Hi All,
Since upgrading a few hypervisors to Debian buster + Xen (11.2 & then
4.12.1), we have noticed a memory leak.
The qemu-system-i386 for each guest continues to consume memory, and
this increases over time. In some guests it takes a week or two to
become large (consuming many hundreds of MB of "RES" memory in top, and
many GB of VIRT). In others, I can literally watch the RES size grow
and grow every few seconds. There is one particular guest that bursts of
http requests incoming, and it leaks many MB per minute.
I am aware of some kernel patches, so we have compiled latest dom0
mainline kernels (4.9, 4.14, 4.19, and 5.2.3) - and the leaks continue
The only success we have had in containing (at least slowing down
dramatically) the leaks are by using qemu 2.11.2 with xen 4.11.2 .
Some additional information:
Didn't see this issue with Stretch + xen 4.8.5 (qemu 2.8)
We are using rbd for the back-end storage.
We had to compile by hand, because we hit these issues (and patch
against them):
https://lists.xenproject.org/archives/html/xen-devel/2018-02/msg01946.html
and
https://www.mail-archive.com/xen-devel@lists.xenproject.org/msg51177.html
We have tried the master branch of xen , however this didn't work for us
(haven't tried with 5.x dom0 kernel however).
I'm not sure if it is dom0 kernel issue, or qemu, xen, or the way it
talks to it. Has anyone else hit an issue like this? I'm keen for
advice/suggestions. There are some downsides on running the older
version of qemu (security, and we can't get the disk/stats from xl top)
Many thanks,
Andrew.
_______________________________________________
Xen-users mailing list
Xen-users@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-users
Since upgrading a few hypervisors to Debian buster + Xen (11.2 & then
4.12.1), we have noticed a memory leak.
The qemu-system-i386 for each guest continues to consume memory, and
this increases over time. In some guests it takes a week or two to
become large (consuming many hundreds of MB of "RES" memory in top, and
many GB of VIRT). In others, I can literally watch the RES size grow
and grow every few seconds. There is one particular guest that bursts of
http requests incoming, and it leaks many MB per minute.
I am aware of some kernel patches, so we have compiled latest dom0
mainline kernels (4.9, 4.14, 4.19, and 5.2.3) - and the leaks continue
The only success we have had in containing (at least slowing down
dramatically) the leaks are by using qemu 2.11.2 with xen 4.11.2 .
Some additional information:
Didn't see this issue with Stretch + xen 4.8.5 (qemu 2.8)
We are using rbd for the back-end storage.
We had to compile by hand, because we hit these issues (and patch
against them):
https://lists.xenproject.org/archives/html/xen-devel/2018-02/msg01946.html
and
https://www.mail-archive.com/xen-devel@lists.xenproject.org/msg51177.html
We have tried the master branch of xen , however this didn't work for us
(haven't tried with 5.x dom0 kernel however).
I'm not sure if it is dom0 kernel issue, or qemu, xen, or the way it
talks to it. Has anyone else hit an issue like this? I'm keen for
advice/suggestions. There are some downsides on running the older
version of qemu (security, and we can't get the disk/stats from xl top)
Many thanks,
Andrew.
_______________________________________________
Xen-users mailing list
Xen-users@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-users