In this blog we will see the Top 10 Technical Interview Questions for CEPH that is likely to be asked when interviewing for a position where CEPH will be used. Please note that these are not just theoretical questions related to the definition and CEPH functionalities rather a conceptual queries based out of the real time scenarios.
If you need to establish a CEPH cluster from scratch you can refer the previous blog post on How to Install 3 Nodes CEPH cluster.
You may also need to install the OS so you can follow How to Install Rocky Linux
In terms of performance, how does CEPH do against a traditional ZFS-based server when it comes to NFS? Less/many files in terms of access performance?
A lone ZFS server and one client accessing the storage will often operate more quickly than a clustered filesystem. Ceph excels because to its parallelism and capacity to provide tera-peta-exabyte scale storage solutions without a single point of failure. At scale is where Ceph will outperform a single ZFS server. A single ZFS server is not as efficient as a cluster with hundreds or thousands of users.
How many OSDs per host do you recommend?
Of course, there are a few things to consider, but the more the better. Which ultimately comes down to personal preference: (a) performance or (b) capacity. Each host should have at least five drives (OSDs). 30 OSDs is the ideal point for performance maximizing. The sweet spot for capacity is 60 OSDs. OSDs on 45 Drives provide a decent balance between the two.
Is Ceph independent of hardware? Do the servers have to be the same model?
Since Ceph is indifferent to hardware, you should take care to avoid having a lot of variance in the storage node density. You want your storage nodes to have the same total amount of storage in both, or as near to weighted as feasible. In general, clusters with servers with 30 and 60 bays should be avoided.
Can pool be nested in CEPH clusters ? Can I create a host failure domain replica pool by merging two OSD level failure domain erasure coded pools.
No pool can be nested in the manner specified. Even if you could, the efficiency of the scenario wouldn’t be optimal.
Let’s say we are constructing a Ceph cluster with three servers. With 100TB of storage available on each server (or node), the cluster’s raw storage capacity would be 300TB.
With a basic three replicated pool, you have 100 TB of usable storage, or 33 percent.
The usable storage capacity would only be 66TB, or 22% effective, if we could establish a 4+2 (66%) on the OSD failure domain and then replicate three times on the host level. Even with two reps instead of three, the useful capacity would only be 98 TB.
In what way would you advise establishing an internet connection to the cluster? The program is a sizable POSTGRES database that serves as the foundation for a search engine.
Don’t connect your cluster directly to the internet; instead, put all of your storage on an internal network.
Instead of allowing users to communicate directly with your storage, you should consider employing a proxy service that can communicate with storage on the back end and with users on the front end.
Is it possible to keep the data on an existing server and build a cluster from it? Is it possible to construct a cluster and keep the data on both of your servers if you have two?
No, in a nutshell . Sadly, all of the data that is currently on a server must be erased because Ceph needs to completely reformat the HDD or SSD when it constructs an OSD.
When some OSDs are failing or reading/writing extremely slowly but are not yet dead, how does the cluster respond?
Due to Ceph’s high level of parallelism, the majority of data will remain unaffected; but, for the portions of your dataset that are impacted by a failing HDD When an OSD has “slow operations Ceph will report it. An administrator can then mark the concerned OSD out of the cluster and receive an alert. The administrator can then replace the failing disk whenever it’s convenient for them, as Ceph will now generate the data on that slow OSD somewhere else in the cluster.
Do more drives have to be from the same manufacturer or size?
Both the vendor and the size do not have to be the same, but you should take care that there is not a significant variance in the density of the storage nodes when it comes to disk size. You want to have the same amount of total storage in both of your storage nodes, or as near to weighted as feasible.
It is therefore OK to have a variety of disk capacities, but strive to distribute them equally among all storage servers.
What happens if there is a power outage and some servers are lost? Does Ceph restore itself automatically when the servers are back up?
Yes, following an unplanned power outage, Ceph will automatically heal. Split brain is avoided by the way Ceph monitors handle quorum; the cluster will restart and resolve itself without assistance from humans. Many of our clients have experienced unexpected power outages, and they are always delighted to learn that the storage cluster resumes its normal state on its own without any help.
Can you explain the distinction between Ansible and Cephadm?
The deployment tool Cephadm was developed for the most recent iteration of Ceph Octopus, which is version 15. Rather of installing bare metal RPMs, it uses containers to deploy and manage the Ceph services. The goal of Cephadm, developed by Ceph developers, is to simplify the process of creating a Ceph cluster for the first time by replacing third-party deploy tools like as Ansible or Puppet. Therefore, you do not need to learn how to utilize Ceph and Ansible simultaneously. Whether you decide to use Cephadm for containerized deployment or not, I believe that Ansible deployments of Ceph will remain a thing.
We have tried our best to provide you the real time answers but you can always refer the Official CEPH Documentation for further learnings Welcome to Ceph — Ceph Documentation