HyperFlex Replication and Stretch Clusters

[Brooke Hammer]

Ask Me Anything Event

Welcome to the Cisco Support Community Ask Me Anything conversation. Have you been struggling in protecting your business applications and minimizing the risk of data loss and keep your business running smoothly? Do you need help in configuring and maintaining replication on hyperflex with HX native replication? Are you looking for assistance in understanding how the stretch cluster operates? Perhaps you would have some queries on the same. The focus of this session is to guide you on how to effectively use the replication feature on Hyperflex and learn how to avoid common pitfalls in understanding the way replication and working of stretch cluster compliments each other. Join our experts in this AMA session starting June 12th, 2023.

Note: Please post your post as a comment below no later than June 23, 2023.

Post your question below by clicking "Reply"

(Answers will be processed depending on the availability of the experts)
Don't forget to thank the expert by giving it a helpful vote!

---

Our experts

Pooja Raj is a Technical Consulting Engineer with Cisco HyperFlex solution team in Asia Pacific time zone. She has expertise in troubleshooting, managing, and deploying Cisco HyperFlex and Cisco UCS as a solution. She holds a bachelor’s degree in computer science and engineering.

Hemanth Anumala is responsible for technical consulting as a member of the Cisco HyperFlex solution team in the Asia Pacific time zone. His technical proficiency lies in the troubleshooting, diagnosing, and resolving complex technical issues related to HyperFlex & UCS . He also helps customers design, implement, and manage HyperFlex solutions. He holds a bachelor’s degree in computer science and engineering. CCNA and DEVNET are among the certifications that Hemanth holds

For more information, visit the Data Center discussion forums . If you have questions or are experiencing technical problems with any of the Cisco products, post your question, request information or use the Cisco Community before opening a case with the TAC.

[Rahil Abrol]

Hi All,

Is there a way to gracefully shutdown one site without affecting workload?

[poraj]

Hi Rahil,

Yes we can gracefully shutdown one site.

As per the official documents, if we have removed the VMware EAM dependency from the cluster (default in HX 4.0(2b) and later releases), then we can use HX maintenance mode in vCenter or through HX Connect to shutdown the controller VMs on each node in a site.

When this occurs, the site will shut down gracefully and the guest VMs will failover to the surviving site. We can also power down ESXi on the nodes if needed.

If EAM is enabled on your CVMs you can refer Appendix A in the below document. It has more detailed procedure on how to shutdown 1 site. The procedure is valid for both EAM enabled CVMs and newer systems with EAM disabled.

ref: https://www.cisco.com/c/dam/en/us/products/collateral/hyperconverged-infrastructure/hyperflex-hx-series/operating-hyperflex.pdf

Please let me know if you have any queries.

[Rahil Abrol]

Thanks Pooja! So that means it is safe to do power shutdown testing over stretched cluster by taking one site down. 

So once the site comes back up, what is the process of rebalancing? How much time it can take? And also do we need to manually move the VMs back to this site?

[poraj]

Hi Rahil,

Yes you can safely shut down one site.

It will take some time for the site to come back online and re-synchronization will take place post this.

The zookeeper and any forward progress made on the datastores while the site was shutdown will be synchronized. There is no definite time mentioned in the official documents.

Yes we need to manually vmotion the Vms back to the site that was shut down.

[Rahil Abrol]

How to make sure VM has proper affinity set? 

[hanumala]

Hi Rahil,

Site host groups and the corresponding affinities are created automatically at build time by the Cisco HyperFlex installer such that the failure scenarios will be handled accordingly. The host affinity rules are always described as “should run” and not “must run”,  You can always go to the Vcenter and check the cluster host/vm groups with corresponding rules created where the they are written in such a way that the the hosts and corresponding VMs of a particular site are grouped together.

The rules are created accordingly so that a VM should always be alive on the cluster instead of VM being must alive on a specific host. Also to add on here we have to make sure that the DRS is turned on such that VM always moves inside a specific site otherwise it will un-necessarily increase the utilization of the WAN link and increase read latency if VM goes to another site.

You can also refer the whitepaper here https://www.cisco.com/c/dam/en/us/products/collateral/hyperconverged-infrastructure/hyperflex-hx-series/operating-hyperflex.pdf 

Please let me know if you have any other queries on the same.

[hanumala]

Hi Rahil,

Would like to add another point here, In order for VMs to be appropriately placed within site affinity, it is necessary to have VMware HA and DRS enabled before creating the VMs; otherwise, the correct VM placement for desired affinity cannot be assured.

Thanks.

[Rahil Abrol]

Thanks Hemanth

[mirandabaiano]

hello everybody