ng_ether_load tunable for VirtualBox - what type?

[Robert Trevellyan]

@cyberjock mentions adding the ng_ether_load tunable when using bridged networking with VirtualBox, but I don't see any mention of what type of tunable it is. I see three options: Loader, rc.conf and Sysctl.

 

[dlavigne]

Any driver load will be a loader.

 

[Robert Trevellyan]

Thanks!

 

[cyberjock]

@cyberjock mentions adding the ng_ether_load tunable when using bridged networking with VirtualBox, but I don't see any mention of what type of tunable it is. I see three options: Loader, rc.conf and Sysctl.

Did you read the guide closely? None of the steps are required anymore because they are built into FreeNAS now. ;)

 

[Robert Trevellyan]

No, I didn't read the guide carefully, because I knew there was a jail available. However, I'm having some stability issues with my VMs, so I went back and noticed the part about the tunable, and I am using bridged mode networking. Is the tunable no longer necessary?

 

[cyberjock]

No, I didn't read the guide carefully, because I knew there was a jail available. However, I'm having some stability issues with my VMs, so I went back and noticed the part about the tunable, and I am using bridged mode networking. Is the tunable no longer necessary?

Nope. Just like you don't have to provide the files (and if you did replace the files that came with FreeNAS by default there's a good chance that's why your system is unstable). I'd recommend you not skim through stuff and read. Failing to read can create major problems, up to and including data loss.

/facepalm

 

[Robert Trevellyan]

I didn't provide or replace any files, I just installed the jail, configured storage for it and started using it. I don't "skim stuff", but at the end of the thread you said,

I'm locking this thread as this thread is no longer necessary since the Virtualbox template is the de-facto standard for which you should be using Virtualbox.

so I saw no point in studying all 5 pages in depth.

Anything I did wrong here?

 

[cyberjock]

Well, you clearly read some of it if you were trying to add a tunable. I thought the first line of that guide would have made things obvious...

View attachment 4022

NOTE: 9.2.1.6 nightly dated 5/24/2014 as well as 9.2.1.6-RELEASE has support for Virtualbox out-of-the-box with only basic configuration required. See post #2 in this thread for those directions. You should still read this whole thread so you understand how some of this works as the template is basically all of this stuff done for you automatically. The template will make things much less hacky...

Because 9.2.1.6+ will include virtualbox as a template for installation there is no reason for anyone to use this guide. If you aren't on 9.2.1.6+ you should be upgrading before trying to use Virtualbox anyway.

Seems pretty straightforward that there is nothing in that guide that you should be doing. That guide doesn't even tell you how to install the plugin. ;)

 

[Robert Trevellyan]

I was paying attention when you first announced your work on a VirtualBox jail, and I was still paying attention when it became part of the RELEASE. The part that seems ambiguous to me is this:

9.2.1.6-RELEASE has support for Virtualbox out-of-the-box with only basic configuration required. See post #2 in this thread for those directions.

So I followed the directions in the manual, and when I ran into trouble I decided to look back to that thread to see if there were any tips. Because I use bridged networking, it seemed appropriate to do some additional "basic configuration". I guess I was wrong.

/facepalm

<shrug>

 

[cyberjock]

What directions were you following in the manual? The section on jail installation? Can you provide a link?

Trying to figure out if the documentation needs clarification on something or if there is a more seamless way to explain everything.

 

[Robert Trevellyan]

http://doc.freenas.org/9.3/freenas_jails.html#using-the-phpvirtualbox-template

IMO the directions there are straightforward, I was just attempting a little self-help with VM stability issues.

 

[cyberjock]

When you say VM stability are you saying that FreeNAS was unstable, the jail was unstable, or the VM OS was unstable?

If the VM OS, which OS?

 

[Robert Trevellyan]

The VM OS, which is Ubuntu Server 14.04 LTS x64. Sometimes I log into phpVirtualBox and discover that a VM is in the Aborted state. I admit that this has sometimes been the result of restarting FreeNAS without shutting down VMs (hence my feature request), but I'm positive it has also happened without any obvious cause. It also happens frequently if I enable or disable Remote Desktop on a running VM, so now I just avoid doing that.

 

[cyberjock]

You running it on that RAIDZ2 system with 16GB of RAM?

 

[Robert Trevellyan]

I am. Each VM is allocated 512MB. Usually only 1 is running, but sometimes I fire up 1 or 2 more for a while. They are used as local web servers for in-house development, never more than 2 simultaneous users.

Your question leads me to expect that your response will be to say my system is inadequate, but maybe I should just wait and see.

Here's how it looks right now with 3 VMs running:

The MFU/MRU ratio looks bad because I just dumped a couple of DVD images onto the machine.

Edit: I forgot, last night I created a VM with 1GB of RAM.

 

[cyberjock]

Your question leads me to expect that your response will be to say my system is inadequate, but maybe I should just wait and see.

Sort of. If the pool can't perform because it gets I/O locked then the hypervisor will decide that the disks are no longer accessible via a timeout and just "turn off the machine". I had that problem on my old RAIDZ2 and RAIDZ3 system, which is why I don't even bother trying to run VMs on my system anymore... even though I have 32GB of RAM.

Even if you don't care if its slow, the hypervisor does, and it will terminate the VMs as it sees fit. That's the only reason I can think of for a VM to be killed in the fashion you are describing.

The only solution is to either add gobs and gobs of RAM to avoid the problem, or have lots of mirrored vdevs. With 4 disks you can do two vdevs and that will double your theoretical iops. But that's also no guarantee the problem won't still happen, and all of this is just going off of what I've seen before. The virtualbox logs should tell you why the VM was killed.

 

[Robert Trevellyan]

OK, I appreciate the insights.

I'm seeing things like this in some individual VM logs:

Code:

00:00:08.794723 AioMgr0-N: Request 0x0000081086b620 failed with rc=VERR_TRY_AGAIN, migrating endpoint /media/IWP/InfiniteWP.vdi to failsafe manager.
00:00:08.794950 AIOMgr: Flush failed with VERR_TRY_AGAIN, disabling async flushes

In VBoxSVC.log I see things like this:

Code:

28:05:25.736550 nspr-4   ERROR [COM]: aRC=NS_ERROR_FAILURE (0x80004005) aIID={c20238e4-3221-4d3f-8891-81ce92d9f913} aComponent={ProgressProxy} aText={Error info is not available, operation is still in progress}, preserve=false

Unfortunately searching on some of these terms isn't helping me to understand what's going on.

If necessary I would sooner move VirtualBox to a different system rather than reconfigure my pool (and my RAM is maxed out on the T110).

 

[cyberjock]

Yeah, those errors mean nothing to me. :(

No clue what those mean. You'd have to ask the Virtualbox guys. I've been to the virtualbox IRC before and the guys there are pretty friendly. Just have to be patient as they don't always answer right away.

 

[Robert Trevellyan]

I was hoping to see state transitions to ABORTED or something like that, but nothing doing.

I will look for any correlation between Aborted VMs and scheduled scrubs, which are the only I/O intense thing my box does as far as I know.

By the way, the Aborts that relate to enabling remote desktop show up in FreeNAS security run outputs as "pid xxxxx (VBoxHeadless), uid 1001: exited on signal 11", which seems to be a known issue with the third party VNC server code.

Thanks for your help.

 

[Robert Trevellyan]

I just wanted to resurrect this thread with another plausible hypothesis [tm].

It's been a long time since any of my VMs wound up aborted for no apparent reason, and I've been trying to figure out why. I had previously tested with and without "Use host I/O cache" checked, without being convinced that one was better for stability than the other (for now I'm keeping it checked).

Late last night it dawned on me that it's also been quite some time since I replaced a drive that had started throwing errors during a long copy operation, leading to long retry delays during the copy. Therefore, my hypothesis is that this drive may have been causing I/O delays long enough to upset VirtualBox, leading to aborted VMs.

The corollary to this would be, even in a modest home FreeNAS like mine, the benefit of TLER support is tangible.