From 6d10fb1cc026d0ba09793643d35e33e2bf2d58f9 Mon Sep 17 00:00:00 2001
From: Anirudh Oppiliappan <anirudh@tangled.sh>
Date: Mon, 23 Jun 2025 15:52:00 +0300
Subject: [PATCH] spindle/{db,engine}: emit StatusTimeout events

Signed-off-by: Anirudh Oppiliappan <anirudh@tangled.sh>
---
 spindle/db/events.go     |  4 ++++
 spindle/engine/engine.go | 22 +++++++++++++++-------
 spindle/engine/errors.go |  9 +++++++++
 3 files changed, 28 insertions(+), 7 deletions(-)
 create mode 100644 spindle/engine/errors.go

diff --git a/spindle/db/events.go b/spindle/db/events.go
index 7a59759d..567e1d85 100644
--- a/spindle/db/events.go
+++ b/spindle/db/events.go
@@ -135,3 +135,7 @@ func (d *DB) StatusFailed(workflowId models.WorkflowId, workflowError string, ex
 func (d *DB) StatusSuccess(workflowId models.WorkflowId, n *notifier.Notifier) error {
 	return d.createStatusEvent(workflowId, models.StatusKindSuccess, nil, nil, n)
 }
+
+func (d *DB) StatusTimeout(workflowId models.WorkflowId, n *notifier.Notifier) error {
+	return d.createStatusEvent(workflowId, models.StatusKindTimeout, nil, nil, n)
+}
diff --git a/spindle/engine/engine.go b/spindle/engine/engine.go
index 7a7a3739..594b9d9c 100644
--- a/spindle/engine/engine.go
+++ b/spindle/engine/engine.go
@@ -104,11 +104,16 @@ func (e *Engine) StartWorkflows(ctx context.Context, pipeline *models.Pipeline,
 
 			err = e.StartSteps(ctx, w.Steps, wid, w.Image)
 			if err != nil {
-				e.l.Error("workflow failed!", "wid", wid.String(), "error", err.Error())
-
-				dbErr := e.db.StatusFailed(wid, err.Error(), -1, e.n)
-				if dbErr != nil {
-					return dbErr
+				if errors.Is(err, ErrTimedOut) {
+					dbErr := e.db.StatusTimeout(wid, e.n)
+					if dbErr != nil {
+						return dbErr
+					}
+				} else {
+					dbErr := e.db.StatusFailed(wid, err.Error(), -1, e.n)
+					if dbErr != nil {
+						return dbErr
+					}
 				}
 
 				return fmt.Errorf("starting steps image: %w", err)
@@ -245,7 +250,7 @@ func (e *Engine) StartSteps(ctx context.Context, steps []models.Step, wid models
 			<-tailDone
 
 			stepCancel()
-			return fmt.Errorf("step timed out after %v", stepTimeout)
+			return ErrTimedOut
 		}
 
 		if waitErr != nil {
@@ -259,7 +264,10 @@ func (e *Engine) StartSteps(ctx context.Context, steps []models.Step, wid models
 
 		if state.ExitCode != 0 {
 			e.l.Error("workflow failed!", "workflow_id", wid.String(), "error", state.Error, "exit_code", state.ExitCode, "oom_killed", state.OOMKilled)
-			return fmt.Errorf("error: %s, exit code: %d, oom: %t", state.Error, state.ExitCode, state.OOMKilled)
+			if state.OOMKilled {
+				return ErrOOMKilled
+			}
+			return ErrWorkflowFailed
 		}
 	}
 
diff --git a/spindle/engine/errors.go b/spindle/engine/errors.go
new file mode 100644
index 00000000..47b052a6
--- /dev/null
+++ b/spindle/engine/errors.go
@@ -0,0 +1,9 @@
+package engine
+
+import "errors"
+
+var (
+	ErrOOMKilled      = errors.New("oom killed")
+	ErrTimedOut       = errors.New("timed out")
+	ErrWorkflowFailed = errors.New("workflow failed")
+)
-- 
2.51.2