CA Release Automation High Availability

[Anon Anon]

We are working on setting up Execution servers for CARA HA architecture. After preliminary research, we have shortlisted 2 methodologies which can best suit our architecture. However we need to understand the pros and cons of both these before we proceed. Please let us know the answers to the below:

 

1. Is it possible to setup supernodes in Ring Execution Server with HA architecture? This method is best suited for us as we are planning to setup 4 execution servers.
2. How do we setup supernodes for both Ring Execution Server HA and Star Execution HA? I found the following article on CA website but it only has information for setting up supernodes in a 2 server environment. https://communities.ca.com/thread/106466161
3. We are going to setup 4 execution servers with 2 servers on each of our data centers. If we go with either of the HA architectures,  should we have supernodes created within the same datacenter or should we create supernodes across the datacenters. For eg. Datacenter X has Execution servers X1 and X2 while Datacenter Y has Y1 and Y2. Should we have supernodes {X1,X2}, {Y1,Y2} or should we have supernodes {X1,Y1},{X2,Y2}? If we use the Ring Execution server method it is pretty clear {X1,X2},{Y1,Y2} approach is the best but would like to know more about the Star Execution server method.
4. Currently we are storing artifacts for some applications on our execution servers. How will creating supernodes affect this? How can we retrieve artifacts stored on execution servers when adopt any of the above mentioned HA architectures?

[Walter Guerrero]

Please download and review the CA Release Automation Architecture and Implementation Guide 5.5 - 6.x at the following link:

 

CA Release Automation 5.x / 6.x Architecture and Implementation Best Practice Guide

 

You can setup super nodes for a ring configuration, if you are going thru this step, why not setup a STAR configuration? That will give you more flexibility for future expansion if needed.

 

Now to setup the super node, please follow the instructions in the guide listed at the beginning of this response and review page 24, which explains what you have to do.

 

So each of the super nodes will contain just two RA agents each? If you create super nodes across data centers, you need to keep in mind the network latency. The description of the make up of the super nodes is very good and again based on the architecture that you decide to use, it will work on both.

 

Based on the information provided in the above guide, you can find additional information about how the artifacts are distributed to the NES. Release Automation KB: Artifact Distribution and Cleanup

 

Artifact NES Distribution

1. Artifacts are retrieved by retrieval agent and downloaded to <Retrieval_Agent_Root>/files/ART<Release_ID>
2. They are moved to <Retrieval_Agent_Root>\files_registry\<File_MD5>
   • This sub-stage is sometimes referred to as the "Retrieval Stage". While the files are being moved, there could be a point where the file exists on the agent at both locations at the same time (thus peaking at two times its size). After the move finishes, they exist only in 'files_registry'.
3. They are copied to the NES at <NES_Root>\artifact_store\releaseId_<Release_ID >
   • Each NES will request and get only the artifacts that are needed by the agents that are connected to it.
4. They are deleted from <Retrieval_Agent_Root>\files_registry\<File_MD5>.

Based on the information above, if you setup a super NES, the artifacts will be distributed to the NES servers that make up that super node.

 

Let me know, if you need any additional information about this.

 

Thanks

 

-Walter

[Keith Puzey]

Hi,

 

      The ring architecture is the best option in this scenario, a star architecture should only be used when there are a very large number of execution servers.  I would recommend that you create supernoded nes's in each data Center and the agents should not be connected across datacenters. I.e all agents in Datacenter x connected to the nes's X1 and X2.

 

     As walter mentioned a lot of these architectures are explained further in the architecture guide,

 

regards

 

keith

[Anon Anon]

Hi guewa01/ Keith-Puzey-CA

 

We have some follow up questions for you . Your answers did help us in our decision. But before proceeding we would like the below questions to be answered:

 

1. If we use the Ring Execution server HA architecture and we need to increase the number of execution servers, is it possible to add more servers to the architecture?
2. We have a deployment which uses agents assigned to supernodes on both datacenters X and Y. If datacenter X goes down, will the deployment still take place on the agents assigned to datacenter Y? The diagram provided in the document does not seem resilient in case of datacenter failure?
3. In the CARA Studio, we have agent groups. What are those agent groups and how do they work? How will the agent groups be affected when we create supernodes? What is the difference between agent group and supernodes?
4. How many agents can one execution server handle? What will happen if the execution server reaches its max capacity? How will the agents be handle in that situation?
5. The document says for load balancing we should go with a software load balancer. We are planning to use 3DNS. Is it something that is supported by CARA?

[Keith Puzey]

Hi,

 

1     We would not normally recommend adding more execution servers to the ring architecture, as your network grows one of the other architectures may be more practical.

2     If a whole datacenter goes dark the execution servers in the remaining data center will still function but the agents in the offline data center will not be accessible

3     The groups in the studio are purely for administration purposes and and have no impact on the artifact distribution.

4     Can you say which version of RA you are using this affects the scaling numbers,

5     I will confirm with engineering on this point.

 

     If you would like to arrange for a webex to discuss these point please email me so we can arrange this , [e-mail]

 

Regards

Keith

[Walter Guerrero]

Based on what Keith mentioned and your future plans. I would recommend that you setup a STAR configuration out of the box just to be on the safe side.

If I understand correctly, you have RA 5.5.2 installed right? Any additional patches installed?

 

About the 3-DNS, are you referring to the F5 Networks 3-DNS box? Based on what I have been finding, the use of that load balancer should not be a problem.

 

Thanks

 

-Walter

[Anon Anon]

Hi Walter

 

Yes I am referring to the F5 networks 3 DNS box.