Will try to make it, keeping an eye for details!

I'd be in for a listen. Currently getting pushed to cert in K8s and have been trying to dodge it. I primarily do Openstack ops support. But lately we have been slinging K8s on top of it. Our Ubuntu offering is at least a bit sane. Our RHEL offering is Openstack virtualized, with K8s on top of that. Damn virtualized inception.

ETA

To docker specifically. I've been surprised how stable pihole runs in a container on my Synology nas. Just had to ignore the guides and engage my brain to get the packets flowing.

Ooooh. I may be in touch, @Subnet. Thanks for the offer. I might be reaching out.

@packingXDs, I jumped on the OpenStack bandwagon early on several years ago but it was purely personal interest. At the time I wasn't even remote involved with the ops side of things and there was no interest at work. I ended up going down the container path which led us to k8s. I really do like it. It's fascinating to me the way it works and the benefits I think we'll gain. We've only recently started rolling things to production so the full benefits are still TBD. What I can say, though, is that deployments are incredibly easy. What would have taken hours before is done in minutes.

I think I'll start an official thread with the presentation details in case people haven't read this one.

Been doing it at the service provider level for many years. Havana release I think. I'm sure at the dev side of things it's probably great. But the evil, layered, abstraction that it does can be a nightmare at scale. Expectations from customers doesn't help.

"After an upgrade one team is reporting 100% increase in deployment times!"

2 minutes vs 4 minutes.

Oh and while smoke testing they deployed 200k vms in a few hours. Turns out even 10g links have their limits when you tell every team to deploy everything at once, over and over.

That's an awful lot of VMs. We're definitely going to have to put some controls on the teams. One team's first response to performance issues was "scale to the hills and beyond!" rather than figuring out why they had a performance issue. Turns out it was (shocker) a code change that was required.



This has been an eye opening experience for me coming from a development role to infrastructure. I'm blamed for everything. Wait, nothing's changed.  I swear, we got on a call last week or so and it was obvious there was a problem with the keys coming from our API gateway. First word out of the API team's mouth was, "could it be an OpenShift problem?".

I always say, "abstraction is good...util it isn't".

"We have a problem. No we didn't change anything" at the outset means they changed something and it's the problem.

The only other constant is "we had 30 second network slowness X days/weeks ago. Please provide a rca".

It's funny because it's true.  

"Our websphere app talks to an Oracle RAC cluster, but I heard a DBA was looking at Postgres on Openshift - could that be our problem?"  

Oh boy. Those both hit home. I was on a call with a client 25 years ago because our system wasn't working. The very first question I asked is "what changed"? Of course, the answer was "nothing". After an hour, the guy let slip that the server had been moved. Sure enough, the thing wasn't even on the network. At least it was billable.

We still get the second scenario a bit. The last time I managed had a transaction retention of < 2 weeks. It was always fun when a client would call in March saying, "hey, we're looking at this transaction from Dec. Can you tell us why it failed?". I guess some people just think everything is stored indefinitely.
I'm expecting a lot more calls. As mentioned above, I just have to get people to slip up and let the truth out. "We have memory issues, it must be Openshift!" Until a developer says on a call that the issue is present on his machine also. We got blamed for issues with a mail server that doesn't even run in the cluster.

On a side note, what do you think of WebSphere? My old team migrated to JBoss after I left and it seems to be much more performant.

I wish our guys would switch to jboss, just from a licensing and resource footprint alone.  Unfortunately, most of our middleware people have sold their soul to big blue.

We pushed hard to move away from ibm/websphere on the Openshift platform, but it's hard to change an addict.  

We even showed them how to run their code on jboss/wildfly in Openshift, response was "that's really nice, we're gonna use liberty".

Now most of us old Redhatters are just wondering what IBM is gonna do with Redhat.    

Hi neighbor! Bet we know a couple co-workers. Don't disable selinux and can haz ip.

I bet several people in this thread know some overlapping names/contacts.  

Wow. So you're running WebSphere INSIDE of OpenShift? Oh my!

We have zero plans to run app servers in the cluster. I know IBM says you can. Just not sure you should.

I totally get the Big Blue mentality. I was shocked when JBoss was approved. Historically, we don't select tools because they're the best in meeting our needs. We select the tools IBM gives away because we spend so much on their mainframes. It's hard to sell another solution when you're competing against "free".

Yes, but "IBM free" can be really expensive.  

If all goes to plan I will be out of the house on Saturday... will this be recorded?

Shoot. I think I can do WebEx recordings with my account. I'll try. If not, I'll try recording through my streaming software behind the scenes.

ETA: I just did a test. I can record and the best part is that after the meeting ends, it translates it to MP4 automatically so I can upload that to my YT channel. The downside is that you see my mug the whole time. I might have to kill video and see if that comes out better. Quality looked pretty good, too.

Most containers in my experience (at least well written ones)... have all extraneous crap stripped out... thus why the container images are so small (many just a few hundred megabytes), compared to a virtual machine which will be a full OS.

I love docker.... currently running 15 containers on my home server.

We start with a minimal image and then tailor it to our needs. Sometimes it sucks when you don't have tools you've come to expect (nc, curl, etc.) The price of keeping things small.

Yeah... If the container your using has a package manager however (apt, dnf, whatever that thing arch uses is, etc.).. it's usually a pretty simple install if you want to add something small like curl.  I've got a couple containers based on Ubuntu and even a distro like that that is usually pretty bloated, the containers are fairly stripped down.  When I needed curl on them, I simply bashed in to the container, apt install curl, and the containers now had curl.

A lot of it depends on how the image was done of course.. as I've saw some that were pretty big as well.

Yup. Many ways to skin a cat. I just keep a busybox image handy which has curl already.