GBase 8 a MPP Cluster in the Cluster a node short-term offline because of a hardware failure, cause a large amount of data inconsistency fevent, back when the fault node, inconsistent data synchronization is automatically; When the cluster business at high load, large amounts of table synchronization will aggravate resource usage and impact performance, there is a need to provide a way, you can adjust the data synchronization of parallelism,

CodePudding user response:

GBase 8 a MPP Cluster Cluster fevent synchronization by gc_recover scheduling and control components, the configuration parameters can be adjusted according to the following gc_recover components, realize the control of synchronous parallelism,

Recover_thread_num gc_recover. Conf configuration parameters regularly re-read function has finished the test, the results are as follows:

1, recover_thread_num parameters as the effective range of [1 ~ 50], this parameter is not configured, or configuration value is not in the range, restart the gcrecover will start 21 thread;

To 1, 2, recover_thread_num parameters configuration to restart gcrecover will start 12 threads, after can modify recover_thread_num parameters, only in the change will take effect, value is greater than the current value of basic will take effect within 10 s, the longest do not exceed 1 minute;

3, effective recover_thread_num parameters, the number of threads gcrecove recover_thread_num + 11

4, gcrecove minimum of 12, the number of threads that recover_thread_num=1

Gcrecove 5, 61, the maximum number of threads that recover_thread_num=50 when

6, the number of threads running gcrecover, can only increase, not decrease; Reduce the number of threads gcrecover only recover_thread_num configuration parameters, and restart the gcrecover