OpenShift 4.1 Installation on Bare metal

[SunilKumar]

Is there any Video Sessions on how to install Openshift 4.1 on baremetal servers?

when i was checking https://docs.openshift.com/container-platform/4.1/installing/installing_bare_metal/installing-bare-m.... i am bit confusing on installeing RHCOS.

Cant we install RHCOS image directly on baremetal? should we compulsory use http server for installation.

Need more details on this section please

Thanks,

Sunil

[penguinpages]

So is it correcft to then state..  Based on this thread.  That if you want to learn OSv4  and evaluate it as platform for running it on-premis as a development and system stack for applications, you are out of luck.  Their is no method, documentation, to facilitate that.

Does this mean go back to Openshift 3 for now for that use case?

What is ETA when , when we sell designs to customers with OS4 liceses, they will then be able to start installing on premise?

[flozano]

No it doesn't. The documentation exists, it was already linked to in this
thread, and a number of people are using it with success. It is not that
easy to follow compared to an installer-manager setup on AWS, but doable.
Take care of getting the correct artifacts for OpenShift 4.1 (older coreos
ISOs and docs won't work) and be prepared to make a few tries and
troubleshoot between them.

By the time OpenShift 4.2 comes out we should get the CoreReady Containers,
that is easier to install, minishift-like experience for developers and
small lab environments.

[penguinpages]

Ok.. Good news.  So we do have method today with OS 4 to deploy on premise.

You noted "AWS is easier" but non of my target customers would / can use that.

I am building out a referance desgin and SOW structure to jump from to do these deployments but need to "eat my own dog food" to learn and make sure I have all the moving parts correct.

As such.  you refer to ""documentation exists".  I think you meant: https://docs.openshift.com/container-platform/4.1/installing/installing_bare_metal/installing-bare-m...    Which is fine... but it states

1) Deployment / boot strap VM must be RHCOS.  

2) To deploy RHCOS I noted what appeared to be documentation... and location to get media.. and process.  But it fails. 

Do you have better documentation or note about where a more detailed guide is, or one that reflects on-site procedures?

I am glad to help document and post my process for the greater community.  But having to get over this initial deployment hump.

Thanks,

[Rydekull]

Hello

OpenShift Container Platform (OCP) 4 does indeed require Red Hat Enterprise Linux CoreOS (RHCOS) for a setup, it is documented in the product documentation OCP but I can understand your problem since it can be a bit tricky to get started with.

The reason being is that RHCOS is still a bit sensitive that you have your DNS / DHCP / Load Balancer settings just right when doing an installation.

I'd be willing to try and help you being a sounding board on this if you'd could give me some more details of your setup and where you are currently stuck.

I would like to know:
Have you configured your DNS with the required settings?
Have you configured a load balancer with the required Virtual IPs?
What is your cluster subdomain?
What is your current DHCP settings?
Do you provide a DNS server through DHCP?
Do you provide a subdomain for DNS search options?
What do you envision the FQDN to be for your bootstrap, control planes(masters) workers?

I was looking at your screenshots earlier in this thread and its a bit unclear how far you are getting in the process.

Do you actually succeed downloading the image?
Does it complete the installation and reboot?
If yes, how far do you get when booting? Do you see anything when the machine is booting up?

One of the "features" of RHCOS is that it tends to redirect the console to an actual serial console. Meaning that it can be a bit frustrating to troubleshoot the initial boot if you are not comfortable with working around such issues. You can see the boot options when booting the RHCOS and it is possible to edit them.

Also, I saw that you mentioned that you are looking to build a offering around OCP 4, It is quite common for us from Red Hat Consulting to work with partners and customers when they build offers to ensure that they are correctly setup, just a good thing to consider.

I do suggest though, since OpenShift comes with subscriptions to run it that you open a support case on issues like these. It is important for us to make sure that we can keep on minimizing the occurences of these traits of the installer and user experience. Since as im sure you can vouch for, it can be a bit suboptimal :-)

[penguinpages]

Thanks for response.  Let me address each item individually.
 
