As in the redis offical site the command provides the CLUSTER FLUSHSLOTS will remove all the slots in the current node. But i am wonder why other nodes can't sense this change . So it only work in the current node and after the slots flushed how the node.confg updated ? will it change ? Please help me to understand
Thanks
Comment From: Cloud-Zhang
Anyone can help to clarify this?
Comment From: itamarhaber
Hello @Cloud-Zhang
The CLUSTER FLUSHSLOTS makes the node "forget" all its assigned hash slots. This changes the node's state, which is persisted to the conf file as well as shared with the cluster's members, and leaves the respective slots unassigned (so you'll need to call CLUSTER ADDSLOTS for these).
Comment From: Cloud-Zhang
Hello Haber. Thanks for your reply. So is it means that when you execute this command in the one node which owns the slot , it will remove the assigned slots . ans in the same time other nodes should know this change and change the config file as well. Now i have a case which is i execute this command in one node . For example one node have slots 0-5460 . after CLUSTER FLUSHTSLOTS executed. these slots will not belong to this node. Then i go to another node whihc own the 5461-10922 slots. Now i will add the 0-5460 slots to it . BUt it will tell me the slots is busy. So it proves that other node will not effect even you execute the command in one node.
Comment From: itamarhaber
I apologize for misleading you somewhat, my cluster knowledge is very lacking. As far as I can tell, you'll need to issue DELNODES before the ADDNODES on the target node.
Comment From: Cloud-Zhang
@itamarhaber after mant times testing . i still found that when you run the cluster flushslots in the one node, other node still not know it . So it means this change will not propagate itself change to the whole cluster. Now if one node run this command and it's slot removed. But if i login into another node(master) to run the Cluster Info it still show the state is ok. But in fact the current cluster is not avilable. Since i do the config change as cluster-require-full-coverage yes. So it should all slots should coverage . that's the key point i raised why this change can't be propagate to the whole cluster.