-sticky queue
@@ -564,23 +573,17 @@ impl WorkflowMachines {
 
         while let Some(event) = history.next() {
             let next_event = history.peek();
-
-            if event.event_type == EventType::WorkflowTaskStarted as i32 && next_event.is_none() {
-                self.handle_event(event, false)?;
-                break;
-            }
-
             self.handle_event(event, next_event.is_some())?;
+            self.last_processed_event = event.event_id;
         }
 
         // Scan through to the next WFT, searching for any patch markers, so that we can
         // pre-resolve them.
         for e in self.last_history_from_server.peek_next_wft_sequence() {
-            if let Some((patch_id, deprecated)) = e.get_changed_marker_details() {
+            if let Some((patch_id, _)) = e.get_changed_marker_details() {
                 self.encountered_change_markers.insert(
                     patch_id.clone(),
                     ChangeInfo {
-                        deprecated,
                         created_command: false,
                     },
                 );
@@ -743,7 +746,6 @@ impl WorkflowMachines {
                             self.encountered_change_markers.insert(
                                 attrs.patch_id,
                                 ChangeInfo {
-                                    deprecated: attrs.deprecated,
                                     created_command: true,
                                 },
                             );

package/sdk-core/src/test_help/history_info.rs

@@ -39,6 +39,7 @@ impl HistoryInfo {
             return Err(HistoryInfoError::HistoryEndsUnexpectedly);
         }
 
+        let is_all_hist = to_wf_task_num.is_none();
         let to_wf_task_num = to_wf_task_num.unwrap_or(usize::MAX);
         let mut workflow_task_started_event_id = 0;
         let mut previous_started_event_id = 0;
@@ -83,7 +84,7 @@ impl HistoryInfo {
             }
 
             if next_event.is_none() {
-                if event.is_final_wf_execution_event() {
+                if event.is_final_wf_execution_event() || is_all_hist {
                     return Ok(Self {
                         previous_started_event_id,
                         workflow_task_started_event_id,

package/sdk-core/src/test_help/mod.rs

@@ -146,6 +146,7 @@ pub struct FakeWfResponses {
 pub struct MocksHolder<SG> {
     sg: SG,
     mock_pollers: HashMap<String, MockWorker>,
+    // bidirectional mapping of run id / task token
     pub outstanding_task_map: Option<Arc<RwLock<BiMap<String, TaskToken>>>>,
 }
 
@@ -381,16 +382,6 @@ pub fn build_mock_pollers(mut cfg: MockPollCfg) -> MocksHolder<MockServerGateway
             }
         }
 
-        // TODO: Fix -- or not? Sticky invalidation could make this pointless anyway
-        // Verify response batches only ever return longer histories (IE: Are sorted ascending)
-        // assert!(
-        //     hist.response_batches
-        //         .as_slice()
-        //         .windows(2)
-        //         .all(|w| w[0] <= w[1]),
-        //     "response batches must have increasing wft numbers"
-        // );
-
         if cfg.enforce_correct_number_of_polls {
             *correct_num_polls.get_or_insert(0) += hist.response_batches.len();
         }

package/sdk-core/src/worker/mod.rs

@@ -329,20 +329,28 @@ impl Worker {
         completion: WfActivationCompletion,
     ) -> Result<(), CompleteWfError> {
         let wfstatus = completion.status;
-        let did_complete_wft = match wfstatus {
+        let report_outcome = match wfstatus {
             Some(wf_activation_completion::Status::Successful(success)) => {
                 self.wf_activation_success(&completion.run_id, success)
                     .await
             }
+
             Some(wf_activation_completion::Status::Failed(failure)) => {
-                self.wf_activation_failed(&completion.run_id, failure).await
+                self.wf_activation_failed(
+                    &completion.run_id,
+                    WorkflowTaskFailedCause::Unspecified,
+                    failure,
+                )
+                .await
+            }
+            None => {
+                return Err(CompleteWfError::MalformedWorkflowCompletion {
+                    reason: "Workflow completion had empty status field".to_owned(),
+                    completion: None,
+                })
             }
-            None => Err(CompleteWfError::MalformedWorkflowCompletion {
-                reason: "Workflow completion had empty status field".to_owned(),
-                completion: None,
-            }),
         }?;
-        self.after_workflow_activation(&completion.run_id, did_complete_wft);
+        self.after_workflow_activation(&completion.run_id, report_outcome);
         Ok(())
     }
 
