I received the following message from my linode early this morning:



The message arrived at 1:39 am, and here's the corresponding graph from my linode panel:



As you can see, disk i/o briefly peaked to 1123.27 and did not 'average' 1123.27 over any two-hour period.

Not a big deal, just thought I'd point it out.

I've gotten the same think - I believe in my case it was prelink running for the first time (via cron) - first time it runs, it does hit the box pretty hard (I disable it on slow hardware, where the benefit seems to be nil and the performance hit while it does its thing is tremendous)

I wouldn't worry about the I/O reports unless you get them daily.

Maybe I didn't explain very well.

I'm not concerned about the alert message. I know what caused the IO spike.

My point is that the alert message is incorrect. It says that my linode was "...averaging 1123.27 for the last 2 hours". But it wasn't. It peaked once to 1123.27. The average over any 2-hour period would be significantly less than that.

Depending a very high number can easily offset your average "threshold score". Might be helpful to do the math. I don't have the time to do it right now.

What is your threshold limit?

The email notice doesn't give units, which it probably should. The units are not KB/s; it happens to be a coincidence that that's about the number where your graph peaks. The email is talking about I/O operations per second. I have no feel at all how this translates into real-world performance, so I don't find it terribly useful, and I have no idea what threshold to set for my alert.

It's a mind-boggling coincidence that the email states "1123.27" and the graph states "1123.27" for "Max"

Wow, I hadn't notice that was the exact number on the graph!

Agree with sleddog, 'averaged' in the disk io alert looks more like 'exceeded'.

I'm setting up a new cluster at Linode, so I've been paying attention to the alerts for the past few weeks. Every single time I get one, the machine is idle, almost no disk activity, and Munin and the Linode graphs show a very very brief spike on an otherwise empty graph.

Is this alert a relic of UML? Is it broken? Am I missing something?

Thanks,
Graham