WSL Becomes Non Responsive, Uses High Amounts Of CPU & Memory

[ThatRex]

Version

Version 10.0.19044.2364

WSL Version

• WSL 2
• WSL 1

Kernel Version

5.15.79.1-microsoft-standard-WSL2

Distro Version

Ubuntu 20.04

Other Software

No response

Repro Steps

1. Install WSL
2. Start Ubuntu (does not seem to depend of the distro)
3. Wait for an undefined amount of time?

Expected Behavior

To not become non responsive and use up high amounts of CPU and memory after a period of time.

Actual Behavior

WLS becomes non responsive and use up high amounts of CPU and memory after a period of time. When this happens I am unable use or interact with WSL and the WSL CLI. In order to restart WSL I have to restart my PC. Before restricting the amount of CPU and Memory in the WSL config it would make my whole system lag out.

Diagnostic Logs

No response

[cristianpuddu]

same problem. I use Vscode, Docker Desktop and run containers for lamp development.
Sometimes It works again with "wsl -.shutdown" and restart docker desktop, but I often have to restart the pc

[dvdmrn]

I have the same problem, this seems to happen after my computer goes to sleep and I leave WSL running. I do what @cristianpuddu does to mitigate it

[ShaleenAg]

similar to this #9383 except in my case disk usage balloons up and as soon as I force stop my react app, WSL becomes responsive again.

[ThatRex]

same problem. I use Vscode, Docker Desktop and run containers for lamp development. Sometimes It works again with "wsl -.shutdown" and restart docker desktop, but I often have to restart the pc

Unfortunately for me wsl --shutdown and other WSL commands dont do anything as the WSL cli is completely non responsive when this issue occurs.

[ThatRex]

I have the same problem, this seems to happen after my computer goes to sleep and I leave WSL running. I do what @cristianpuddu does to mitigate it

hmmm, I do put my PC in hibernate, sometimes multiples times a day, so maybe that has something to do with it.

[codebymikey]

Just like @ThatRex, I also hibernated my PC fairly regularly, however this never led to any non-responsiveness until around last week, when my Ubuntu 18.04 image started prompting me to run wsl --update .

So it's probably something to do with that recent WSL update.

[dvdmrn]

same problem. I use Vscode, Docker Desktop and run containers for lamp development. Sometimes It works again with "wsl -.shutdown" and restart docker desktop, but I often have to restart the pc

Unfortunately for me wsl --shutdown and other WSL commands dont do anything as the WSL cli is completely non responsive when this issue occurs.

@ThatRex What I typically do is force quit WSL and run this command in PowerShell

[mklueh]

I'm constantly running into this problem, using Jetbrains Gateway.

When

wsl --shutdown

does not work anymore and hangs itself, I use

taskkill /f /im wslservice.exe

But now, after I did this, Jetbrains Gateway does not work anymore at all for one of my projects where the freeze happened, even after re-installation.

There is also this issue with a lot of active comments, that has been closed for whatever reason!

#8529

[sba923]

Same here.

wsl --shutdown hangs

I hibernate the PC every day at the end of my work day, so maybe that's linked with the fact I had a Ubuntu WSL tab open in Windows Terminal?

Restarting the LxssManager doesn't help.

Can't restart wslservice.exe.

[cristianpuddu]

same problem. I use Vscode, Docker Desktop and run containers for lamp development. Sometimes It works again with "wsl -.shutdown" and restart docker desktop, but I often have to restart the pc

i did other tests, in my case the problem seems to occur only when i use vscode.

[thangnq1001]

Try my steps: #8725 (comment)

[benzman81]

Same problem, also hibernating (nowadays, who shuts down a computer anyway if you not have to, i.e. win update or some installation). We use rancher desktop with wsl and we constantly need to restart wsl and wait for all kubernetes pods to spin up again when this problem occurs. This is pretty annoying and seems to have started with some latest windows update.

[jeffska]

Just dropping in from #8696 which has 150+ comments and still no official response.

[OneBlue]

Thanks everyone for all the info.

We have introduced a new feature to diagnose issue like this in wsl 1.1.0: wsl.exe --debug-shell

We have a couple theories on what the root cause of this could be, but we need more information from a live repro to move forward. Instructions:

