SOLVED Alert: Control Domain Memory Usage

delaf

I tried to install the kernel as described, but I got an error :

# yum install http://koji.xcp-ng.org/kojifiles/work/tasks/7620/17620/kernel-4.19.19-6.0.12.1.kmemleak.1.xcpng8.1.x86_64.rpm
Loaded plugins: fastestmirror
kernel-4.19.19-6.0.12.1.kmemleak.1.xcpng8.1.x86_64.rpm                                                     |  30 MB  00:00:03
Examining /var/tmp/yum-root-Uyd1Lb/kernel-4.19.19-6.0.12.1.kmemleak.1.xcpng8.1.x86_64.rpm: kernel-4.19.19-6.0.12.1.kmemleak.1.xcpng8.1.x86_64
/var/tmp/yum-root-Uyd1Lb/kernel-4.19.19-6.0.12.1.kmemleak.1.xcpng8.1.x86_64.rpm: does not update installed package.
Error: Nothing to do

The host is a 8.1 up to date (latest patchs are installed but I did not reboot yep after).

stormi

Announcement

Here's an updated ixgbe driver package that is meant as an update candidate without updating to a higher version. If need all feedback we can get on this one, because this is the candidate for the official fix as would be delivered to everyone. Of course the intel-igb-alt will remain available for those who need a more recent driver, and I even moved it to the updates repository so that one doesn't need to add --enablerepo=xcp-ng-testing to install it anymore.

To test the official update candidate on XCP-ng 8.1 or 8.2.

If you had previously installed intel-ixgbe-alt:
- yum remove intel-ixgbe-alt -y
- Check that the /lib/modules/4.19.0+1/override/ixgbe.ko file was properly deleted. I've seen, once, a situation where a .ko file from an -alt package wasn't deleted, so I'm being cautious here and ask you to report you see that it is still present. In theory, this is impossible, but as I've seen it once I don't know anymore
- Run depmod -a
Update the intel-ixgbe package from the testing repo: yum update intel-ixgbe --enablerepo=xcp-ng-testing
Reboot

Note: I'm not 100% sure that I picked the right patch, nor that this patch alone is sufficient.

delaf

FYI : On one host that has the problem I only have 1 Debian VM that does only one thing : netdata. It is a netdata that get flows from other netdata and that is polled by a prometheus server (which is on another host).

Screen Shot 2020-11-13 at 09.26.10.png

delaf

As soon as I stopped the last VM on the host, there is no more memleak.

delaf

More informations:

# rpm -ivh http://koji.xcp-ng.org/kojifiles/work/tasks/7620/17620/kernel-4.19.19-6.0.12.1.kmemleak.1.xcpng8.1.x86_64.rpm
Retrieving http://koji.xcp-ng.org/kojifiles/work/tasks/7620/17620/kernel-4.19.19-6.0.12.1.kmemleak.1.xcpng8.1.x86_64.rpm
Preparing...                          ################################# [100%]
    package kernel-4.19.19-6.0.12.1.xcpng8.1.x86_64 (which is newer than kernel-4.19.19-6.0.12.1.kmemleak.1.xcpng8.1.x86_64) is already installed
    file /boot/System.map-4.19.0+1 from install of kernel-4.19.19-6.0.12.1.kmemleak.1.xcpng8.1.x86_64 conflicts with file from package kernel-4.19.19-6.0.12.1.xcpng8.1.x86_64
    file /boot/config-4.19.0+1 from install of kernel-4.19.19-6.0.12.1.kmemleak.1.xcpng8.1.x86_64 conflicts with file from package kernel-4.19.19-6.0.12.1.xcpng8.1.x86_64
    file /boot/vmlinuz-4.19.0+1 from install of kernel-4.19.19-6.0.12.1.kmemleak.1.xcpng8.1.x86_64 conflicts with file from package kernel-4.19.19-6.0.12.1.xcpng8.1.x86_64
    file /lib/modules/4.19.0+1/kernel/fs/nfs/nfs.ko from install of kernel-4.19.19-6.0.12.1.kmemleak.1.xcpng8.1.x86_64 conflicts with file from package kernel-4.19.19-6.0.12.1.xcpng8.1.x86_64
    file /lib/modules/4.19.0+1/kernel/net/netfilter/nf_conntrack.ko from install of kernel-4.19.19-6.0.12.1.kmemleak.1.xcpng8.1.x86_64 conflicts with file from package kernel-4.19.19-6.0.12.1.xcpng8.1.x86_64
    file /lib/modules/4.19.0+1/modules.order from install of kernel-4.19.19-6.0.12.1.kmemleak.1.xcpng8.1.x86_64 conflicts with file from package kernel-4.19.19-6.0.12.1.xcpng8.1.x86_64

stormi

I built a new kernel with memleak, which should install correctly. I've updated my post with the instructions above.

delaf

@stormi Ok thank you. The installation works.

I made another test: use the kernel-alt 4.19.108 on one box. It seems I did not have the issue anymore!

delaf

On server with kmemleak kernel, I get this error:

# echo scan > /sys/kernel/debug/kmemleak 
-bash: echo: write error: Device or resource busy

Digging a litlle bit, and I found :

# dmesg | grep memleak
[    0.677307] kmemleak: Kernel memory leak detector disabled
[    2.701225] kmemleak: Early log buffer exceeded (5128), please increase DEBUG_KMEMLEAK_EARLY_LOG_SIZE

So kmemleak is still disabled as DEBUG_KMEMLEAK_EARLY_LOG_SIZE seems to be too small
@stormi Could you rebuild a kernel with an increased DEBUG_KMEMLEAK_EARLY_LOG_SIZE?

