|
Seriously, just run a couple of oVirt boxes on the side if you don't want to pay the VMware tax. It's not especially full-featured, but at least it doesn't often break itself in half and you're still only supporting one hypervisor config.
|
#
?
Jul 14, 2015 22:19
|
|
|
|
| # ? Apr 23, 2024 12:45 |
|
|
evol262 posted:DCs in openstack just seems like such a bad idea. Nova doesn't even set up libvirt auth or anything. Why don't they run DNS/LDAP/DHCP/etc in normal VMs (on KVM if they want) with some kind of HA manager instead of dumping it in openstack? If I wasn't clear it wasn't literally Windows DC's. It was PowerDNS, OpenLDAP, and ISC DHCP. So many of the core Windows DC services, just the Linux version  Not that that makes it better.As for "why" it goes back to resources. We had OpenStack up and running, and a skeleton crew of engineers, so that's where poo poo got built. No one had time to build a production oVirt/RHEV environment alongside it. Yes it was a terrible idea. But we got "HA" for the services that were an awful fit for the cloud by build a couple bare metal boxes and putting them in the same load balancer pool. No I am not proud of it. This was a company trying to cut everything to the bone. That's not why I left, but it sure made it easier to give notice.
|
|
#
?
Jul 14, 2015 22:30
|
|
|
Well, oVirt/RHEV is a mess there, which is why I mentioned the libvirt bit. vdsm (the glue between the ovirt engine/rhev-m and the system) expects to be able to do whatever it wants, and configures auth for libvirt, which will break nova. But a trivial pacemaker+libvirt config happily hosts traditional VMs alongside nova. It's a great solution if it owns the box. It's poo poo at playing nice with other services. Not blaming the environment, and I've seen worse (production auth as mysql-backed LDAP with mysql running in a single linux-vserver instance on Gentoo with a publicly-routable IP in 2k13), just curious.
|
|
#
?
Jul 14, 2015 22:48
|
|
|
kiwid posted:
Implement affinities and you'll find it more quickly after a crash.
|
|
#
?
Jul 17, 2015 21:24
|
|
|
Potato Salad posted:Implement affinities and you'll find it more quickly after a crash. In vSphere 5.5 and earlier, HA will ignore should DRS rules. Must DRS are honored but potentially dangerous here because if your only valid targets are down, then the VM(s) won't power back on. Note that vSphere 6 allows you to set HA to honor should rules. Some more detail If you run a virtualized vCenter (we do, and it is rarely an issue), you should absolutely use should DRS rules to keep it on a couple of known physical systems to make it easier to track down if it has problems - just know the limitations 
|
|
#
?
Jul 17, 2015 22:00
|
|
|
Is it normal for VCSA 6 to take like 10 minutes to become available? The generic "welcome to vsphere 6 bla bla" site comes up almost immediately, but then it just sits in a broken state and won't let me click through to the actual webUI without cryptic errors for a good 10 minutes. My VCSA is at minimum specs but I still kind of  at 10 minutes.
|
|
#
?
Jul 17, 2015 23:35
|
|
|
Martytoof posted:Is it normal for VCSA 6 to take like 10 minutes to become available? The generic "welcome to vsphere 6 bla bla" site comes up almost immediately, but then it just sits in a broken state and won't let me click through to the actual webUI without cryptic errors for a good 10 minutes. as far as I can tell, yes
|
|
#
?
Jul 17, 2015 23:59
|
|
|
Yea, that's normal.
|
|
#
?
Jul 18, 2015 03:15
|
|
|
Thank god I don't reboot it all that often then 
|
|
#
?
Jul 18, 2015 05:38
|
|
|
Martytoof posted:Thank god I don't reboot it all that often then the best is when you reboot it and you hit the limit where it wants to check the disk. You don't notice until it has taken too long to come back and you start logging into host servers, so then you have to reboot again with fastboot enabled.
|
|
#
?
Jul 18, 2015 06:01
|
|
|
Fancy_Lad posted:In vSphere 5.5 and earlier, HA will ignore should DRS rules. Must DRS are honored but potentially dangerous here because if your only valid targets are down, then the VM(s) won't power back on. Note that vSphere 6 allows you to set HA to honor should rules. Some more detail poo poo, I have my work cut out for me. Thanks.
|
|
#
?
Jul 20, 2015 17:37
|
|
|
Martytoof posted:Is it normal for VCSA 6 to take like 10 minutes to become available? The generic "welcome to vsphere 6 bla bla" site comes up almost immediately, but then it just sits in a broken state and won't let me click through to the actual webUI without cryptic errors for a good 10 minutes. Sounds like you just said "Rhyme, don't upgrade to Vsphere 6."
|
|
#
?
