Very high Load / CPU Utilization

marschi

Hello everyone,
I’m running Mailcow on a VM hosted by ETH Services (Gianfar, 4 VCore, 8GB RAM, 80GB SSD) at a “family scale,” meaning 7 mail accounts.

At somewhat regular intervals, the CPU load spikes to an unhealthy level, causing mail retrieval and calendar/address book sync to fail.

Usually, at that point, an update is also available, and by running update.sh (which includes restarting the containers), the problem resolves itself.

Here is a screenshot of htop by the time the problem occurs:

Do you have any hint what could cause this and where i can start to trace it down?

thank you
Markus

esackbauer

Htop using one whole core? That seems to be a problem on provider side, like oversubscription of CPU and/or memory.
Also all of that 8 GB memory is eaten up plus all of swap space (which is rather small for a 8 GB VM).
Could be that swapping makes everything slow.

Try this;:
https://raspberrytips.com/check-linux-performance/

marschi

Thanks for the reply, i wait for the next event and will collect more data with the commands described. And with whose probably also consult the provider 👍️

ETNyx

like @esackbauer said

Although I do not think this is CPU over-subscription, if reported correctly, by your provider, it should be turquoise or yellow color not red. Red are kernel threads, while you have high load and over-pressure on your memory subsystem could be partially swapping like @esackbauer said. On 8GB and 6 user system you could disable swap to avoid this bottleneck, if it is it.

On the note of memory, what is allocating it? Usually clamd have high usage, yours shows 5%, also seems quiet fresh, but it is marked by yellow color, this should mean that its executable has been change on disk, this is quiet funky you got there,..

In the past I got similar problem of overloading system by clamd cause by some fork bomb test of specific mail attachment, it solved it self by omm killer but mail was always lost,…