Purpose
The load balancing policy. load_based indicates balancing based on node load (such as disk usage), and replica_count_based indicates balancing based on the number of replicas.
Attributes
Attribute |
Description |
|---|---|
| Parameter type | String |
| Default value | "replica_count_based" |
| Value range | load_based / replica_count_based |
| Effective mode | dynamic |
Usage notes
load_based: Uses the load-based balancing policy. All nodes in the cluster must have valid (non-zero)disk_max_bandwidthanddisk_max_iopssettings.- When you attempt to set
load_basedat runtime, if any node has the preceding configuration set to 0, the change is rejected and the parameter value remains unchanged. - If a node's configuration is missing at runtime (for example, a newly added node is not configured, or its configuration is changed to 0), the cluster automatically switches to
replica_count_basedscheduling and periodically emits ERROR logs to remind you to complete the node configuration. The displayed parameter value remainsload_based. After the missing configuration is completed, the cluster automatically detects it and switches back toload_basedwithout manual intervention. - The current version does not provide an interface or monitoring metric to query the currently effective policy. You can detect the degraded state through periodic ERROR log alerts.
- When you attempt to set
replica_count_based: Uses the balancing policy based on the number of log stream replicas, without relying on node I/O capacity configuration.Configuration changes take effect immediately. Running balancing or scheduling tasks continue according to the old policy without interruption. A policy switch triggers a regular scheduling assessment. Whether to actually initiate a full rebalance is determined by the scheduler based on the new policy and is not mandatory.