Jul 20, 2015 22:54
|
|
|
Rhymenoserous posted:Sounds like you just said "Rhyme, don't upgrade to Vsphere 6." He actually just said :java:
|
|
#
?
Jul 20, 2015 23:08
|
|
|
Is there a dominant application container for Windows? Looking for the same kind of stuff as Docker does on Linux. My particular use-case is containerizing IE (and maybe Firefox and Chrome as well) along with a Selenium instance for doing a bunch of web integration tests in parallel. Microsoft's pre-built unit test VMs are nice in theory but I'd really like to avoid the memory+CPU overhead of running full VM instances. Or is there another way I could do this? Thin client type stuff maybe? I've also been wondering if maybe piggybacking on the corporate Citrix environment would be a viable approach, but I don't think that would give me enough control over versions (I want to run tests on IE9, 10, and 11). I strongly prefer "free" because my company doesn't believe in spending money on software when they have a surplus of perfectly good testers on staff. Piggybacking on existing infrastructure is fine though, and I can probably eventually have my way if it's pitifully cheap (say less than $10 a month). Paul MaudDib fucked around with this message at 00:14 on Jul 23, 2015 |
|
#
?
Jul 23, 2015 00:05
|
|
|
Paul MaudDib posted:Is there a dominant application container for Windows? Looking for the same kind of stuff as Docker does on Linux.
|
|
#
?
Jul 23, 2015 02:52
|
|
|
Martytoof posted:Is there any way to tell VCSA to not force a URL redirect and just use the URL I came in on? Late reply I know but I think you can resolve this by setting the VCSA hostname to localhost. I had the same issue with a VCSA upgraded from 5.5 that had a hostname set, but a newly created one with a default hostname of localhost does not.
|
|
#
?
Jul 26, 2015 02:35
|
|
|
Paul MaudDib posted:Is there a dominant application container for Windows? Looking for the same kind of stuff as Docker does on Linux. MS is working on some sort of windows container thing, but that doesn't really help you now. If you're using 100% WebDriver tests then you can dockerize selenium and phantomjs/slimer and spin containers up in a Linux VM. I assume you're not based on the content of your post, though.
|
|
#
?
Jul 26, 2015 17:55
|
|
|
This is coming from someone who only knows that "containers are VMs as VMs are to hardware" and that's it: Isn't Azure container-y? That is, you're never really aware of the underlying VMs when deploying apps in Azure and all you care about is the framework presented to you?
|
|
#
?
Jul 27, 2015 00:44
|
|
|
Potato Salad posted:This is coming from someone who only knows that "containers are VMs as VMs are to hardware" and that's it: "I'm only aware of the framework" is pretty much the definition of PaaS. Could be a container, may not be.
|
|
#
?
Jul 27, 2015 02:12
|
|
|
I'm trying to set up a single server with CentOS7 and oVirt 3.5, and when I tell it to add itself as host, it spits out logs in webadmin about installing packages, then after a minute webadmin hangs completely, then ten minutes later it pops up some error about :quote:Error while loading data from server: Download of https://192.168.2.201/ovirt-engine/webadmin/deferredjs/8F788F85BF8B5BF5E5233CF55EF14245/11.cache.js?autoRetry=3 failed with status 0() If I run engine-cleanup then install engine-setup all over again I still can't seem to make it through. Any ideas? I did notice that hostnames from centos on this server don't resolve on my network, but I edited the /etc/hosts and they do resolve on the server itself. Not sure if that has anything to do with it. All I'm planning for now is spinning up a VM for FreeNAS taking half the storage on the server, wish there was an idiot's guide beyond the oVirt documentation.
|
|
#
?
Jul 27, 2015 05:39
|
|
|
Zero VGS posted:I'm trying to set up a single server with CentOS7 and oVirt 3.5, and when I tell it to add itself as host, it spits out logs in webadmin about installing packages, then after a minute webadmin hangs completely, then ten minutes later it pops up some error about : Are you using all in one? You should be If you have the memory (2gb is enough, 4gbis better), I'd use a hosted engine, which will let you easily expand your setup with resiliency on the engine. If you don't, use all in one. When you install the engine and try to add the host it's running on as a compute node, vdsm is trying to set up a bridge and waits for it to come up, but that itself will take the engine offline. Then otopi (the framework which does all this) rolls the state machine all the way back and reverts it all. Unchecking the "verify connectivity" box in the new host may get you past this, but all on one machine really isn't a recommended setup without the all in one plugin.
|
|
#
?