Have you configured your DNS with the required settings?  Yes.. I was using router with its DNS functions but tearing that out as I am not sure it can do all the zone features and SRV records / keys that are needed... 
Have you configured a load balancer with the required Virtual IPs?  I have pacemaker and a VIP setup with the two cluster front end IP (one for master and one for infrastructure.. if you have documentation on details of how they want the VIP defined.. I can follow.  I am using the KVM nodes as the IPLB layer (vs Dedicated vms) so I can also use that as abstraction for other LB needs in my demo environment.
What is your cluster subdomain?  penguinpages.local
What is your current DHCP settings?  A and PTR records for all physical KVM systems and VM.  I have no STONITH setup as SuperMIcro IPMI is being pain... so no records for that.  Standard NTP, zone and such.  I was running down rat hole of PXE and file image name and location in DHCP that RHCOS would need.  If this is still needed I can do that.  I was to host all this kind of PXE out of a common glusterfs for cluster on the KVM nodes again.. Ripping it out of the switch as I wan to avoid some feature in the OS4 stack that it could not support..  Manual as to what record types are needed and I can follow that fine.
Do you provide a DNS server through DHCP? Yes
Do you provide a subdomain for DNS search options? Yes.  NTP, bootp server and file is the plan... once I get manual to what is needed
Do you actually succeed downloading the image?  RHCOS download was fine... as I completed per screen shots installation.
Does it complete the installation and reboot?  Sure..  just not to anything useful :)
If yes, how far do you get when booting? Do you see anything when the machine is booting up? Screen shot shows.  Just sits there at pxe such that it found no target MBR (chain load) or direct path (EFI file target).   I did not boot up recovery image and mount file space to see what it plopped out there.
actual serial console. Meaning that it can be a bit frustrating to troubleshoot the initial boot..  -> I can check that via "virsh console rhcos"  next boot attempt. 
What do you envision the FQDN to be for your bootstrap, control planes(masters) workers?  boot node will just be osmstr01 (first of the three master nodes) .. .but what I am confused is in production would it not be better that their be IPLB load balance and against dual DHCP with split scopes..  But ... for first round to get it up and running I can just KISS and follow guide.
 
OpenShift comes with subscriptions to run it that you open a support case on issues like these. -> Sure.  THis is demo so I can learn it to start to sell it via demo of it.  BUt... trying to contribute back to community of guide.   
You mention that RHCOS is needed... that RHEL not is ok.    AKA... keep clawing at isseus with RHCOS... or...  to get infrastrucutre round 1 up... just KISS and use RHCOS. Is their beta for ppcle of RHCOS?
I am open to suggestions.

[Rydekull]

First of, RHCOS is required, that is not an option to skip, the bootstrap and control plane requires it.

But, for workers it is possible to use RHEL-nodes, but it just doesn't make much sense to mix at that stage, since at that point you've already gone through the pain to setup RHCOS. But sure, there are circumstances where you still want to use RHEL on your workers.

---

>  Yes.. I was using router with its DNS functions but tearing that out as I am not sure it can do all the zone features and SRV records / keys that are needed... 

Great, so SRV records are supposedly setup correctly then. Good.

>  if you have documentation on details of how they want the VIP defined.. I can follow.

https://docs.openshift.com/container-platform/4.1/installing/installing_bare_metal/installing-bare-m...

> What is your cluster subdomain?  penguinpages.local

Okay, that's good, but most likely not sufficient, since penguinpages.local is probably your TLD in general for your environment. OCP 4 requires a subdomain per cluster you install.

https://docs.openshift.com/container-platform/4.1/installing/installing_bare_metal/installing-bare-m...

> Do you provide a subdomain for DNS search options? Yes.  NTP, bootp server and file is the plan... once I get manual to what is needed

See previous answer, worth noting. RHCOS expects to get the subdomain of the cluster as the search path through DHCP.

> If yes, how far do you get when booting? Do you see anything when the machine is booting up? Screen shot shows.  Just sits there at pxe such that it found no target MBR (chain load) or direct path (EFI file target).   I did not boot up recovery image and mount file space to see what it plopped out there.

I wouldn't declare an unbootable machine as a succesful installation. I also do not see in any previous screenshot what your boot options for your VM was either.

Also, just noticed the screenshot which you opted to "skip" the ignition config, Which is a large indication to me you havnt been following the documentation:

https://docs.openshift.com/container-platform/4.1/installing/installing_bare_metal/installing-bare-m...

>What do you envision the FQDN to be for your bootstrap, control planes(masters) workers?  boot node will just be osmstr01

Yeah, that isn't a FQDN, so I can just assume you mean osmstr01.penguinpages.local. Again, this wouldn't be a good domain for OCP since it requires its own subdomain to live within.
Also, the first masters cannot be a bootstrap node. Its either a master (Control Plane) or bootstrap, not both. The Bootstrap node will be deleted after initial installation.It is just used to "Bootstrap" the cluster.

I can recommend just using a normal RHEL iso, and boot with that until you get your basic networking and DNS setup correct, It is way easier to verify it using that. At that point, proceed to the RHCOS installation.

As i mentioned before, RHCOS is very picky about the correct setup, you need to be very particular and follow the documentation to a great detail.