When matching service restarts, it kills all long poll connections, and SDK would retry them, the new polling request would come almost at the same time, and they would timeout at 60s if there is no tasks. This would result in long poll request rate oscillating at interval of 60s. We could add some short jitter to the long poll timeout to make the requests smooth out faster.
When matching service restarts, it kills all long poll connections, and SDK would retry them, the new polling request would come almost at the same time, and they would timeout at 60s if there is no tasks. This would result in long poll request rate oscillating at interval of 60s. We could add some short jitter to the long poll timeout to make the requests smooth out faster.