You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi,
Currently when we call rebalancing of any kind, it starts instantly and continues until it finishes.
But in reality, clusters have some times which are at their peak load and some times at their lowest.
I think this is a very useful feature to do rebalancing when the load is at its lowest, so it affects the cluster/producer/consumers the least.
This can be achieved in two methods: partial rebalancing and scheduled rebalancing.
In former, when we call rebalancing, we can define some thresholds that when the load is at its lowest, it does a batch of rebalancing. This can also be done automatically in a way that CC looks at historical data, e.g. when we define a 1 week of windows, and finds times with lowest peak and then does its job.
In latter, e.g. we want to do a rebalancing at night because we know the load is lower, but it might be a crucial cluster that we're not allowed to have access during midnights, so we can schedule a time to start rebalancing and it doesn't start instantly.
Though I know that it's possible to stop execution to mimic partial rebalancing, and write a job to curl CC at our desired time to mimic the scheduling method, I believe these two features makes CC much more desirable and standalone.
Thanks.
The text was updated successfully, but these errors were encountered:
Hi,
Currently when we call rebalancing of any kind, it starts instantly and continues until it finishes.
But in reality, clusters have some times which are at their peak load and some times at their lowest.
I think this is a very useful feature to do rebalancing when the load is at its lowest, so it affects the cluster/producer/consumers the least.
This can be achieved in two methods: partial rebalancing and scheduled rebalancing.
Though I know that it's possible to stop execution to mimic partial rebalancing, and write a job to curl CC at our desired time to mimic the scheduling method, I believe these two features makes CC much more desirable and standalone.
Thanks.
The text was updated successfully, but these errors were encountered: