SOLVED CPU usage breakdown?

[nebulon]

One way is to determine the busy container right now first by using systemd-cgtop. It would would display the docker containers like /docker/003d4ff9290ce4b92223b3bb57195a73fe8a4f5b8e473e74f6e54575e4bdadcb where the uuid then can be grepped for to find the container with

docker ps --no-trunc | grep 003d4ff9290ce4b92223b3bb57195a73fe8a4f5b8e473e74f6e54575e4bdadcb

[ruihildt]

@nebulon Thanks, I'll do that.

Is there a way to keep track of that in time or it only works in real time?

Would it be possible to have it in the same way as we can keep track of ram in each app and in the system page per app, somewhere in the future? I have no idea if that's possible to implement or not.

[rmdes]

What does it mean ?

[nebulon]

The output is similar to top or htop, but that tool is aware of cgroups in linux, so it is easier to determine the app based on the container id.

[rmdes]

Problem is the parts that are using the most CPU are :

/ is using 793-796 % CPU 14.3 Gb MEMORY
cron.service 300+ % CPU 13 GB MEMORY
system.slice
docker.service
box.service
user.slice
containerd.service

Another problem or discrepancies I'm seeing if I dive into containers themselves is that it seems the containers that are using more CPU, the ones that often comes on top, are not tied to any domain/app in particular.

if I docker ps a specific container i see on the systemd-cgtop no matter what I try to match them I cannot associate these containers ID to my list of ID-Domains when I type cloudron list

it's like our box is running more containers than the number of app we have ?
can a container be on the loose and still running and using resources ?

It does not make sense I know

[rmdes]

I keep having this drop_caches process eating all my CPU

I tried to kill this process but no matter the way I choose to terminate/kill it it just keeps going, something is occupying the system,it seems to me, outside of the docker environnement.

[rmdes]

@nebulon running apps are not the problem, what I see is high CPU usage at the / root level of the box

[nebulon]

Ok this is interesting, according to https://linux-mm.org/Drop_Caches this means the system tries to free up memory pages. And this essentially from time to time.
From Cloudron side we do not actively do this at all, so this is likely Ubuntu behavior. If this takes up a lot of cpu and for longer time periods, this may be related to the system I/O performance as such.

I have to read up further though, for example is this is swap related and then it will indeed be painful if the swap is on some slow disk.

[rmdes]

After last reboot,the process is now gone and CPU is more stable, quite crazy how this process completely freeze the cloudron box, Swap is being used at 338M over 4GB available and seem to increase very slowly, CPU is now around 50% more or less, still with Highs of 89%

i guess it will be better when on Ubuntu 20.x ? and moving to a new VPS or VDS soon probably

[nebulon]

Not sure if Ubuntu 20.04 will change anything here, this seems to be a pretty normal thing since a long time for linux distros.

Also this may not indicate that you need more main memory or so, it might simply mean that the process of discarding and finding those resources takes long on your system. Thus my concern about system I/O speeds.

Which VPS provider are you currently using and is this an SSD system?

[rmdes]

@nebulon This is the VPS specs I'm using : https://contabo.com/?show=configurator&vserver_id=237
8 CPU cores
30 GB RAM (guaranteed)
800 GB disk space (100% SSD)
100% SSD disk space

[nebulon]

This looks ok and good enough to me. I am always a bit unsure about disk I/O measurements and also if this is the root cause here at all, but anyways maybe you can do some sanity check with hdparm -t /dev/vda1 (replace /dev/vda1 with you main disk partition)

It should be somewhere above 500MB/s at least on provider with SSD in my experience.

[ruihildt]

FYI, CPU usage is now stable and around 13%, more or less since Contabo did, and I quote their email, a "technical adjustement".

You can clearly see on that monthly CPU chart, how it evolved (granted it's early days, but still):

Not only this, but the dashboard now loads nearly instantly whereas beforehand, it could take up to 30 seconds.

Taking into account previous random issues with cheaper Contabo VPS in the past, I tend to think the issue was from Contabo from the start.

[robi]

@ruihildt
nice to see it resolved outside of Cloudron.

If you want a similarly sized and less expensive system, I can share an invite link for ssd nodes that's been working well for us.

[ruihildt]

@robi Yes please.

I also think next time I encounter a stability issue, I'll explore the provider angle sooner, maybe by replicating it elsewhere the same stack.