table distribution in hana multi node

[Former Member]

Hi Team,

I have a query, as everyone knows tables are distributed to different nodes presuming no one is partitioned. Then if one of the node goes down and standby comes up during that time will we not have problems accessing that table since it wont be accessible since standby is taking over.

Thanks,

Message was edited by: Tom Flanagan

[Former Member · Accepted solution]

Lars/Richard,

What would be your take on this, how does hana handle it during the switch over there has to be some kind of glitch right.

[lbreddemann · Accepted solution]

Would be easier to notify me or anybody else, if you'd use the @<username> feature... ...

And sure enough: the stand-by host concept allows for fast recovery from a host failure - not for uninterrupted data access.

So, yes, whenever the host switch occurs, data will become temporarily unavailable.

- Lars

[Former Member · Accepted solution]

Lars,

Thanks a ton for the response.

Obvisouly you know me. I mailed you as well.

Thanks lars. so will short dumps occur if that particular tables are accessed during that time.

[Former Member]

Hi,

Regarding your question, I've made a test in SAP HANA SPS07 rev. 70 early this year. You can take it as a reference.

In my test environment, there were three hosts, one master, one slave and one standby as follows.

I created a very simple partitioned table named "STUDENT".

After the insertion, when you opened the definition of the table and switched to the "Runtime Information", you could find there were 2 records on the master and 1 record on the slave. There was no record on the standby.

At this time, when you made a select, everything was fine. You could see three records.

Then I stopped the slave.

At this time, I ran the select statement but failed. This is because there was one record on the slave and the slave was stopped (the standby was not started yet). So you get everything or you get nothing. In this case, you got nothing.

After the standby was started and took over the data which was on the slave, I could succeed to run the SQL again.

In short, if you use host-level HA which means you have only one SAP HANA system and there is(are) standby host(s) in your SAP HANA system, "seamless" is impossible. This is because of the "shared nothing" concept in SAP HANA distributed system.

Best regards,

Wenjun

[Former Member]

Wow that was some analysis, impressive analysis.

Wenjun so if the table is not available then if the application accesses it would it throw some kind of short dump during the failover.

Also is the failover is it dependant on the node size redo log size? Will it matter since a larger appliance will be much faster.

[Former Member]

Mr.Zhou:

Thanks for your efforts.

Would you please post the screenshots for following 2 scenarios?

1) stop both master and slave, then start standby, what would be the result form the selection?

2) stop only master but leave slave running, then start standby, what would be the result of the selection?

If you have no time, only scenario one is needed.

Best regards.