• Update to Store WSL 1.1.0 with: wsl.exe --update --pre-release (Administrator prompt)
• Reproduce the issue where WSL is stuck
• Open a debug shell via: wsl.exe --debug-shell
• Inside that shell, run (just paste into the shell):

stack_log=stacks.txt
fd_log="fd.txt"
for pid in $(pgrep init); do
  echo -e "\nProcess: $pid" >> "$stack_log"
  echo -e "\nProcess: $pid " >> "$fd_log"
  cat "/proc/$pid/cmdline" >> "$fd_log"
  echo -e '\n' >> "$fd_log"
  for tid in $(ls "/proc/$pid/task" || true); do
    echo "tid: $tid" >> "$stack_log"
    cat "/proc/$pid/task/$tid/stack" >> "$stack_log" || true
  done

  ls -la "/proc/$pid/fd" >> "$fd_log" || true
done

ss -lap --vsock > "sockets.txt"

echo "stacks:"
cat stacks.txt

echo "fds:"
cat fd.txt

echo "sockets:" 
cat sockets.txt

echo "dmesg:"
dmesg

echo "meminfo:"
cat /proc/meminfo

• Share the output of that script with a dump of wslservice.exe on this issue (TaskManager -> Details -> select wslservice.exe -> Right click -> Create dump file)

Note: The debug shell is not user friendly. By design it doesn't use an hvsocket relay like the other shells, but a named pipe. That means that it should work even if the service is completely crashed, as long as the VM is still running. Sometimes the prompt doesn't show, if that's the case, pressing 'return' should be enough to display it.

If the VM is completely crashed, the debug-shell will just remain empty. If that's the case, please capture a dump of the wsl.exe --debug-shell process and share it here

[alex-reach]

@OneBlue I followed your instructions above. My DMP File is too big to be shared here (only 25 MB allowed). You can download it here: https://bitbit.de/wslservice.zip

[OneBlue]

@OneBlue I followed your instructions above. My DMP File is too big to be shared here (only 25 MB allowed). You can download it here: https://bitbit.de/wslservice.zip

Thank you @alex-reach. Can you also share the script's output ? We need both the script output and the .dmp file to investigate this issue

[alex-reach]

@OneBlue I followed your instructions above. My DMP File is too big to be shared here (only 25 MB allowed). You can download it here: https://bitbit.de/wslservice.zip

Thank you @alex-reach. Can you also share the script's output ? We need both the script output and the .dmp file to investigate this issue

wsl_debug.txt

@OneBlue Sure, I attached the script output now

[OneBlue]

Thanks a lot @alex-reach. Pasting my answer from the other issue to get more information:

With this we can see that the issue appears to be related to virtio (since that's where the guest is stuck).

We need to dig deeper to understand why things are stuck here though, let's gather more information. Can you please:

Repeat the same steps as before to get wsl hung and share the wslservice.exe dump and the script output again
Run the following command and share its output (elevated cmd.exe): tasklist /m vp9fs.dll
Its output should look like this:
tasklist /m vp9fs.dll

Image Name PID Modules
========================= ======== ============================================
dllhost.exe 28572 vp9fs.dll
vmwp.exe 4956 vp9fs.dll
The second column gives you the PID of the processes (there should be one vmwp.exe, and one or two dllhost.exe). Please dump all of them (Use the PID to differentiate if there are processes running with the same name) and share the dumps here.

To summarize, we need:

The output of the script I shared #8696 (comment)
The output of tasklist /m vp9fs.dll
The dumps for wslservice.exe, and all the vmwp.exe, wmwpWSL.exe, and dllhost.exe processes returned by tasklist

[alex-reach]

@OneBlue Sure, here we go:

You can download everything here:

https://www.bitbit.de/wsl-debug.zip

But I have no "wmwpWSL.exe" running ..

[EvanSchalton]

Just redirected here from #8529
I'm having the same issue, someone over there indicated it might be a ram issue -- I have 64GB of ram and nothing else runningm, 56 is accessible via the .wslconfig (w/ 16 slot for good measure) -- I don't think it's a ram issue

[EvanSchalton]

For me this seems to happen when my build context is over 24mb (despite 64GB ram, 16GB Swap & 128GB SSD space) -- I suspect, thought I'm not sure how to validate that the build kit is hitting a recurssion issue w/ symlinks in the build dir and isn't exiting properly -- I think as my build context gets larger I loose insight into the dir structure so I've been conflating the two issues -- if I have time I'll create a minimal example w/ a recursive symlink build context and see if it causes a hang

[OneBlue]

Thank you @alex-reach.

This is really interesting because your repro is a bit different from the one in #8696.

In your case, dllhost.exe is still there after resuming from hibernation, but it doesn't seem to be notified properly.
Let's capture some logs to figure out why. Can you please:

• Start a log collection (see 8) Collect WSL logs (recommended method)
• Reproduce the issue
• Stop the log collection
• Run the debug shell script I shared earlier
• Dump wslservice.exe, dllhost.exe and vmwp.exe as I explained earlier
• Share the dumps, the debug shell output, and the log result here

[alex-reach]

@OneBlue Thanks for your feedback. I had to revert in the meantime to WSL 1.0.3.0, because I had issues with the Apache Service stopping to respond frequently in the preview WSL 1.1.0.0. I will install the preview again in a few days to send you a new log collection.

