Row store tables in HA environment

[chris_clarke]

Hi,

As I understand it, when a HANA database is started, all row store tables are preloaded into memory, while column store tables can either be configured to be preloaded or by default they are loaded on first access.

When a failure occurs, and a HA node takes over what happens to row store tables and access to them?  Is it queued until they are preloaded into memory or are they loaded on access?

My interest is in working out what outage could be expected in a worst case scenario.  Say there is a 300GB row store table, this could take a significant amount of time to load into memory and I don't know if it will still be accessible while this load is taking place.

Does anyone out there know?

Chris

[SK-EA]

Chris,

Here are my views.

In a fail over scenario, the server N+1 (standby) reads indexes from the shared storage and connects to the logical components of failed server. Hence I believe there wont be any explicit loading of system tables into memory again.

Also row tables are meant for tables having minimal records. So, they shouldn't be consuming huge space.

Will try to come back with more details. Hope it helps for now.

-Srinivas K.

[chris_clarke]

Thanks for your input.

I know the main function of HANA is supposed to be around Column based storage, but when people start to use HANA for ERP or other more transactional systems I assumed the row tables would be used.  Maybe my assumption is wrong and column store tables will be used.

Referring to my previous question, I guess doubt is put into my mind as I don't know why row tables have to be pre-loaded, whereas column tables don't.  So if they have to be in memory when the node starts, don't they have to be in memory when the HA node takes over?  If it's possible for these tables to be loaded on demand on the HA node after a failover, why is it not possible to do this in normal operation on system startup?

Thanks again for your comment

Chris

[former_member184768]

Hi Chris,

I haven't had a chance to see the ERP on HANA instance yet, but I think the data would still be in Column Store for that. Typically Row stores are system tables which are absolute necessary for the application to start functioning (like BW or ERP). The application data would be in column format to gain the advantage of compression.

So the system tables which are not very big in size should be easily pre-loaded to the memory after the restart.

I am not much an expert on HA, but I think the data is replicated across the nodes for the failover node also to take over without issues.

Regards,

Ravi