#7037 fix: Track ResourceVersion in TaskAction watcher to prevent duplicate ADDED events Pull request opened by majiayu000 Fixes #6988 Changes Track ResourceVersion across watch reconnects to prevent duplicate ADDED events when the Kubernetes watch connection drops and re-establishes. Root cause:

doWatch()

creates watches without passing ResourceVersion, so the API server replays all existing resources as ADDED events on reconnect. Fix: 1. Added

lastResourceVersion

field to

ActionsClient

struct 2. In `doWatch()`: read

lastResourceVersion

under RLock and pass it in

metav1.ListOptions.ResourceVersion

when creating the watch 3. In `handleWatchEvent()`: update

lastResourceVersion

from

taskAction.ResourceVersion

under write Lock 4. Handle

watch.Error

events: on 410 Gone, reset

lastResourceVersion

to

""

(triggers fresh list); on other errors, preserve version for retry Thread safety uses the existing

mu sync.RWMutex

. Test Plan • `TestHandleWatchEvent_UpdatesResourceVersion`: verify ResourceVersion is extracted and stored from watch events • `TestDoWatch_PassesResourceVersionOnReconnect`: verify stored version is passed in ListOptions on reconnect • `TestDoWatch_HandlesGoneError`: verify 410 Gone resets to empty string • `TestDoWatch_PreservesResourceVersionOnOtherErrors`: verify non-Gone errors preserve version All tests pass with

-race

flag. •

main

• #6583 • fix: Track ResourceVersion in TaskAction watcher to prevent duplicate ADDED events 👈 flyteorg/flyte