How Data Rebalance operate?

Hi

Here are some posts to help us understanding rebalance and finding out answers of your questions:
https://www.vertica.com/kb/PrepareYourClusterforRebalance/Content/BestPractices/PrepareYourClusterforRebalance.htm
https://www.vertica.com/kb/Understanding-Rebalancing-Part-1-What-Happens-During-Rebalancing/Content/BestPractices/Understanding-Rebalancing-Part-1-What-Happens-During-Rebalancing.htm
https://www.vertica.com/kb/Understanding-Rebalancing-Part-2-Optimizing-for-Rebalancing/Content/BestPractices/Understanding-Rebalancing-Part-2-Optimizing-for-Rebalancing.htm

In general, it's better adding all nodes at one, because adding fewer nodes need more time. And as there are locks when table is rebalancing, DML and DDL operators will be blocked until it finished or failed.

To save time for splitting data when rebalancing, we can leverage local resegment to split ROS containers before really adding nodes. The process is :
1. prepare as suggestions from upper posts.
2. turn on local segment.
SELECT ENABLE_LOCAL_SEGMENTS();
SELECT SET_SCALING_FACTOR (11);
3. force spliting data. If business is busy, the 2nd optional paramenter can help us split tables one by one.
SELECT DO_TM_TASK('mergeoutsingleros' [, ':tablename']);
4. really adding 4 new nodes and reblancing according to our documents. It's better to stop DDLs/DMLs operations at the step.
5. turn off local segment after No.4 finished.
SELECT DISABLE_LOCAL_SEGMENTS();
SELECT SET_SCALING_FACTOR(4);

Hope those can help you.
DQ

On 2018/11/23, 17:17, "mkheir" wrote:

Vertica Forum https://forum.vertica.com/
mkheir started a new discussion: How Data Rebalance operate?

Hello team,

We are preparing to add more nodes to a cluster, the cluster actually counts 7 nodes and want to extend it to 11 nodes.

I have the following questions:

- Is it faster to add 4 nodes at once then rebalance, or add two nodes then rebalance , followed by adding two other nodes then rebalance?

- Are there any operations (DML,DDL?) that need to be avoided or should not run during the rebalance?

- The cluster have 18TB of data, does it mean that >36TB (K-Safety=1) will be moved? details on how data will be moved is highly appreciated.

Note:

=> SELECT scaling_factor, is_enabled, maximum_skew_percent FROM ELASTIC_CLUSTER;

scaling_factor | is_enabled | maximum_skew_percent

Many thanks Dingqiang,

Very useful inputs, I will go throw them in details and prepare a plan following your recommendations.

Regards.