cPanel High Availability Roadmap

[cPanelFelipe]

Will the db deployments also be per-site? The web pods are per-site basis. If the db pods are also per-site basis and they are HA (like some operators out there), that would help a lot in isolation of sites.

Clarification: Web pods will be per-user, not per-site. The current plan is for all users to access a single DB pod.

We expect to reassess these sorts of things as we deploy and refine the offering.

Also NFS is a major pain point. If you could select a good K8 operator (rook-ceph etc) to provide it from within cPanel itself, that would help greatly. Otherwise operating an NFS would be problematic for a majority.

From what we found, k8s-based storage solutions generally focus on providing nonvolatile storage for the cluster itself; we need something that serves the cluster and the cPanel server concurrently. Some technologies, like Rancher Longhorn, do expose iSCSI, but iSCSI doesn’t really accommodate concurrent access.

NFS is a simple, flexible, and ubiquitous solution to that end. It can be deployed standalone or (via NFS-Ganesha) as a frontend/proxy to Ceph, Gluster, etc. So customers can choose the replicated-filesystem technology that best suits their needs.

That all said, this is another area we expect to reassess as we deploy & refine.

 

[ern008]

Clarification: Web pods will be per-user, not per-site. The current plan is for all users to access a single DB pod.

We expect to reassess these sorts of things as we deploy and refine the offering.

It's great that you are working on it.
However, it seems there are issues with the following:

• There will be no option to limit databases resources per cPanel account.

It shouldn't be possible without running multiple database pods per cPanel account.
Currently it's possible using CloudLinux MySQL governor.

• Running LiteSpeed web server enterprise seems to be problematic in this configuration.
• CloudLinux Imunify360 might not work too.
• Web pods per-user means that a reseller won't have security isolation between hosting accounts.
• Web pods per-user might cause higher resources utilization due to running many web servers.

There will be a webserver running for each user.
I can think of a reason but this might happen.

 

[ern008]

It shouldn't be possible without running multiple database pods per cPanel account.

I meant it shouldn't be possible without running database pod/s for each cPanel account.
This is how it seems to be for now.

 

[Maccess]

We have begun heavy research phases of High Availability for cPanel. We have a forecasted roadmap for HA and we will continue adding additional information to it as we flesh out the project further.

*edit* x-linking Configuration Cluster thread

Hi,
Can i ask for new link? i found 404 error for "forecasted roadmap for HA" link.

 

[cPRex]

@Maccess - this is fixed now!

 

[WorkinOnIt]

Great to see progress. I think the links are all pointing to 404. I think it should be this one now? Correct?
https://docs.cpanel.net/knowledge-base/roadmaps/cpanel-cloud-edition-roadmap/

For my own usage on this HA setup (and I would suspect I am rather typical of a Small / Medium hoster), most of our servers contain 90% WordPress sites. So this solution - while obviously being agnostic of all software etc - does need to consider speed and reliability for end usage. If you have e.g. multiple WooCommerce sites all running ecommerce transactions, the solution needs to be able to replicate that data efficiently and reliably across backup databases. If one server goes down, the next one kicks in, but there needs to be multiple databases to this solution.

I am not 100% familiar with the proposed technology K8, Pods and NFS solution technically - but it sounds comprehensive! I will trust that the architects' concept will take into account some of the 'non-expert' users like myself! After all, that's the beauty of cPanel and WHM - they do bridge the gap between expert and intermediate - so please keep that in mind.

It's not entirely clear from the roadmap, but in an ideal world, the HA solution would consider the following types of scenarios :

WHM Server Admin experience

• Be able to set up 2 or more servers and databases to act as 'fail over' - if one goes down, next one kicks in to provide as 'seamless as possible' service.
• Admin be able to log in to a single 'parent' WHM or some kind of management console and be able to see replication and failover event status.
• Any changes made within server 1 WHM would auto-populate to the additional connected servers and visa versa.
• Syncing should be configurable to suit different cases: Either instant or able to be scheduled.

cPanel Hosting Customer experience:

• They simply need to be able to log into their cPanel any be able to see their data in a visualised sense, that it has some kind of replication happening (e.g. sync status icon?) and perhaps a visual log showing failover events?
• Will the proposed solution give any capability to cPanel users?

Website End User experience

• A website user will not experience any noticeable delay before / during / after replication or failover events.

 

[cPRex]

@WorkinOnIt - I thought I got all the links fixed - can you link me to a specific post where they aren't working?

 

[Mind on the Net]

Revisiting this, it feels like it would be instrumental to have the NFS service built-in. Not having it would limit the reach of the HA edition of cPanel, since it would be difficult and expensive for small to medium hosts to run their own NFSes. Even larger hosts could avoid it for the reason that maintaining an NFS service is a major commitment. Considering how the two widely available and reliable managed NFS services are AWS's own NFS service and Google's Filestore, this is pretty limiting. AWS's NFS service is told to suffer from low performance while Google Filestore works quite well but Google Cloud's egress costs are a killer for any consumer hosting operation.

But if cPanel created/forked a shared storage operator (ceph, rook, gluster etc) and maintained it themselves, this would make it possible for the cloud edition to have its own NFS built inside and allow the cloud edition to be adopted more widely than it would otherwise.

 

[WorkinOnIt]

Any updates? I see nothing since I provided feedback.

 

[cPRex]

I reached out to @cPanelLauren and she's going to provide an update on this a bit later.