From 875125b1cad4efdbf4636e73bc82d752b1381616 Mon Sep 17 00:00:00 2001
From: Pranay Prakash <pranay.gp@gmail.com>
Date: Wed, 12 Nov 2025 20:53:02 -0500
Subject: [PATCH 1/2] Fix workflow failing because of function timeouts

---
 .changeset/cyan-ducks-wonder.md |  5 +++++
 packages/core/src/runtime.ts    | 30 ++++++++++++++++++++----------
 2 files changed, 25 insertions(+), 10 deletions(-)
 create mode 100644 .changeset/cyan-ducks-wonder.md

diff --git a/.changeset/cyan-ducks-wonder.md b/.changeset/cyan-ducks-wonder.md
new file mode 100644
index 000000000..123897616
--- /dev/null
+++ b/.changeset/cyan-ducks-wonder.md
@@ -0,0 +1,5 @@
+---
+"@workflow/core": patch
+---
+
+Allow step retrying if it fails without proper cleanup
diff --git a/packages/core/src/runtime.ts b/packages/core/src/runtime.ts
index 069accbb3..ba20bc08b 100644
--- a/packages/core/src/runtime.ts
+++ b/packages/core/src/runtime.ts
@@ -640,12 +640,14 @@ export const stepEntrypoint =
           let result: unknown;
           const attempt = step.attempt + 1;
           try {
-            if (step.status !== 'pending') {
-              // We should only be running the step if it's pending
-              // (initial state, or state set on re-try), so the step has been
+            if (!['pending', 'running'].includes(step.status)) {
+              // We should only be running the step if it's either
+              // a) pending - initial state, or state set on re-try
+              // b) running - if a step fails mid-execution, like a function timeout
+              // so the step has been
               // invoked erroneously.
               console.error(
-                `[Workflows] "${workflowRunId}" - Step invoked erroneously, expected status "pending", got "${step.status}" instead, skipping execution`
+                `[Workflows] "${workflowRunId}" - Step invoked erroneously, expected status "pending" or "running", got "${step.status}" instead, skipping execution`
               );
               span?.setAttributes({
                 ...Attribute.StepSkipped(true),
@@ -698,11 +700,24 @@ export const stepEntrypoint =
               () => stepFn(...args)
             );
 
+            // NOTE: None of the code from this point is guaranteed to run
+            // Since the step might fail or cause a function timeout and the process might be SIGKILL'd
+            // The workflow runtime must be resilient to the below code not executing on a failed step
+
             result = dehydrateStepReturnValue(result, ops);
 
             waitUntil(Promise.all(ops));
 
-            // Update the event log with the step result
+            // Mark the step as completed first. This order is important. If a concurrent
+            // execution marked the step as complete, this request should throw, and
+            // this prevent the step_completed event in the event log
+            // TODO: this should really be atomic and handled by the world
+            await world.steps.update(workflowRunId, stepId, {
+              status: 'completed',
+              output: result as Serializable,
+            });
+
+            // Then, append the event log with the step result
             await world.events.create(workflowRunId, {
               eventType: 'step_completed',
               correlationId: stepId,
@@ -711,11 +726,6 @@ export const stepEntrypoint =
               },
             });
 
-            await world.steps.update(workflowRunId, stepId, {
-              status: 'completed',
-              output: result as Serializable,
-            });
-
             span?.setAttributes({
               ...Attribute.StepStatus('completed'),
               ...Attribute.StepResultType(typeof result),

From a5f6eee077cd249b38ab17ce45efd97ae4a5b498 Mon Sep 17 00:00:00 2001
From: Pranay Prakash <pranay.gp@gmail.com>
Date: Sun, 16 Nov 2025 12:03:39 -0800
Subject: [PATCH 2/2] Update packages/core/src/runtime.ts

Co-authored-by: Peter Wielander <mittgfu@gmail.com>
---
 packages/core/src/runtime.ts | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/packages/core/src/runtime.ts b/packages/core/src/runtime.ts
index ba20bc08b..f9200535c 100644
--- a/packages/core/src/runtime.ts
+++ b/packages/core/src/runtime.ts
@@ -644,8 +644,7 @@ export const stepEntrypoint =
               // We should only be running the step if it's either
               // a) pending - initial state, or state set on re-try
               // b) running - if a step fails mid-execution, like a function timeout
-              // so the step has been
-              // invoked erroneously.
+              // otherwise, the step has been invoked erroneously
               console.error(
                 `[Workflows] "${workflowRunId}" - Step invoked erroneously, expected status "pending" or "running", got "${step.status}" instead, skipping execution`
               );