Antsle Forum

Welcome to our Antsle community! This forum is to connect all Antsle users to post experiences, make user-generated content available for the entire community and more. 

Please note: This forum is about discussing one specific issue at a time. No generalizations. No judgments. Please check the Forum Rules before posting. If you have specific questions about your Antsle and expect a response from our team directly, please continue to use the appropriate channels (email: [email protected]) so every inquiry is tracked. 

Please or Register to create posts and topics.

Performance issues related to KVM (possibly)

I was seeing a situation where, every evening, the load average on the hypervisor (that is, the antsle itself) would start to climb, even though CPU and I/O were virtually idle.

Googling around, I saw that this was not an uncommon KVM/QEMU issue, and one recommendation was to enable the QEMU Guest Agent in the KVM guests. This in turn requires a virtio-serial device added to the guests' domain configurations with virsh edit.

Since implementing this change, it appears so far as if this has, in fact, fixed the issue, although I've been messing around with things so much (this is a brand new device, so I'm playing a lot) that it may be too early to be certain. Either way, it seems to me it would be a good thing to have these things added by default to KVM templates and domain files, so this is a request to do so!

I will update this thread in a few days with further information, positive or negative, regarding this experiment, since I see other people have had issues with load average spikes or occasionally having their units become unresponsive (which happens if the load average spikes too high!)

Either way, it seems to me it would be a good thing to have these things added by default to KVM templates and domain files, so this is a request to do so!

@mshappe, thanks for the helpful post. I'll ask some team members to take a look at this and see if we can implement it in the templates. Please keep those updates coming.

mshappe has reacted to this post.
mshappe
Quote from Daniel Scott on February 10, 2020, 7:44 am

@mshappe, thanks for the helpful post. I'll ask some team members to take a look at this and see if we can implement it in the templates. Please keep those updates coming.

Feel free to reach out to me (or have them do so) if they have any questions. I'm still trying to find good docco as to WHY these pieces should make a difference (and more to the point, why KVM causes the load average spikes without them), but so far, it does appear to be the solution.

As an update: while it is possible that the KVM changes above improve performance and make odd behaviour less likely, I have encountered an entirely unrelated-to-KVM (I think) issue with the NVM driver, or hardware, on my device, which I have filed tickets for with support. For the time being, at their request, I have pulled the nvme cache partition out of the zpool while they dig into possible causes.

Stay tuned!

@mshappe @ddmscott

Did this ever get resolved?  I have a One XD where the NVME keeps running into issues and essentially going offline.  I have the logs and cache running on it for each Zpool.

 

Thanks for any insight on this.

Unfortunately, this wound up being a hardware issue, that had to be addressed by sending it back to have the NVMe drive replaced. It's been running like a dream since, tho'.

lancem has reacted to this post.
lancem