@@ -509,7 +517,7 @@ impl Worker {
         &self,
         run_id: &str,
         success: workflow_completion::Success,
-    ) -> Result<bool, CompleteWfError> {
+    ) -> Result<WFTReportOutcome, CompleteWfError> {
         // Convert to wf commands
         let cmds = success
             .commands
@@ -553,7 +561,10 @@ impl Worker {
                         .await
                 })
                 .await?;
-                Ok(true)
+                Ok(WFTReportOutcome {
+                    reported_to_server: true,
+                    failed: false,
+                })
             }
             Ok(Some(ServerCommandsWithWorkflowInfo {
                 task_token,
@@ -563,9 +574,15 @@ impl Worker {
                 self.server_gateway
                     .respond_legacy_query(task_token, result)
                     .await?;
-                Ok(true)
+                Ok(WFTReportOutcome {
+                    reported_to_server: true,
+                    failed: false,
+                })
             }
-            Ok(None) => Ok(false),
+            Ok(None) => Ok(WFTReportOutcome {
+                reported_to_server: false,
+                failed: false,
+            }),
             Err(update_err) => {
                 // Automatically fail the workflow task in the event we couldn't update machines
                 let fail_cause = if matches!(&update_err.source, WFMachinesError::Nondeterminism(_))
@@ -574,30 +591,13 @@ impl Worker {
                 } else {
                     WorkflowTaskFailedCause::Unspecified
                 };
-
-                warn!(run_id, error=?update_err, "Failing workflow task");
-
-                if let Some(ref tt) = update_err.task_token {
-                    let wft_fail_str = format!("{:?}", update_err);
-                    self.handle_wft_reporting_errs(run_id, || async {
-                        self.server_gateway
-                            .fail_workflow_task(
-                                tt.clone(),
-                                fail_cause,
-                                Some(Failure::application_failure(wft_fail_str.clone(), false)),
-                            )
-                            .await
-                    })
-                    .await?;
-                    // We must evict the workflow since we've failed a WFT
-                    self.request_wf_eviction(
-                        run_id,
-                        format!("Workflow task failure: {}", wft_fail_str),
-                    );
-                    Ok(true)
-                } else {
-                    Ok(false)
-                }
+                let wft_fail_str = format!("{:?}", update_err);
+                self.wf_activation_failed(
+                    run_id,
+                    fail_cause,
+                    Failure::application_failure(wft_fail_str.clone(), false).into(),
+                )
+                .await
             }
         }
     }
@@ -608,35 +608,46 @@ impl Worker {
     async fn wf_activation_failed(
         &self,
         run_id: &str,
+        cause: WorkflowTaskFailedCause,
         failure: workflow_completion::Failure,
-    ) -> Result<bool, CompleteWfError> {
+    ) -> Result<WFTReportOutcome, CompleteWfError> {
         Ok(match self.wft_manager.failed_activation(run_id) {
             FailedActivationOutcome::Report(tt) => {
+                warn!(run_id, failure=?failure, "Failing workflow activation");
                 self.handle_wft_reporting_errs(run_id, || async {
                     self.server_gateway
-                        .fail_workflow_task(
-                            tt,
-                            WorkflowTaskFailedCause::Unspecified,
-                            failure.failure.map(Into::into),
-                        )
+                        .fail_workflow_task(tt, cause, failure.failure.map(Into::into))
                         .await
                 })
                 .await?;
-                true
+                WFTReportOutcome {
+                    reported_to_server: true,
+                    failed: true,
+                }
             }
             FailedActivationOutcome::ReportLegacyQueryFailure(task_token) => {
+                warn!(run_id, failure=?failure, "Failing legacy query request");
                 self.server_gateway
                     .respond_legacy_query(task_token, legacy_query_failure(failure))
                     .await?;
-                true
+                WFTReportOutcome {
+                    reported_to_server: true,
+                    failed: true,
+                }
             }
-            FailedActivationOutcome::NoReport => false,
+            FailedActivationOutcome::NoReport => WFTReportOutcome {
+                reported_to_server: false,
+                failed: true,
+            },
         })
     }
 
-    fn after_workflow_activation(&self, run_id: &str, did_complete_wft: bool) {
+    fn after_workflow_activation(&self, run_id: &str, report_outcome: WFTReportOutcome) {
         self.wft_manager.after_wft_report(run_id);
-        if did_complete_wft {
+        if report_outcome.reported_to_server || report_outcome.failed {
+            // If we failed the WFT but didn't report anything, we still want to release the WFT
+            // permit since the server will eventually time out the task and we've already evicted
+            // the run.
             self.return_workflow_task_permit();
         }
         self.wft_manager.on_activation_done(run_id);
@@ -718,6 +729,11 @@ impl WorkerConfig {
     }
 }
 
+struct WFTReportOutcome {
+    reported_to_server: bool,
+    failed: bool,
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;

package/sdk-core/src/workflow/workflow_tasks/mod.rs

@@ -135,7 +135,7 @@ pub enum ActivationAction {
 }
 
 macro_rules! machine_mut {
-    ($myself:ident, $run_id:ident, $task_token:ident, $clos:expr) => {{
+    ($myself:ident, $run_id:ident, $clos:expr) => {{
         $myself
             .workflow_machines
             .access($run_id, $clos)
@@ -143,7 +143,6 @@ macro_rules! machine_mut {
             .map_err(|source| WorkflowUpdateError {
                 source,
                 run_id: $run_id.to_owned(),
-                task_token: Some($task_token.clone()),
             })
     }};
 }
@@ -256,6 +255,7 @@ impl WorkflowTaskManager {
         debug!(
             task_token = %&work.task_token,
             history_length = %work.history.events.len(),
+            attempt = %work.attempt,
             "Applying new workflow task from server"
         );
         let task_start_time = Instant::now();
@@ -325,19 +325,20 @@ impl WorkflowTaskManager {
             return Ok(None);
         }
 
