fix(mothership): answer a browser run still running only once its log exists, and never discard a run that settled meanwhile

Author: waleedlatif1

apps/sim/lib/mothership/request/tools/workflow-client-fallback.ts

@@ -24,6 +24,7 @@ import {
   createStructuralWorkflowToolCompletionData,
   getWorkflowToolCompletionMessage,
 } from '@/lib/mothership/tools/workflow-tools'
+import { getWorkflowExecutionLogStatus } from '@/lib/workflows/executor/execution-state'
 import type { ResolvedSecretTraceRegistry } from '@/executor/utils/resolved-secret-trace-registry'
 
 const logger = createLogger('CopilotWorkflowClientFallback')
@@ -67,7 +68,8 @@ const CLIENT_CLAIM_POLL_MS = 5000
  * Once `waitTimeoutMs` has passed, settles a browser-run workflow call as running in the
  * background, the same detach a browser sends on `pagehide`: its waiter answers the call with
  * the run's ids and the browser's run goes on, bound to the call, so a Stop still cancels it.
- * A call no browser has claimed yet is re-read until one does or the wait ends.
+ * A call no browser has claimed yet, or whose run has no log yet, is re-read until it does or the
+ * wait ends.
  */
 async function detachClientRunAfterWait(params: {
   toolCallId: string
@@ -86,6 +88,11 @@ async function detachClientRunAfterWait(params: {
     if (!row || row.status !== ASYNC_TOOL_STATUS.running) return
     const executionId = getClaimedWorkflowExecutionId(row.claimedBy)
     if (!executionId || row.clientWorkflowExecutionId !== executionId) continue
+    // The claim precedes the run's log; answer running only once the run is readable by its id.
+    const logged = await getWorkflowExecutionLogStatus(executionId, workflowId).catch(
+      () => undefined
+    )
+    if (!logged) continue
     const status = ASYNC_TOOL_CONFIRMATION_STATUS.background
     await settleClientToolCall({
       toolCallId,

apps/sim/lib/mothership/request/tools/workflow-client-settlement.integration.ts

@@ -156,7 +156,7 @@ describe.runIf(Boolean(redisUrl))('settled client-claimed workflow tools', () =>
   async function executionLog(
     toolCallId: string,
     executionId: string,
-    status: 'completed' | 'failed' | 'cancelled' | 'pending'
+    status: 'completed' | 'failed' | 'cancelled' | 'pending' | 'running'
   ) {
     const snapshotId = generateId()
     snapshotIds.push(snapshotId)
@@ -496,6 +496,8 @@ describe.runIf(Boolean(redisUrl))('settled client-claimed workflow tools', () =>
 
   it("settles a browser run that outlasts the worker's wait as running, bound to the call", async () => {
     const { toolCallId, executionId } = await claimed()
+    // The execute route's log of the browser's run, still going: what the worker waits on.
+    await executionLog(toolCallId, executionId, 'running')
     const runOnServer = vi.fn()
 
     const outcome = await raceWorkflowToolClientPickup({

apps/sim/lib/mothership/tools/handlers/workflow/mutations.ts

@@ -301,6 +301,8 @@ async function answerWithinWait(
     while (!result && !(await runIsReadable(executionId, workflowId))) {
       result = await Promise.race([settled, waited(RUN_LOG_POLL_MS)])
     }
+    // A run that settled while its log was being read answers with its result.
+    result ??= await Promise.race([settled, Promise.resolve(null)])
     if (result) return result
   } finally {
     stopWaiting.abort()