[fontanka16]

wsl.exe --debug-shell has as little effect as wsl --shutdown for me (none). I have been waiting for five minutes.
Whe looking into TaskManager, there is a lot of activity though:

I am not sure if any of the above commands kicked off that level of CPU usage

[N0xFF]

• Start a log collection (see 8) Collect WSL logs (recommended method)
• Reproduce the issue
• Stop the log collection
• Run the debug shell script I shared earlier
• Dump wslservice.exe, dllhost.exe and vmwp.exe as I explained earlier
• Share the dumps, the debug shell output, and the log result here

@OneBlue I repeated all your steps and collected logs and dumps.

My steps to reproduce:

1. Run Docker Desktop.
2. Open Bash.
3. Sleep (not hibernate).
4. Wake up.
5. Collect logs and dumps.

-> LOGS

Doesn't have dllhost.exe for vp9fs.dll:

PS C:\WINDOWS\system32> tasklist /m vp9fs.dll

Image Name                     PID Modules
========================= ======== ============================================
vmwp.exe                     11812 vp9fs.dll

[jak-sdk]

This happens to me daily, for me it seems to be excessive interrupts from Hyper-V?

top - 10:21:03 up 10:07,  3 users,  load average: 1.04, 1.20, 1.12
Tasks:  72 total,   1 running,  71 sleeping,   0 stopped,   0 zombie
%Cpu0  :  0.3 us,  0.0 sy,  0.0 ni, 99.7 id,  0.0 wa,  0.0 hi,  0.0 si,  0.0 st
%Cpu1  :  0.0 us,100.0 sy,  0.0 ni,  0.0 id,  0.0 wa,  0.0 hi,  0.0 si,  0.0 st
%Cpu2  :  0.4 us,  0.0 sy,  0.0 ni, 99.6 id,  0.0 wa,  0.0 hi,  0.0 si,  0.0 st
%Cpu3  :  0.0 us,  0.0 sy,  0.0 ni,100.0 id,  0.0 wa,  0.0 hi,  0.0 si,  0.0 st
%Cpu4  :  0.9 us,  0.0 sy,  0.0 ni, 98.2 id,  0.0 wa,  0.0 hi,  0.9 si,  0.0 st
%Cpu5  :  0.0 us,  0.0 sy,  0.0 ni,100.0 id,  0.0 wa,  0.0 hi,  0.0 si,  0.0 st
MiB Mem :   5930.4 total,   1543.8 free,    789.2 used,   3597.4 buff/cache
MiB Swap:      0.0 total,      0.0 free,      0.0 used.   4855.8 avail Mem

Perf:

   PerfTop:    4561 irqs/sec  kernel:99.4%  exact:  0.0% lost: 0/0 drop: 0/0 [2200Hz cycles],  (all, 6 CPUs)
---------------------------------------------------------------------------------------------

    31.01%  [kernel]          [k] hv_ce_set_next_event
    23.17%  [kernel]          [k] native_apic_mem_write
    22.73%  [kernel]          [k] asm_sysvec_hyperv_stimer0
     3.16%  [kernel]          [k] read_tsc
     1.96%  [kernel]          [k] __sysvec_hyperv_stimer0
     1.88%  [kernel]          [k] read_hv_clock_tsc
     1.80%  [kernel]          [k] add_interrupt_randomness
     1.66%  [kernel]          [k] hrtimer_interrupt
     1.15%  [kernel]          [k] clockevents_program_event
     0.84%  [kernel]          [k] irq_exit_rcu

# cat /proc/interrupts  | grep HVS
HVS:  254943891  294333063  651585438  293415102  274733401  217126954   Hyper-V stimer0 interrupts

And to compare to normal behaviour:

   PerfTop:      24 irqs/sec  kernel:95.8%  exact:  0.0% lost: 0/0 drop: 0/0 [4000Hz cycles],  (all, 6 CPUs)
---------------------------------------------------------------------------------------------

    24.13%  perf              [.] __symbols__insert
    15.97%  perf              [.] rb_next
     5.58%  [kernel]          [k] kallsyms_expand_symbol.constprop.0
     4.60%  perf              [.] rb_insert_color
     3.60%  perf              [.] kallsyms__parse
     2.68%  libc.so.6         [.] 0x00000000001b2ef9
     2.25%  [kernel]          [k] vsnprintf
     2.06%  libelf-0.186.so   [.] gelf_getphdr

# grep HVS /proc/interrupts
HVS:       3439       3326       3454       4390       3559       3150   Hyper-V stimer0 interrupts

[TheWizier]

+1

[microsoft-github-policy-service]

This issue has been automatically closed since it has not had any activity for the past year. If you're still experiencing this issue please re-file this as a new issue or feature request.

Thank you!