Thank you!

stormi

Pinging @r1 about this.

r1

@stormi @delaf will check this and do required changes.

r1

@delaf and others, you can download and install and update from link which should work fine.

After the system is running for some time, user can # echo scan > /sys/kernel/debug/kmemleak and then # cat /sys/kernel/debug/kmemleak to see if there are any unreferenced objects floating in memory.

delaf

@r1 installed and it works :

# dmesg | grep kmem
[    6.181218] kmemleak: Kernel memory leak detector initialized
[    6.181223] kmemleak: Automatic memory scanning thread started

I will check the leaks tomorrow.

delaf

@r1 @stormi for now I don't have any memleak detected :

# uname -a
Linux bb0272.aquaray.com 4.19.0+1 #1 SMP Thu Nov 19 15:15:04 CET 2020 x86_64 x86_64 x86_64 GNU/Linux
# yum list installed | grep kernel
kernel.x86_64                   4.19.19-6.0.12.1.2.kmemleak.xcpng8.1
# cat /sys/kernel/debug/kmemleak
# echo scan > /sys/kernel/debug/kmemleak
# cat /sys/kernel/debug/kmemleak

delaf

FYI: I confirm that on the server running the alt kernel I do not have anymore the memleak problem :

# yum list installed | grep kernel
kernel.x86_64                   4.19.19-6.0.12.1.1.kmemleak.xcpng8.1
kernel-alt.x86_64               4.19.108-3.xcpng8.1         @xcp-ng-base
# uname -a
Linux bb0266.aquaray.com 4.19.108 #1 SMP Mon Mar 23 15:04:05 CET 2020 x86_64 x86_64 x86_64 GNU/Linux
#

Memory graph from this server :
Screen Shot 2020-11-20 at 12.14.54.png

Obviously, the alt-kernel has been installed ~2020-11-13.

On the server running the memleak kernel (since yesterday) :
Screen Shot 2020-11-20 at 12.16.21.png

olivierlambert

@delaf Weird but interesting!

r1

@olivierlambert @delaf what we know from kmemleak so far is that it will only scan and report unreferenced objects. If any kernel module / kernel itself is still holding(referencing) the memory then it may not show up. We are evaluating other options to find this.

kernel-alt is more related to upstream, so either this issue is known and fixed in upstream or it might have been introduced from kernel updates.

The oldest kernel available is 4.19.19-6.0.10.1.xcpng8.1, is it possible to install it and see if the issue repeats?

stormi

kernel-alt also means that all the drivers used are buillt-in, no those from the driver packages that take precedence over the built-ins when running the main kernel.

r1

Yes, all drivers are stock kernel modules for kernel-alt. It would be interesting to see the behavior by disabling updates and override. I think we can try both. 1st check if the downgraded kernel shows same symptoms and then disabling update drivers.

stormi

Ok, so now we know for sure that it's related to the kernel or one of the drivers.

Let me summarize all the tests that users affected by the issue can do to help find what causes it:

Test with kmemleak, hoping that it may be able to detect something. No luck for now for @delaf who tried.
Test with the oldest kernel (4.19.19-6.0.10.1.xcpng8.1). I doubt it will yield results, but that would allow to be sure. If it does yield results, then it will allow to search towards a specific patch.
With the current kernel, give priority to built-in drivers. If this gives good results, this will mean that the leak is in one of the drivers that are provided through driver RPMs. Two ways:
- A bit riskier but we'd still be interested in the results: disable them all so that only built-in kernel drivers are used. For this, edit /etc/modprobe.d/dist.conf and change search override updates extra built-in weak-updates into search extra built-in weak-updates override updates, then run depmod -a and reboot. Don't forget to restore the original contents after the tests.
- Another way, that allows to select specific drivers one by one:
  - Identify a few drivers that you want to check in the output of lsmod. For example ixgbe.
  - Find where the currently used driver is on the filesystem: modinfo ixgbe | head -n 1
  - If the path contains "/updates/", it's not a kernel built-in. Rename the file to name_of_file.save. depmod -a. reboot. The kernel will then use its built-in driver.
  - If nothing changes, restore the file and try another.

I also intend to build a new ixgbe driver, just in case we're lucky and it's the culprit, since every affected user uses it.

delaf

@r1 I do not manage to install the old kernel. Any idea?

# yum downgrade "kernel == 4.19.19-6.0.10.1.xcpng8.1"
Loaded plugins: fastestmirror
Loading mirror speeds from cached hostfile
Excluding mirror: updates.xcp-ng.org
 * xcp-ng-base: mirrors.xcp-ng.org
Excluding mirror: updates.xcp-ng.org
 * xcp-ng-updates: mirrors.xcp-ng.org
Resolving Dependencies
--> Running transaction check
---> Package kernel.x86_64 0:4.19.19-6.0.10.1.xcpng8.1 will be a downgrade
---> Package kernel.x86_64 0:4.19.19-6.0.12.1.xcpng8.1 will be erased
--> Finished Dependency Resolution
Error: Trying to remove "kernel", which is protected

Danp

@delaf I ran into the same thing recently. See solution below --

https://xcp-ng.org/forum/post/33291