Linode transfer/CPU/disk stats

There are tools that you can run inside your linode that give you a lot more information than the Dashboard graphs do. Try munin, for example. It's very good for monitoring memory usage and other metrics.

@hybinet:

There are tools that you can run inside your linode that give you a lot more information than the Dashboard graphs do. Try munin, for example. It's very good for monitoring memory usage and other metrics.

I'm more interested in:
* My monthly network transfer as Linode sees it
* Host (i.e. Xen dom0, Linode's physical box) load and status
* Host job queue length
* Physical disk IO
* My CPU usage as the Linode host sees it *

I'm most interested in the first three, as there's obviously no way to check any even vaguely related metric from domU.

Where did these data feeds go?

• Due to time slicing, there can potentially be large discrepancies between CPU usage reported inside a VM and in its host. If for some reason the VM host gives unfair shares to another VM, or is CPU-bound itself in the host OS / dom0, the guest itself will never see this.

Most of those must be measured from inside the linode (I also recommend Munin). There is no automated way to retrieve them from the manager in most cases.

@Guspaz:

Most of those must be measured from inside the linode (I also recommend Munin). There is no automated way to retrieve them from the manager in most cases.

Maybe I should just open a ticket for this…

I'm looking at

http://www.linode.com/forums/viewtopic.php?t=1558

and

http://www.linode.com/wiki/index.php/Bandwidth

Which state that there is (or was) an XML feed of this information. I know that network transfer/bandwidth usage is available via the accounts.info() API call.

Can ANYONE give me an informed answer on whether these XML feeds have moved, or were discontinued, or what the status is?

And, if they've been permanently shut off, whether there are plans to add at least host (dom0) status/load/job queue length and physical disk IO to the API?

I know Munin very well, though personally I perfer Cacti or Nagios/Icinga with pnp4nagios. I thought I did an adequate job of explaining why some of these metrics are IMPOSSIBLE to collect in the guest, and the others are not reliable.

The official answer is to do it yourself in your linode:

http://forum.linode.com/viewtopic.php?p=30938#30938

You've done an adequate job of explaining why YOU think the metrics are impossible to collect in the guest, but that doesn't mean you're right. Of the metrics that you've listed, they're all either available for accurate measurement from within the guest, not available outside the management interface, or not available in any place.

It works for me, but only via IPv4.

@jantman:

Host (i.e. Xen dom0, Linode's physical box) load and status

Host job queue length

My CPU usage as the Linode host sees it *

• Due to time slicing, there can potentially be large discrepancies between CPU usage reported inside a VM and in its host. If for some reason the VM host gives unfair shares to another VM, or is CPU-bound itself in the host OS / dom0, the guest itself will never see this.
  AFAIK there are ways to measure stolen CPU cycles from inside the VM, though this would have to be supported by the kernel and your measurement tools (munin, nagios, whatever). I've personally never had to worry about receiving an unfair amount of CPU here, and CPU usage measured from inside the VM is usually very close to CPU usage reported in the Dashboard.

Also, why would you want to monitor the host's load and queue length? Seriously, what the host machine is doing is none of your business. I never quite understood the point of showing it in the Dashboard in the first place, since it's always either "low" or "medium" and the difference between the two doesn't affect the VM's performance in any way.

@hybinet:

I never quite understood the point of showing it in the Dashboard in the first place, since it's always either "low" or "medium" and the difference between the two doesn't affect the VM's performance in any way.

Mine shows "idle" :P

@Guspaz:

@hybinet:

I never quite understood the point of showing it in the Dashboard in the first place, since it's always either "low" or "medium" and the difference between the two doesn't affect the VM's performance in any way.

Mine shows "idle" :P

From what I heard, the "host load" readout was correct for UML, got broken and always shown "idle" for Xen, then got removed with the switch to new manager, and readded when some people complained(including me, but I was complaining about lack of the physical host name on the dashboard, not the load status).

Wouldn't be surprised if it again always show "idle" no matter what, I've never saw it say anything else. But maybe Linode's hardware is just THAT good. :)

I've seen it on high, a few days ago a host had a hardware failure and the drives had to be hot-swapped to a backup host (very quickly I might add thanks to the linode-awesomeness).