High Availability PA server

[JamiseBondi]

Hi guys,

I've seen a post by IBM describing a high availability PAW application layer cluster where a distributed architecture of web servers is used to spread the load behind a load balancer. (https://na.eventscloud.com/file_uploads ... eAzizi.pdf)
What I haven't yet come across is an architecture where the server running the databases/instances is clustered, such that if the primary server, either at the operating system or the Planning Analytics software level, had a hissy fit, that a standby PA server running these same instances could take over. Same concept as the web servers in that there is some redundancy between the web servers, but for a different role: the database servers. This could either be an active-active cluster or active-passive cluster, just some more redundancy that we currently have in this architecture. Has anyone installed the PA database servers / TM1 servers on a cluster or similar configuration for the purposes of high availability?

Thanks all.

[orlando]

Hi,

i don't think that's possible with the current TM1 technology. The data is stored in RAM. So the "main server" would be permanently busy "somehow" exporting the data to other servers. But when data is read, no data can be written in the cell at that moment and vice versa.
This data exchange can be done at night, but during "working hours" I don't see any chance.

Best regards,
orlando

[lotsaram]

Yes with Rest API delta queries it's possible. After you initialize a monitoring/mirroring service each subsequent query can just retrieve cells with changed values since the last delta request and push these to the hot synced instance.

For some customers requiring high availability on failover we have this set up using tm1py based apps. Syncing of metadata is also possible so that any new elements and changes in subsets are also picked up (but this isn't delta based so needs a bit of care and thought how to set it up and what exactly needs to be mirrored).

[moby91]

What I haven't yet come across is an architecture where the server running the databases/instances is clustered, such that if the primary server, either at the operating system or the Planning Analytics software level, had a hissy fit, that a standby PA server running these same instances could take over. ... This could either be an active-active cluster or active-passive cluster, just some more redundancy that we currently have in this architecture. Has anyone installed the PA database servers / TM1 servers on a cluster or similar configuration for the purposes of high availability?

You may look into

https://www.ykud.com/blog/cognos/tm1-hi ... balancing/
TM1 High Availability or Load Balancing


There used to be an IBM document on how to install and run the TM1 server software in a Microsoft Wolfpack cluster aka Microsoft Cluster Server (MSCS), for instance mentioned by

https://jwakefield.blogspot.com/2011/03 ... -load.html
Does TM1 support clustering /load balancing?

but it seems it is no longer online. You may ask your IBM contacts for a copy of this document. Although the TM1 server software is not cluster-aware it can use certain cluster functionalities to enable an automatic cold failover:
Imagine there are two cluster nodes, the primary node A and the secondary node B. On both the TM1 server software is installed. The instance's database directory is on a shared NAS. The TM1 instance on the primary node A is up&running. It is stopped on the secondary node B. If the TM1 instance on the primary node A fails, the cluster software will automatically start the TM1 instance on the secondary node B. As the TM1 instances on the primary node A and secondary node B share the same database directory, the newly started TM1 instance on the secondary node B will have all the new data gathered by the failed TM1 instance on the primary node A.

[ykud]

Yes with Rest API delta queries it's possible. After you initialize a monitoring/mirroring service each subsequent query can just retrieve cells with changed values since the last delta request and push these to the hot synced instance.

For some customers requiring high availability on failover we have this set up using tm1py based apps.

Just out of curiosity: do you use a queue in the middle to capture transactions from source to target or the tm1py app itself is buffering the transactions? I.e. if your target server is down (or unavailable due to network partition), will something capture the transactions happening in this period? And is there a mechanism to check that each transaction has arrived on target, or you just do full restore / reset every now and then?
Transaction logging needs to be enabled on every cube for this to work so all the 'disable log and load a lot of data' TIs need be adjusted?

Thanks,
Y