MaxMrgOutROSSizeMB changes suggested by VAdvisor

Tim_1

Hi,
Just had a scrutinize analyzed by VAdvisor and one of the recommendations was setting MaxMrgOutROSSizeMB to 20480 to reduce the time taken for mergeout.

Once this parameter is set the documentation implies that a "rebalance" has to be run for this to be implemented, is that correct?

Any other ways of triggering ROS containers over 20GB to be split?

Regards
Tim

Bryan_H

What is the current setting for MaxMrgOutROSSizeMB? Vertica won't split containers automatically, instead this setting should only prevent new containers from exceeding the size. Rebalance_cluster will fix container sizes to fall below MaxMrgOutROSSizeMB but this can be time- and resource-intensive since it will try to run in parallel across all containers. Instead, you can identify projections that have containers exceeding 20GB in storage_containers, then run REBALANCE_TABLE only on the tables that have containers needing split.

Tim_1

Current settings is default.
Thanks for the update, we have a number of tables with Hierarchical Partitioning, some of the old ROS containers are 100+GB in size. Occasionally they do insert data into these partitions that does cause lengthy MergeOuts running for hours. This has a knock on impact to the performance of the cluster, so need to mange when the ROS containers get rebuilt.