Jul 27, 2015 05:52
|
|
|
evol262 posted:"I'm only aware of the framework" is pretty much the definition of PaaS. Could be a container, may not be. Containerization and PaaS are pretty distinct things, though. (Full Disclosure: I work for a PaaS company, Apprenda) Containerization allows you to define the dependencies for your application and bundle them together in some sort of semi-portable format that doesn't necessarily care about the underlying hardware or software (I can run a docker container on whatever machine I'm sitting at, oblivious to the guest OS, because my container software abstracts all of the calls to the OS via a translation layer), allowing for easier? deployments and such. PaaS does care about the underlying OS and such to a greater extent, and offers more features. For instance, a PaaS may offer services such as authentication, multi-tenancy, or DB services through a single API. However, applications are built in a particular language or runtime that has to comport to some supported runtime of the PaaS - although several are starting to offer container deployment options now. Finally, PaaS allows for scaling, HA, and other fun features that don't really have anything to do with containers. The twist is that a lot of containers are starting to become more PaaS-like - Docker has some sort of container orchestration thing now, which can handle scaling and HA and such. So I guess, in short, you're pretty much right, but I felt like providing more depth. I haven't had my coffee yet, though, so take it all with a grain of salt.
|
|
#
?
Jul 27, 2015 12:49
|
|
|
uncurable mlady posted:Containerization and PaaS are pretty distinct things, though. (Full Disclosure: I work for a PaaS company, Apprenda) Full disclosure: I work for Red Hat, and I'm hitting the limitations of containers all over the place. My reply wasn't to say that containers couldn't be PaaS, but that there's no way of knowing. Containers and PaaS are pretty distinct things. Being presented with nothing but a platform (whatever platform, really -- node.js, python, .NET) doesn't give you any indication at all of whether that's running inside a container or not. Openshift is. Engine Yard isn't (or didn't, they might now). Containers are a technology. PaaS is a solution built on top of technologies. I didn't want to go into a "PaaS vs Containers" thing because there's some overlap, but they're not really in the same business category, and I was in a rush. Containers don't provide a meaningful abstraction of any of the system calls. There's isolation in kernel namespaces and cgroups, but docker (and containers) absolutely care what underlying hardware and software is under them. It seems like they don't, because the kernel is very good at presenting a stable interface, but the containers are making syscalls to the host kernel with no abstraction at all. If you hypothetically run a docker container (or systemd-nspawn, since that actually works on alternative architectures, which Docker currently has issues with that IBM and others are working on) on a non-x86 host, a i686 host, or a host whose kernel is missing all the bits you're used to from Ubuntu or RHEL or CoreOS or whatever (hypothetically -- a container which relies on some missing ipv6 or ipsec features), it'll break. It'll also break if your application tries to load a kernel module, run VMs, run docker containers itself, and a lot of other cases also break without privileged or super privileged containers. Even then, they can break. The advantage of containerization isn't that it lets you define your dependencies, at least not in any real way. There are a lot of ways to do that, and Docker has no mechanism at all for handling it other than "build from this base image". The advantage is a rapid deployment and testing cycle, trivially reproducible tests and dev environments (it's easy enough to get the same with image-based deployment in openstack/ec2/whatever, but a little slower and with a little more overhead), and not needing a lot of underlying infrastructure. It's very good for deploying microservices and related services which may not benefit from a full-blown VM and don't need remote access or a real init system. They're terrible for a lot of use cases. PaaS cares about the underlying OS. Which could be a container. Or could not be. PaaS consumers don't need to care about the underlying OS, which is the entire point. Services like authentication, multi-tenancy (running backed by containers means security problems and lack of isolation here), DB services, autoscaling, HA, and all the rest are actually built on technologies, all of which may or may not also be in containers. PaaS doesn't allow for anything the underlying technologies don't support, and the kubernetes people (and especially the kubernetes on mesos people) would disagree with many of your assertions about what "doesn't really have anything to do with containers". Orchestration, HA, and auto-scaling are IaaS-like, because PaaS is a solution built on technologies which can provide those things (mostly). I'm really not trying to nitpick here, but all of this is confusing enough to people reading the thread who are primarily running Hyper-V or VMware or whatever and don't deal with containers and STUFFaaS may be confused enough without muddying the waters.
|
|
#
?
Jul 27, 2015 16:29
|
|
|
evol262 posted:Are you using all in one? You should be I have 48gb ram but Hosted Engine says it outright requires two hosts and shared storage. I'll try out the All In One thing and see how it goes, thanks.
|
|
#
?
Jul 27, 2015 16:38
|
|
|
Zero VGS posted:I have 48gb ram but Hosted Engine says it outright requires two hosts and shared storage. Hosted engine does not require two hosts. It does require shared storage, though, but gluster support has landed, so you can use a local gluster brick if you want to.
|
|
#
?
Jul 27, 2015 16:54
|
|
|
Zero VGS posted:I have 48gb ram but Hosted Engine says it outright requires two hosts and shared storage.
|
|
#
?
Jul 28, 2015 00:21
|
|
|
loving awesome answers.
|
|
#
?
Jul 28, 2015 11:54
|
|
|
evol262 posted:I'm really not trying to nitpick here, but all of this is confusing enough to people reading the thread who are primarily running Hyper-V or VMware or whatever and don't deal with containers and STUFFaaS may be confused enough without muddying the waters. Yeah, good point. You did a better job of laying out the differences than I did, though.
|
|
#
?
Jul 28, 2015 12:31
|
|
|
Anyone know of a place that offers a free VPS trial besides AWS? I used them before switching to 2 paid VPS instances, 1 running ZNC/Mumble/TS3 and another as an rtorrent/Dropbox solution for family members. So far that only thing I've seen is the 2 month/$300 trial from Google Cloud. I'm just looking to learn more about CLI linux hands on, which I'd rather avoid spending another $5 a month if possible.
|
|
#
?
Jul 28, 2015 22:01
|
|
|
EconOutlines posted:Anyone know of a place that offers a free VPS trial besides AWS? I used them before switching to 2 paid VPS instances, 1 running ZNC/Mumble/TS3 and another as an rtorrent/Dropbox solution for family members. Is there a reason you can't virtualize this on your laptop or whatever? You won't find much cheaper than $5/mo.
|
|
#
?
Jul 28, 2015 22:45
|
|
|
NFOservers has a one month free trial and they're pretty cheap. I used them for ~18 months to run a mumble server among other things and was very happy with them. Over the the Raspberry Pi thread they're discussing the pros and cons of Seafile vs Owncloud, and a Pi B+ is plenty fast to run a mumble server.
|
|
#
?
Jul 28, 2015 23:03
|
|
|
evol262 posted:Is there a reason you can't virtualize this on your laptop or whatever? You won't find much cheaper than $5/mo. No reason, I was just hoping to find something free since it won't be doing anything other than basic SSH linux learning. I'd rather use something attached to a datacenter vs inside VMWare or whatever though. Hadlock posted:NFOservers has a one month free trial and they're pretty cheap. I used them for ~18 months to run a mumble server among other things and was very happy with them. Thanks, I'll check them out.
|
|
#
?
Jul 28, 2015 23:14
|
|
|
This thread has been pretty useful in getting me up to speed on VMware, but I'm at a new company now and it looks like they're using OpenStack. How do I start learning this poo poo?
|
|
#
?
Jul 30, 2015 05:41
|
|
|
Poke around the RedHat OpenStack docs, they are miles better than the official docs. Start with the Component Overview. Google around for articles that give overviews. Read this post from evol262. If you want to play around with an installation, use RedHat's packstack, it's a much better installer than the vanilla tools. If you have to administer OpenStack (as opposed to just use it), then go ahead and shave off all your hair because you're going to lose it anyway. It's a goddamn nightmare.
|
|
#
?
Jul 30, 2015 06:34
|
|
|
Dr. Arbitrary posted:This thread has been pretty useful in getting me up to speed on VMware, but I'm at a new company now and it looks like they're using OpenStack. That's a broad question. In general, knowing what vendor you're using can help a lot, because some of their installers are quite good. For a "I wanna play with everything locally" scenario, canonical's juju crap is ok if you like Ubuntu, RDO is good if you like Redhat and puppet (the entire installer is basically standalone puppet), Rackspace's stuff is good if you like chef, and devstack is reasonable for a potentially broken setup which doesn't install service scripts or anything. I'm happy to give answer files for RDO. If you have access to horizon (the web interface, which is pretty friendly for new users), you can check the list of endpoints. If not, "keystone service-list && keystone endpoint-list" will give you a list of services, their ids, and a list if api endpoints (with ids). Absolutely every component in openstack talks across APIs, and the list of services and endpoints will tell you, "hey, if I connect to 10.0.0.2, it's running glance/heat/whatever, and I can ssh there and look at the configs to learn what a running config looks like and how we're set up". This matters because openstack is like the standard oil political cartoon octopus, and the default config files are thousands of lines long (including comments), and even the developers of those projects have not mapped their murky depths. But seriously, don't try to read the default configs to try to guess common/basic settings. Everything talks to keystone. You get a token (keystone_rc) which says you're allowed some set of permissions to some projects/tenants. When you start a VM with Nova (let's ignore the "I'm just a neutron/glance/whatever" thing for now), keystone auths you. Nova pulls an image from glance to the target machine, maps storage from cinder if you told it to, starts up, gets an address from a network segment neutron manages dnsmasq on, pulls metadata for cloud-config (usually including an ssh key for a predefined user), and you're in. These images are basically sysprepped Linux disks waiting to be configured, but there's no installer phase. They boot and go, with expected configuration to happen from cloud config and/or a configuration management system. Neutron and SDN is a big topic. Neutron has defined networks, which are virtual, and which are segmented with vxlan or GRE or vlans. They're isolated from each other in the kernel with network namespaces. There are virtual routers for how these networks can talk to each other (which are like real routers conceptually), and you can assign "floating IPs" which are (generally, but not always) publicly routable. There's also a software firewall of "security groups" which allow and disallows traffic on rules that you define. There are ancillary bits for making guests sort-of HA, and autoscaling (with haproxy balancing), and billing tenants for CPU time, and metrics, and some other stuff, but you may not be using any of that. The big thing is that your whole workflow needs to operate around it, in the same way as AWS or GCE or whatever. It's not traditional virt, and you should have relatively anonymous VMs which configure themselves and join application pools when provisioned. Ideally, you'd also have Dev working on the same images, with Jenkins or Travis or something automatically testing new releases against those images. You should be using a configuration management system. There's a lot of "shoulds". Any specific questions?
|
|
#
?
Jul 30, 2015 06:38
|
|
|
minato posted:If you have to administer OpenStack (as opposed to just use it), then go ahead and shave off all your hair because you're going to lose it anyway. It's a goddamn nightmare. Nope, I'm a Windows guy. evol262 posted:Any specific questions? I think that should be enough for me to start asking questions at work. I didn't realize there are different versions and vendors for OpenStack so I'm one step closer to sorta knowing what's going on.
|
|
#
?
Jul 30, 2015 06:48
|
|
|
Dr. Arbitrary posted:I didn't realize there are different versions and vendors for OpenStack so I'm one step closer to sorta knowing what's going on. It's basically all the same (or the configs and projects are the same, and it all works together the same). The differentiation is in Red Hat or Canonical or Rackspace or Mirantis or VMware saying "we know configuring all of this is really confusing and awful, so we've provided a convenient, sensible way for you to define configuration for 20 projects that pretend to be unrelated but really aren't". So rackspace gives nice chef tools, and we give nice puppet stuff, and somebody gives salt stuff (maybe the salt people? I don't know, but theirs is really good), etc. That's the vendor differentiation, basically.
|
|
#
?
Jul 30, 2015 06:53
|
|
|
evol262 posted:It's basically all the same (or the configs and projects are the same, and it all works together the same). The differentiation is in Red Hat or Canonical or Rackspace or Mirantis or VMware saying "we know configuring all of this is really confusing and awful, so we've provided a convenient, sensible way for you to define configuration for 20 projects that pretend to be unrelated but really aren't". I just want to bitch about how awful the Red Hat OSP installer is right now. It's poorly documented and the reference architecture it spits out isn't really scaleable. Red Hat's openstack support is about like a proxy for google with a 48 hour delay "yes thank you we saw that as the first hit on google 3 days ago when we opened the ticket!" I wish Red Hat's installer was as good as the Mirantis one. The good news is it's probably going away for the next release.. That release can't come soon enough.
|
|
#
?
Jul 30, 2015 07:12
|
|
|
The RDO community forums are probably better support, unfortunately. We know the installer sucks and that Mirantis' is better (I think I've talked about it here before, too). Staypuft is actually good if you're running Foreman, though
|
|
#
?
Jul 30, 2015 14:57
|
|
|
|
| # ? Apr 23, 2024 12:45 |
|
|
How do you guys reconcile the fact that sometimes on vCenter, a VM may show that a it is relatively underutilized, but the VM owner says that they can't run their app properly because of resource contention issues? I'm starting at a new position managing something like 1000 vm's, about half are staging/qa environments where the VM owners are howling that they don't have the resources do their jobs properly. That doesn't jive with what I'm seeing in both vCenter and vCOPS reports showing that the VM's in question are actually underutilized according to the vCOPS report.
|
|
#
?
Jul 30, 2015 22:21
|
|