KDE Will Nicely Notify You When Apps Are Being Killed Due To Out-Of-Memory

[WileEPyote]

This might be self-inflicted pain.

Without swap, the kernel can't reclaim anonymous data pages, so they all use up system RAM, even effectively unused ones. The only way the kernel can reclaim memory is by evicting file-backed pages, a lot of which are from executable binaries. As memory pressure increases, the kernel spends increasingly more time faulting back in pages to execute.

Orrrrrrr, hear me out. Install an ungodly amount of ram and never worry about OOM or swap again.

[Rovano]

A few months ago, I was at a machine about 13 years old with AMD 4cores. The machine has 2GB DDR2 RAM. I ran KDE6 there for a test. It ran beautifully smoothly.

Avis could still rely on the GPU memory consumed.

[mrg666]

I just did another check on Plasma RAM usage on my laptop. It is Latitude 7400 with 16 GB RAM and Intel integrated GPU. The first is from text console (CTRL-ALT-F2) without Plasma session. Second is after logging into Plasma session using Konsole. Command is "free -h". Plasma version 6.2.2, QT 6.8 running in FC41. Plasma is using just 300MB.

No Plasma

               total        used        free      shared  buff/cache   available
Mem:            15Gi       1.3Gi        10Gi       223Mi       4.2Gi        14Gi
Swap:          8.0Gi          0B       8.0Gi

Plasma running

               total        used        free      shared  buff/cache   available
Mem:            15Gi       1.6Gi        10Gi       199Mi       4.2Gi        13Gi
Swap:          8.0Gi          0B       8.0Gi

[moonwalker]

And brace for a bad time. Linux user space code generally either assumes that malloc never fails, or if there are error handling paths, they tend to be not well-tested. Long story short, malloc returning NULL is likely to result in fireworks.

I believe I did imply as much.

[moonwalker]

Windows doesn't do overcommit, so an application gets an out of memory error and deals with it hopefully, probably often by terminating.
Linux does overcommit which means the kernel must deal with the problem, and can therefore make different choices. The kernel takes a holistic approach and doesn't necessarily kill the process making the demand which caused out of memory. I don;t know if Windows server is different, because the Linux approach seems a lot better for a server.

That heavily depends on what you're serving. E.g., on my home server I keep memory overcommitment at its default value, enabled, because if some process crashes - I just have service units configured to automatically restart it and keep on keeping on, and the home users can just retry whatever it is they were doing - reload the page, restart file transfer over local network. But for mission-critical workloads you may actually want memory overcommitment completely disabled.

[moonwalker]

Note: You can build Krusader, it will use Qt6.

I know, I'm just too lazy to do it.

[openminded]

avis: trolls
everyone: well aware of it
also everyone: triggered anyway

[Nth_man]

Currently the only Qt5 application I still have installed is Krusader [...]

Note: You can build Krusader, it will use Qt6.

[MrCooper]

Very nice!

I don't have a swap file or a swap partition, my experience when my system runs out of RAM is that everything freezes, then if I wait a minute or two then some process gets killed and things work again.

This might be self-inflicted pain.

Without swap, the kernel can't reclaim anonymous data pages, so they all use up system RAM, even effectively unused ones. The only way the kernel can reclaim memory is by evicting file-backed pages, a lot of which are from executable binaries. As memory pressure increases, the kernel spends increasingly more time faulting back in pages to execute.

Again, if you want Linux to behave more like Windows in that regard - set `vm.overcommit_memory` to 2 in your /etc/sysctl.conf.

And brace for a bad time. Linux user space code generally either assumes that malloc never fails, or if there are error handling paths, they tend to be not well-tested. Long story short, malloc returning NULL is likely to result in fireworks.

[andreduartesp]

swapspace, an ubuntu/debian package, does dynamic swap, I use it on my installs, as well as well configured zswap.

On Ubuntu if you don't create a swap partition it installs swapspaces and configure it