-        let task_token = if let Some(entry) = self.workflow_machines.get_task(run_id) {
-            entry.info.task_token.clone()
-        } else {
-            if !self.activation_has_eviction(run_id) {
-                // Don't bother warning if this was an eviction, since it's normal to issue
-                // eviction activations without an associated workflow task in that case.
-                warn!(
-                    run_id,
-                    "Attempted to complete activation for nonexistent run"
-                );
-            }
-            return Ok(None);
-        };
+        let (task_token, is_leg_query_task) =
+            if let Some(entry) = self.workflow_machines.get_task(run_id) {
+                (entry.info.task_token.clone(), entry.legacy_query.is_some())
+            } else {
+                if !self.activation_has_eviction(run_id) {
+                    // Don't bother warning if this was an eviction, since it's normal to issue
+                    // eviction activations without an associated workflow task in that case.
+                    warn!(
+                        run_id,
+                        "Attempted to complete activation for run without associated workflow task"
+                    );
+                }
+                return Ok(None);
+            };
 
         // If the only command in the activation is a legacy query response, that means we need
         // to respond differently than a typical activation.
@@ -364,7 +365,6 @@ impl WorkflowTaskManager {
                             return Err(WorkflowUpdateError {
                                 source: WFMachinesError::Fatal("Legacy query activation response included other commands, this is not allowed and constitutes an error in the lang SDK".to_string()),
                                 run_id: run_id.to_string(),
-                                task_token: Some(task_token)
                             });
                         }
                         query_responses.push(qr);
@@ -375,30 +375,32 @@ impl WorkflowTaskManager {
             }
 
             // Send commands from lang into the machines
-            machine_mut!(self, run_id, task_token, |wfm: &mut WorkflowManager| {
+            machine_mut!(self, run_id, |wfm: &mut WorkflowManager| {
                 wfm.push_commands(commands).boxed()
             })?;
             // Check if the workflow run needs another activation and queue it up if there is one
             // by pushing it into the pending activations list
-            let next_activation = machine_mut!(
-                self,
-                run_id,
-                task_token,
-                move |mgr: &mut WorkflowManager| mgr.get_next_activation().boxed()
-            )?;
+            let next_activation = machine_mut!(self, run_id, move |mgr: &mut WorkflowManager| mgr
+                .get_next_activation()
+                .boxed())?;
             if !next_activation.jobs.is_empty() {
                 self.pending_activations.push(next_activation);
                 let _ = self.pending_activations_notifier.send(true);
             }
             // We want to fetch the outgoing commands only after any new activation has been queued,
             // as doing so may have altered the outgoing commands.
-            let server_cmds =
-                machine_mut!(self, run_id, task_token, |wfm: &mut WorkflowManager| {
-                    async move { Ok(wfm.get_server_commands()) }.boxed()
-                })?;
+            let server_cmds = machine_mut!(self, run_id, |wfm: &mut WorkflowManager| {
+                async move { Ok(wfm.get_server_commands()) }.boxed()
+            })?;
+            let is_query_playback = is_leg_query_task && query_responses.is_empty();
             // We only actually want to send commands back to the server if there are no more
-            // pending activations and we are caught up on replay.
-            if !self.pending_activations.has_pending(run_id) && !server_cmds.replaying {
+            // pending activations and we are caught up on replay. We don't want to complete a wft
+            // if we already saw the final event in the workflow, or if we are playing back for the
+            // express purpose of fulfilling a query
+            if !self.pending_activations.has_pending(run_id)
+                && !server_cmds.replaying
+                && !is_query_playback
+            {
                 Some(ServerCommandsWithWorkflowInfo {
                     task_token,
                     action: ActivationAction::WftComplete {
@@ -448,7 +450,7 @@ impl WorkflowTaskManager {
         } else {
             // Blow up any cached data associated with the workflow
             let should_report = self
-                .request_eviction(run_id, "Activation failed by lang")
+                .request_eviction(run_id, "Activation failed")
                 .map_or(true, |attempt| attempt <= 1);
             if should_report {
                 FailedActivationOutcome::Report(tt)
@@ -507,11 +509,7 @@ impl WorkflowTaskManager {
 
                 Ok((wft_info, activation))
             }
-            Err(source) => Err(WorkflowUpdateError {
-                source,
-                run_id,
-                task_token: Some(wft_info.task_token),
-            }),
+            Err(source) => Err(WorkflowUpdateError { source, run_id }),
         }
     }