fix(admin): replace useEffect restart poll with useQuery pattern in KiloclawInstanceDetail#1881
fix(admin): replace useEffect restart poll with useQuery pattern in KiloclawInstanceDetail#1881pandemicsyn wants to merge 1 commit intomainfrom
Conversation
…iloclawInstanceDetail Replace the useEffect-based polling (which modified the main data query's refetchInterval and used a useRef to track previous status) with a dedicated useQuery polling hook, matching the VolumeReassociatePanel pattern. The new approach: - Adds a separate 'machine-restart-poll' useQuery that invalidates the get query every 3 seconds while isRestartPending (awaitingRestartCompletion) - Uses a render-time check (not useEffect) to detect when the machine returns to 'running' and re-invalidate controllerVersion/gatewayStatus - Removes awaitingRestartCompletion from the main data query's refetchInterval This fixes the race condition where controllerVersion was refetched while the machine was still mid-restart, caching a stale value for the full 5-minute staleTime.
![kilo-code-bot[bot]](https://avatars.githubusercontent.com/in/2193792?s=60&v=4){kind=small}
| }), | ||
| }); | ||
| } | ||
| if (awaitingRestartCompletion && data?.workerStatus?.status === 'running') { |
There was a problem hiding this comment.
WARNING: This can stop polling before the restart actually begins
awaitingRestartCompletion is set to true in the mutation onSuccess handler before the invalidated get query has a chance to observe the machine leaving running. On the first rerender after that state update, data?.workerStatus?.status is often still the pre-restart cached value (running), so this branch clears awaitingRestartCompletion immediately and re-invalidates controllerVersion/gatewayStatus against the old machine state. The previous prevMachineStatus guard avoided this false positive; this version needs to wait until the status has been observed as non-running at least once before treating running as restart completion.
Code Review SummaryStatus: 1 Issue Found | Recommendation: Address before merge Overview
Issue Details (click to expand)WARNING
Other Observations (not in diff)No additional issues found outside the diff. Files Reviewed (1 files)
Reviewed by gpt-5.4-20260305 · 133,002 tokens |
1 participant
Summary
Replaced the
useEffect-based polling approach for machine restart completion detection with a dedicateduseQuerypolling hook, matching the existingVolumeReassociatePanelpattern already used in the same file.Changes in
KiloclawInstanceDetail.tsx:awaitingRestartCompletionfrom the main data query'srefetchInterval(onlyawaitingRestoreCompletiondrives it now)'machine-restart-poll'useQuerythat polls every 3s whileawaitingRestartCompletionis true, invalidating thegetquery on each tickuseEffect+useRef(prevMachineStatus) approach with a render-time check: whenawaitingRestartCompletion && data?.workerStatus?.status === 'running', sets state to false and re-invalidatescontrollerVersionandgatewayStatusThis ensures
controllerVersionis only refetched after the machine is confirmed running, avoiding the race condition where a stale value could be cached for the full 5-minutestaleTime.Verification
VolumeReassociatePanelpolling pattern at lines 412–431 of the same file.setStatepattern is consistent with existing codebase usage.Visual Changes
N/A
Reviewer Notes
The original
prevMachineStatus/wasRestartingguard prevented false positives on first render when the machine was alreadyrunning. That guard is safely removed becauseawaitingRestartCompletionstarts asfalseand is only set totrueon mutation success — so the condition cannot fire spuriously on mount.The poll query key includes
awaitingRestartCompletionas a dependency (consistent withVolumeReassociatePanel's pattern), which causes the query to be remounted when polling stops. This is harmless and follows the established pattern.