Portal 4.2.x High Availability - 2 or 3 servers at minimum?

[Welington Strutz]

Hi everyone, I´ve looked on other community topics/questions (such as Why does the API Developer Portal have such high hardware requirements? ) as well as the documentation, and I would like to understand the requirements for having a High Availability architeture for the Portal.

 

Looking on this page from documentation ( High Availability - CA API Developer Portal - 4.2 - CA Technologies Documentation ) , including its diagram, I would realize that a minimum of 2 servers (plus external Database and a Load Balancer) would be enough to define an HA environment.

 

However, that other page ( Deployment Topology - CA API Developer Portal - 4.2 - CA Technologies Documentation ) it leads to realize that only a 3-node (servers?) scenario is the minimum set for an HA, Production environment.

 

Probably I am missing something in concept, nevertheless even CA experienced folks might get confused on this, without mentioning the risk of either have fewer production licenses than needed for a "starter" HA-enabled API Portal setup (if 3 is the answer) or over the customer budget (if 2 is the right one).

 

Can we have a 2-server-HA scenario in CA API Developer Portal?

 

Thanks in advance. Best,

Welington Strutz

[Cornelia Postma]

Hello Wellington,

 

The 2 pages, that you are mentioning, are about two different things.

The first has 2 different portal setup. One docker swarm is setup in Datacenter 1 and the other is setup in Datacenter 2. Both portal swarms have been setup identically with the same configuration, and the same certificates and keys.

You will need to setup your own load balancer to route the traffic to these 2 instances.

 

The second page is about the setup within one single portal installation. That is setup on different nodes (machines) using the docker swarm cluster. For this you will need to have 3 machines at least. One will be the so called swarm manager, and the other 2 would be the workers. 

 

Basically the setup you see at the second page is the internal setup you would have for one of the DataCenters in the diagram of the first page.

 

 

Hope this helps along.

 

[Welington Strutz]

Thank you Conny for the asnwer... May I ask you for a bit more detail?

 

Thinking on a single site in which we want HA for Portal (the second page). If I understood correctly, I need, at least, three nodes (machines), being 2 of them acting as Swarm workers, and one as Swarm manager.

 

Does it require three CA API Developer Portal licenses to define this scenario?

Does the Swarm manager machine run any specific CA software on it?

Does the Swarm manager machine require the use of the CentOS hardened image (and their hw requirements)? ( CentOS Hardened Image - CA API Developer Portal - 4.2 - CA Technologies Documentation )

 

Thanks again for your contribution! Best,

Welington

[Cornelia Postma]

Hi Wellington,

 

No, this would just make use of one portal license.

All three machines can (but don't need to be) the use of the CentOS hardened image.

You would first install the portal and create a swarm cluster on the Manager node (see Create a Swarm Cluster - CA API Developer Portal - 4.2 - CA Technologies Documentation ) 

After that you would use the horizontal scaling (Horizontal Scaling - CA API Developer Portal - 4.2 - CA Technologies Documentation ) to add the additional worker nodes.

[panka03]

Dear Conny,

 

Will the three 3 nodes in a swarm (single data center) deployment provide high availability? From my limited testing, I found out that when the swarm manager node is down, the entire swarm cluster is not working fine.

 

Only the 2 data center approach seems to provide HA to an extent. For which, the customer must purchase 2 licenses and also must have their own mysql enterprise database license (Only InnoDB). Even in this scenario, the Jarvis data is not propagated among the 2 clusters since the analytics data is still stored in postgres and not externalized to mysql. There is no way today to replicate postgres between these 2 clusters. So, in my opinion, none of the methods provide 100% HA with all portal data & analytics.

 

Please correct me if I am missing something here.

 

Regards,

Karthik

[Dirk Bleyenberg]

Hi 

I believe your questions has been answered, I will mark this as the correct answer.
When your question is not answered or you still have additional questions please let us know.
With Kind Regards

Dirk