Hello,
I have one question mainly to know if the kind of set up I could test was supported by SAP or not.
I have a system in which HDB 3tier system replication is set up.
Node 1 -> Node 2 ( sync )
Node2 -> Node 3 ( async )
Node2 is here in order to provide some HANA "pseudo" HA.
I say pseudo because in this landscape, there is no clustering involved...
Node 3 is here for DR. No clustering involved here neither.
An SAP abap application server is running on top of this, secured by VMWare HA ( for HA ) and Veeam system replication ( for DR ).
Doing some tests, I changed the default hdbuserstore key from :
KEY DEFAULT
ENV : Node1:30041
USER: SAPABAP1
To :
KEY DEFAULT
ENV : Node1:30041,Node2:30041,Node3:30041
USER: SAPABAP1
As if I was scaled out , but I'm not scaled out. This is only here in order not to reconfigure the ABAP application server in case any hana node fails.
Testing this set up I could see that things seem to be working fine.
My test was :
- Connect to Node 1
- HDB Stop
- Connect to Node 2
- hdbnsutil -sr_takeover
I checked SM21 on the ABAP system and could see the workprocesses going in and then out of reconnect mode.
In ST04, I see I'm connected to the Node2 db host.
I also checked the persistence location seen at HANA level and it went from Node1 to Node2 :
select count(*) from M_TABLE_PERSISTENCE_LOCATIONS
where
schema_name='SAPABAP1' and PERSISTENCE_HOST='Node1'
COUNT(*)
0
select count(*) from M_TABLE_PERSISTENCE_LOCATIONS
where
schema_name='SAPABAP1' and PERSISTENCE_HOST='Node2'
COUNT(*)
98877
Replication between nodes 2 and 3 is still O.K.
So everything seems to be O.K.
The advantage I find in this set up is that I don't have to reconfigure the ABAP AS to point to the surviving HDB Node.
It avoids me to restart the AS.
Whenever one of the surviving Hana node is started up ( manually ) , The AS reconnects to it and that's it !
And this would be useful in this situation where, ( don't ask me why ... ) , i'm not allowed to use any clustering software for the hana hosts .
My question is : can I use this kind of setup and be supported ?
What would be the drawbacks you would see in this ?
One I can put on the table is split brain situation if Node 1 is set back in the game while Node 2 is also on.
Although the system sees the host holding the tables is Node2 I wonder if this would still stay consistent no matter the situation
Thanks & BR,
Steve.