[TiDB Usage Environment] Production Environment
[TiDB Version] v6.1.1
[Encountered Issue]
After modifying a system configuration online, I executed tiup cluster reload xx. According to the official documentation, the cluster should still be available during this period, but the reality is that it cannot be connected.
btw, the IP that the program connects to is an internal load-balanced IP, with equal weights for each TiDB instance.
The reload process operates on each node individually, and it is possible that while this node is being processed, the load balancer directs traffic to this node.
That makes sense. So, how can we minimize service unavailability issues? I’m using Alibaba’s SLB and have already set the health check interval and threshold to the minimum, but I guess it’s still unavoidable?
The official statement says that the cluster remains available during reload, but isn’t that a bit of an empty promise? For connections like JDBC to TiDB nodes, can we specify a cluster mode? If we can only specify one IP to access the database, then isn’t it unavoidable?
During peak business hours, using tiup reload may cause jitter. It is recommended to perform it during off-peak hours. Additionally, some parameters can be modified online without restarting the cluster. You can refer to:
When reloading, you can use -R to restart a specific type of component or -N to restart a specific node component to minimize the impact on the cluster.
Well, when I reload, I use the -R operation, and there are limited TiDB configuration items that can be modified online. However, generally, changes are not made frequently. Thanks!
|
username: 海石花47
|
username: 
|
username: