stottle

It is hard to recommend a GPU in general. Best I can say is that I am doing GPU passthrough with an nvidia 750 TI. It works for what I need it to. As for passthrough, once you pass the device to your VM, it works in the VM. You should be able to connect HDMI/DVI/VGA to it. It isn't available to any other "machine" though, host or another VM. Not sure what you are asking in terms of multi-monitor.

Is anybody using docker compose? Are there any plans to integrate it with unRAID?

Thanks for the help. Been reading up today. I didn't find the official docker user guide until after I started this thread, unfortunately!

Ahh, I looked at a couple of example dockfiles and they were pointing to 0.9.15, which is what is recommended on one of the follow-up pages of this thread. Looks like you've already updated, so nevermind!

It looks like the current version of phusion/baseimage is 0.9.16. According to the release notes, it disables ssh. Is there a reason the recommended version is 0.9.15 instead? According to the release notes:

Gerrit is a bit like GitHub, a wrapper around git that provides a web interface and code review. Running it locally means not having all of that (potentially proprietary) code in the cloud. Jenkins is a Continuous Integration (CI) tool which can integrate with git/Gerrit to run automated tests when code is checked in. Both are common workflow tools for software developers. Maybe I'm misusing the webui term. I think of it as a ui for setting up the docker. If that is the case, it would be convenient to provide a UI to set up (for instance) the authorization type for Gerrit. If it is meant to be a front-end to the running docker, I don't think it applies to Gerrit/Jenkins. For Gerrit, I assume it would be necessary to customize the docker to put the database in /mnt/user/appdata, and not have it deleted if the docker is updated. Still learning.

Yep, that would be them. I found the same dockers you did. Correct me if I'm wrong, but it seems like there are several things unRAID is doing that are different from just grabbing a docker from somewhere. 1) Ideally, the docker would be based on the unRAID friendly OS (Phusion currently recommended) 2) Providing a webui 3) Since the host OS is known, there are several best practices (user accounts, etc) that are recommended [if it isn't clear, a lot of my information about dockers comes from the guide] I could create a linux VM and set up Gerrit myself. But it seems like dockers are a better approach in this case. I'm just struggling with the initial learning curve.

I see there are dockers for Jenkins and Gerrit out there. Is it difficult to convert these for use with unRAID (phusion and a web ui)? I've never created a docker before, but the instructions for Gerrit (simple setup anyway) look pretty straight forward. I'd appreciate some pointers on getting started. Thanks

One of two things come to mind. 1) On windows, the way a virtual disk is exposed from kvm to the guest isn't known. So you need to use the virtio drivers during OS install before it sees a "disk". I doubt this is your issue, as I *think* Linux includes these "drivers" as part of the OS. 2) You aren't passing the entire SSD drive to the VM, only a disk image. There could be a setting wrong in your xml there, or the file itself (on disk on the cache drive) could be locked (wrong permissions, not writable). Check the file permission, and if that doesn't help, post the disk part of your config.

Maybe this should be stickied?

+1 from me. This would be really handy. It seems like it is close to what you already have for cdrom images. Also, there are no changelog entries for post #1 for the 2015.03.05a release. Can you tell us what is different? Thanks for the great work!

The top post says this can be done in the webgui. It isn't obvious how in the new dynamix gui. I went with the command line version instead, to create a non-array disk for VMs. Maybe the instructions need an update?

When installing a Windows 7 guest, I use the virtio drivers twice. Once to recognize the iso during installation of the OS, and then after the OS is installed to work with drives (like an optical drive). Can someone explain the difference between the two options (scsi and scsi-passthrough) and when you should use one vs. the other? Thanks!

Nice! I will have to give that a shot.

That doesn't help. But it looks like the USB ids changed since I started the VM. I'm wondering if this is a (virtualization) KVM issue. That is, when I first started the server, the two instances were bus 2, device 4 and bus 2, device 7. Now they are still on bus 2, but are devices 17 and 22.

I did indeed update the firmware. Different set of lines? You mean different usb port? The KVM switch works fine between multiple windows machines not running in VMs. It just seems to have trouble with unRAID.

I've got a 4 port KVM switch (http://www.newegg.com/Product/Product.aspx?Item=N82E16817399062). I've got it connected to a desktop computer, unRAID and an unRAID VM (using the *other* KVM). When I start the unRAID server, keyboard input works fine. But if I switch away and come back to it, no input is accepted. The video works, i.e. I can see the prompt for unRAID. If I telnet to the server, I can interact with it, but I wanted to have the "direct line". Any suggestions for how to resolve this?

I'm making progress on my VM. If I start my unRAID, start libvirt, then start my VM I get passthru of the GPU and Ceton card. I can run it and restart it. But I have problems if I shut down. If shut down and run virsh start or virsh create again, but VM starts (according to virsh list) but doesn't show up on the monitor. Since I can't use VNC with GPU passthru, I have no way to interact with the VM and have to destroy it. Are others having luck with shutdown and power up again? If so, any suggestions? Thanks

You need to telnet into the unRAID or have keyboard/video attached to the base linux OS to use these commands on the system itself. At this point, the graphical plugins don't support the type of VM image you want to use. If you aren't comfortable from the shell, I'm not sure I'd try this. You will want to create your image in an appropriate place, which is not a quick thing to describe.

\o/ Well, it isn't using MSI, but it is working in the VM!

I have the ceton infinitv as well, and am interested in seeing if it works in a KVM virtual windows box. I currently have it in a "bare-metal" windows box, where it is not using MSI. I tried to enable MSI, following the linked pages' advice. I was able to set several other PCI devices to MSI, but for the ceton it didn't take for some reason. The device still works, but device manager lists the IRQ as a positive #, rather than negative. Is it possible I will have different results (on different hardware) with KVM? Should I even bother if MSI isn't supported? Appreciate any advice suggestions.

How do you know it isn't working? Is libvirt giving an error? Or is the VM starting, but networking isn't available? It isn't clear if this is a host or guest problem. Did you install drivers for the networking controlling in the guest?

Thanks, that helps me a lot. I guess I am unusual in not using my cache drive for VMs. Can you give a bit more detail on how I could have virtMan configured to start/stop with host power up/down? For vmMan, I didn't realize that was how it worked. I thought it was closer to your webvirtman plugin. One suggestion would be to have a way to hand an input xml file to libvirt for creating a VM. That way you don't have all of the hardware specific stuff you get from dumpxml. Another suggestion would be to differentiate between settings you can change on-the-fly (like attaching devices) from ones that require the VM to be restarted, as I described in my original post.

I have a VM that works *manually*. The trouble is getting this to work the way I want automatically. Which leads to this question. I'm currently on beta 10a, not sure when 11 will be out, nor how 11 will affect my question. It seems conceptually you have three problems to solve with a VM 1) boot configuration (passthrough hardware like GPU, controllers, able to support shutdown/restart) 2) runtime configuration (attached devices (i.e., usb), triggering vm power state changes) 3) interaction with host OS (auto start of VM, shutting down the VM when the unRAID is shut down) In my mind, these three ought to be handled independently. I'm struggling because at this point, they don't seem to be. I have been testing with my array off. Having had some issues where I had to power off the hardware, this avoids the parity check. I have to manually start libvirt, but then I can use the command line to start/test the VM, without worrying about the xml being modified. As I mentioned elsewhere, I think we should be sharing the XML *input* to libvirt - before libvirt inserts default stuff for your hardware - instead of libvirt's dumpxml output. It isn't clear, though, what happens when I start the array. I think the dmacias' plugins modify libvirt to help with clean startup/shutdown, and with resetting things when the VM is restarted (GPU reset). It isn't clear what this does to an already running VM. Am I right about this? If so, is it possible to do the same thing without needing the array to be on, or manually? It seems like I should be able to validate a VM (step #1 above) without needing the array to be on. I think a web UI for #2 is ideal, both in terms of exposing the available options in human readable form, and making it easy to make changes. Arguably this should be possible without the array being on as well. #3 is where the real value is, in terms of being able to count on the VM being available without intervention. But I'm not sure how to get there. I'm not sure how to step from "my VM works manually" to "my VM works". I think this is because of the extra variables mixed in with the plugins and array state. It seems like the #3 functionality is built into the plugins, but tied to array stop/start instead of host power up/down. I hope this question doesn't offend limetech or dmacias or anyone else. I really appreciate the work going into adding this feature. It feels like I'm waiting for the next beta to fix something, as I feel a bit stuck getting my VM where I want. This is my attempt to put my finger on the problem I'm having. My tendency in troubleshooting is to remove layers, but I'm not sure what the plugins do.

Not working (as in no groups listed in /dev/vfio). My go script is: #!/bin/bash mkdir -p /mnt/btrfs mount /dev/disk/by-id/ata-TOSHIBA_THNSNH128GBST_231S10DWTE4Y-part1 /mnt/btrfs # Start the Management Utility /usr/local/sbin/emhttp & vfio-bind 0000:01:00.0 0000:01:00.1 0000:00:14.0 /boot/unmenu/uu cd /boot/packages && find . -name '*.auto_install' -type f -print | sort | xargs -n1 sh -c I see the following in the log: Oct 12 18:17:12 Tower2 logger: Starting go script Oct 12 18:17:12 Tower2 kernel: BTRFS: device fsid 69be0add-8368-4d88-935d-f2db1e71b57a devid 1 transid 567 /dev/sdh1 Oct 12 18:17:12 Tower2 kernel: BTRFS info (device sdh1): disk space caching is enabled Oct 12 18:17:12 Tower2 emhttp: unRAID System Management Utility version 6.0-beta10a Oct 12 18:17:12 Tower2 emhttp: Copyright (C) 2005-2014, Lime Technology, LLC Oct 12 18:17:12 Tower2 emhttp: shcmd (1): mkdir -p /boot/config/plugins/webGui Oct 12 18:17:12 Tower2 kernel: BTRFS: detected SSD devices, enabling SSD mode Oct 12 18:17:12 Tower2 emhttp: shcmd (2): mkdir -p /boot/config/shares Oct 12 18:17:12 Tower2 unmenu-status: Starting unmenu web-server So I see "Starting go script" and "Starting unmenu" with nothing about vfio-bind. So I'm officially confused!