Details
-
Bug
-
Status: Open (View Workflow)
-
Major
-
Resolution: Unresolved
-
None
-
None
-
None
Description
From Customer:
New cluster created
Server version: 10.5.6-4-MariaDB-enterprise-log
MaxScale version: 2.5.5
Columnstore_version: 5.4.3
It still takes a long time to load data, no change
We have always made all the tables columnstore tables.
With recent versions we have had the ability to make all of the tables copied from the UI database InnoDB tables, and only make the summary tables Columnstore tables. This would make this cluster more comparable to the setup in Redshift where the UI tables are all installed as reference tables, so they appear in their entirety on each node in the cluster. In that configuration the sample table provided loads in about 1 sec on Redshift, about 2 sec on CS 1.1.6 using the columnstore engine and far faster than X5's 23 sec using the columnstore engine.
If we were to do this, the CrossEngine Join feature would be used for multiple tables in almost every query of a table using the columnstore engine. Where as in the past the CrossEngine Join feature was only used occasionally.
What is your opinion on changing to this alternate table engine configuration?
Do you think using the CrossEngine Join feature on 8 to 18 tables (or more) in a single query would be problematic?
I take it all of those CrossEngine joins would be processed on pm1.
Is there any hope that X5 using the columnstore engine can get its load data times back down in the 2 sec range for the the sample table provided?
A side effect of this new configuration would be that the convert_tz function would be working to create the view on an InnoDB table instead of a Columnstore table.