fix(metrics): count static markExecutionAsFailed terminal failures and omit unknown durations

Author: TheodoreSpeaks

apps/sim/lib/logs/execution/logging-session.test.ts

@@ -627,6 +627,7 @@ describe('completeWithError cancelled-status guard', () => {
 describe('LoggingSession.markExecutionAsFailed workflowId scoping', () => {
   beforeEach(() => {
     vi.clearAllMocks()
+    dbMocks.selectLimit.mockResolvedValue([{ status: 'running', trigger: 'api' }])
     dbMocks.updateWhere.mockResolvedValue(undefined)
   })
 
@@ -674,7 +675,7 @@ describe('LoggingSession workflow metrics', () => {
       loops: {},
       parallels: {},
     })
-    dbMocks.selectLimit.mockResolvedValue([{ status: 'running' }])
+    dbMocks.selectLimit.mockResolvedValue([{ status: 'running', trigger: 'api' }])
     dbMocks.execute.mockResolvedValue(undefined)
   })
 
@@ -745,6 +746,8 @@ describe('LoggingSession workflow metrics', () => {
   it('does not double-emit when markAsFailed runs after a completed session', async () => {
     const session = new LoggingSession('wf-1', 'exec-1', 'api', 'req-1')
     await session.complete({ totalDurationMs: 500 })
+
+    dbMocks.selectLimit.mockResolvedValue([{ status: 'completed', trigger: 'api' }])
     await session.markAsFailed('timeout')
 
     expect(recordExecutionCompletedMock).toHaveBeenCalledTimes(1)
@@ -769,6 +772,22 @@ describe('LoggingSession workflow metrics', () => {
     )
   })
 
+  it('static markExecutionAsFailed emits failed only for non-terminal rows', async () => {
+    dbMocks.selectLimit.mockResolvedValue([{ status: 'running', trigger: 'webhook' }])
+    await LoggingSession.markExecutionAsFailed('exec-1', 'crash', undefined, 'wf-1')
+
+    expect(recordExecutionCompletedMock).toHaveBeenCalledTimes(1)
+    expect(recordExecutionCompletedMock).toHaveBeenCalledWith({
+      trigger: 'webhook',
+      status: 'failed',
+    })
+
+    recordExecutionCompletedMock.mockClear()
+    dbMocks.selectLimit.mockResolvedValue([{ status: 'failed', trigger: 'webhook' }])
+    await LoggingSession.markExecutionAsFailed('exec-1', 'crash again', undefined, 'wf-1')
+    expect(recordExecutionCompletedMock).not.toHaveBeenCalled()
+  })
+
   it('skips the completion metric when the run was already cancelled elsewhere', async () => {
     dbMocks.selectLimit.mockResolvedValue([{ status: 'cancelled' }])
     const session = new LoggingSession('wf-1', 'exec-1', 'api', 'req-1')

apps/sim/lib/logs/execution/logging-session.ts

@@ -519,7 +519,10 @@ export class LoggingSession {
       })
 
       this.completed = true
-      this.emitExecutionCompletedMetric('failed', Math.max(1, durationMs))
+      this.emitExecutionCompletedMetric(
+        'failed',
+        typeof totalDurationMs === 'number' ? Math.max(1, totalDurationMs) : undefined
+      )
 
       try {
         const { PlatformEvents, createOTelSpansForWorkflowExecution } = await import(
@@ -613,7 +616,10 @@ export class LoggingSession {
       })
 
       this.completed = true
-      this.emitExecutionCompletedMetric('cancelled', Math.max(1, durationMs))
+      this.emitExecutionCompletedMetric(
+        'cancelled',
+        typeof totalDurationMs === 'number' ? Math.max(1, totalDurationMs) : undefined
+      )
 
       try {
         const { PlatformEvents, createOTelSpansForWorkflowExecution } = await import(
@@ -992,7 +998,10 @@ export class LoggingSession {
       this.requestId,
       this.workflowId
     )
-    this.emitExecutionCompletedMetric('failed')
+    // The static helper emits the failure metric when the row transitions from
+    // a non-terminal status; mark this session as emitted either way so a later
+    // in-process completion attempt can't add a second point.
+    this.completionMetricEmitted = true
   }
 
   static async markExecutionAsFailed(
@@ -1003,6 +1012,21 @@ export class LoggingSession {
   ): Promise<void> {
     try {
       const message = errorMessage || 'Run failed'
+      const current = await db
+        .select({
+          status: workflowExecutionLogs.status,
+          trigger: workflowExecutionLogs.trigger,
+        })
+        .from(workflowExecutionLogs)
+        .where(
+          and(
+            eq(workflowExecutionLogs.executionId, executionId),
+            eq(workflowExecutionLogs.workflowId, workflowId)
+          )
+        )
+        .limit(1)
+        .then((rows) => rows[0])
+
       await db
         .update(workflowExecutionLogs)
         .set({
@@ -1029,6 +1053,12 @@ export class LoggingSession {
           )
         )
 
+      // Only a transition from a non-terminal status is a new terminal outcome;
+      // rows already completed/failed/cancelled emitted their point elsewhere.
+      if (current && (current.status === 'running' || current.status === 'pending')) {
+        workflowMetrics.recordExecutionCompleted({ trigger: current.trigger, status: 'failed' })
+      }
+
       logger.info(`[${requestId || 'unknown'}] Marked execution ${executionId} as failed`)
     } catch (error) {
       logger.error(`Failed to mark execution ${executionId} as failed:`, {