High CPU usage from `systemd-journal-gatewayd` | Home Assistant | Page 1

heavy trellis Jul 15, 2025, 6:34 PM

#

Hi!
Recently I have been experiencing an issue with my Home Assistant install where the CPU usage will spike to 100% and stay there. This is evident in the local tools within HA, and also on my hypervisor.

This is my system info:

arch: amd64
channel: stable
docker: 28.3.0
features:
- reboot
- shutdown
- services
- network
- hostname
- timedate
- os_agent
- haos
- resolved
- journal
- disk
- mount
hassos: "16.0"
homeassistant: 2025.7.2
hostname: homeassistant
logging: info
machine: qemux86-64
machine_id: <removed>
operating_system: Home Assistant OS 16.0
state: running
supervisor: 2025.07.1
supported: true
supported_arch:
- amd64
- i386
timezone: America/Chicago```


I was able to determine that is seems to be caused by this `systemd-journal-gatewayd` process as shown in the Glances screenshot. 
As you can see in the VM CPU average graph, it will just stay pegged like that for days until I reboot the VM. After the first time it happened, it did not re-occur for 2 days, this time it came back within 30 minutes.
Any pointers on how I would troubleshoot this or ideas on what causes this?
Thanks!

dawn gull Jul 15, 2025, 8:41 PM

#

I'd probably try to strace the process but not sure how best to do that on HAOS.

light gust Jul 15, 2025, 9:01 PM

#

heavy trellis Hi! Recently I have been experiencing an issue with my Home Assistant install wh...

machine: qemux86-64
have you tried using host cpu type on the vm? so it can talk to cpu directly using its instruction set. this is easy to change in proxmox but may be different on other hypervisors

#

this may not help but something to try

dawn gull Jul 15, 2025, 9:01 PM

#

I suppose you can do something like this but with a different command: https://gist.github.com/Impact123/e9a4a07b184eb393d2ff762e3b1b0a05#through-advanced-ssh-addon

# Start and enter alpine container
docker run --rm --privileged --pid=host -it alpine sh

# Install strace inside container
apk add strace

# Follow
strace -ttTfp $(pgrep systemd-journal-gatewayd)

# Summarize
timeout 10 strace -cfp $(pgrep systemd-journal-gatewayd)

dawn gull Jul 15, 2025, 9:02 PM

#

light gust this may not help but something to try

They are using host, otherwise the model wouldn't be shown in glances.

light gust Jul 15, 2025, 9:03 PM

#

dawn gull They are using `host`, otherwise the model wouldn't be shown in glances.

oh yeah, right enough

heavy trellis Jul 15, 2025, 9:36 PM

#

light gust > machine: qemux86-64 have you tried using host cpu type on the vm? so it can ta...

That's interesting. It is already set to host in proxmox. If you look at the Glances screenshot, the correct processor is detected there. Not sure why the ha info output makes different claims.

dawn gull Jul 15, 2025, 9:46 PM

#

Because that isn't the CPU but the machine type: https://developers.home-assistant.io/docs/operating-system/board-metadata/#supervisor-related-variables

heavy trellis Jul 16, 2025, 12:40 AM

#

dawn gull I suppose you can do something like this but with a different command: <https://...

I'll give that a shot. I assume I would need to do that when the issue is happening?

dawn gull Jul 16, 2025, 12:41 AM

#

Indeed.

heavy trellis Jul 16, 2025, 12:45 AM

#

If I log in to the HA OS through the Proxmox console and enter login it takes me to a shell as well - is that the HAOS shell or one of the containers? I could do it there, too, right?

dawn gull Jul 16, 2025, 12:45 AM

#

Yes and yes.

heavy trellis Jul 16, 2025, 12:50 AM

#

The follow is listing a ton of stuff...
Output from the summarize is here, but I am not sure how to interpret this:

% time     seconds  usecs/call     calls    errors syscall
------ ----------- ----------- --------- --------- ----------------
 93.59   21.705481      803906        27           ppoll
  6.37    1.476753           8    177933           poll
  0.01    0.002630          48        54           fcntl
  0.01    0.002362          42        55        27 read
  0.01    0.002078          76        27           openat
  0.01    0.001535          56        27           close
  0.00    0.001070          39        27           fstat
  0.00    0.000105          35         3           lseek
  0.00    0.000041          41         1           write
  0.00    0.000028          28         1           restart_syscall
  0.00    0.000017          17         1           sendto
------ ----------- ----------- --------- --------- ----------------
100.00   23.192100         130    178156        27 total
Terminated```

dawn gull Jul 16, 2025, 1:02 AM

#

Me neither but the idea is that this gives some clues for the devs what could cause it.

heavy trellis Jul 16, 2025, 1:15 AM

#

Makes sense. Well at least now I know I can use systemctl restart systemd-journal-gatewayd to temporarily fix it rather than rebooting the VM.
Not sure what the etiquette is and visibility for devs here vs Github. Does it make sense to post this in Github in the HAOS repo as well?

dawn gull Jul 16, 2025, 1:17 AM

#

It might be a good idea to open a bug report and share that output too, yeah. The problem is that unless it's reproducible it might be hard to fix but we'll see.

heavy trellis Jul 16, 2025, 1:20 AM

#

Yeah, I've been scouring all over to see if anybody else has seen this issue. It seems like it may have started with the recent HAOS 16 update, but I am not sure on that. If that is related, we might start seeing more reports of this as adoption goes up.
It seems to be happening pretty frequently now and is having some performance impacts, so I'm thinking I may need to rebuild with a fresh image and restore from backup. Thanks for the help with this!

heavy trellis Jul 18, 2025, 1:29 PM

#

I downgraded to HAOS 15.2 3 days ago and the issue never returned. I just upgraded to 16 again to see if the issue will return or if it was maybe just a glitch.

#High CPU usage from `systemd-journal-gatewayd`