As someone who had to tweak many times LMK and Android's way to compute which process/app has which priorities, and how is a service restarted, I want to scream very loudly. But I guess that works.

Ideally you do this in a bank app, so that when you swotch over to that PSD2 forced nonsense the app that made the payment request has died and loses the session and you get to start all over.

I was going to ask if Android exposes mlock, which is a call that implements memory locking on normal UNIX and LINUX machines, but it looks like it is restricted to 64kb on Android.

So Android will kill your fake process.

Even when that is done, does it keep on killing other apps? Why would it do that? Because once your fake process is gone, the memory pressure is gone.

Oh please, say more, it seems very interesting.

I didn't really get the last two bullet points

For batch jobs I would really like a scheduler that will pause and fully swap out processes until memory is available again. For example, when compiling a C++ project, some source files or some link steps will require vast amounts of memory. In that case you would want to swap out all the other currently running compiler processes so the memory hungry one can do its job, then swap them back in. I don't want to punish the memory hungry process, actually I want exactly the opposite - I want everything else to get out of its way. The build system will eventually finish running processes that take up a lot of memory and will continue the ones that require little memory.

> I really wonder if a "simple" memory-aware scheduler that punished tasks whose memory behavior (allocation or access) slows down the system would be enough. What if such a process was punished (for slowing down the system) by being slowed down (not being scheduled or getting lower cpu times) in a way that, no matter what it did, the other tasks continued fast enough so that it could be (even manually) killed without soft-crashing the system?

This approach is hard to make work, because once the system is in memory shortage, mostly all processes will be slowing the system. There's already a penalty for accessing memory that's not currently paged in --- the process will be descheduled pending the i/o, and other processes can run during that time ... until they access memory that's not paged in. You can easily get into a situation where most of your cpu time is spend in paging and no useful work gets done. This can happen even without swap; the paging will just happen on memory mapped files, even if you're not using mmap for data files, your executables and libraries are mmaped, so the system will page those out and in in an effort to manage the memory shortage.

To make a system easier to operate, I like to run with a small swap partition and monitor swap usage both in % and by rate. You can often get a small window of a still responsive system to try to identify the rogue process and kill it without having to restart the whole thing. A small partition means a big problem will quickly hit the OOM killer without being in swap hell for ages.

There might be research or practice from commercial unix and mainframe where multi-tenancy is more common? What I've seen on the free software side is mostly avoiding the issue or trying to addressing it with policy limits on memory usage. Probably more thorough memory accounting is a necessary step to doing a better job, but adding more ram when you run into problems is effective mitigation, so....

I think the LD_PRELOAD is automatically inherited by whatever processes make executes. Otherwise you’d have to wrap each individual build step.

Ya that's what I wanted too. I'm actually flabbergasted that malloc() isn't a blocking call on most OSs, that would wait until the requested amount of memory was available before returning. That way programs would just suspend/resume as needed in low memory situations, rather that crashing. A process viewer could show which programs are blocked waiting for more memory, and users could optionally resume them manually. We could even serialize and unserialize entire programs and have them resume when enough memory is available.

This one simple thing would have freed users from running under the assumption that programs can crash at any time, and allowed them to operate at a higher level of abstraction to get more real work done.

This missed opportunity with a blocking malloc() is one of any number of obscene design decisions that I can't unsee in the tech status quo.

In my experience, basically all software bugs stem from asynchronous/nonblocking behavior. Because it's difficult to prove that async code is deterministic without coming full circle and restructuring it as sync. For example, higher-order methods like map/reduce and scatter/gather arrays can replace iterators. And this can now be done automagically by LLM code assistants, and static analyzers since the middle of last century. Once we see that this is possible, especially if we use all const variables to avoid mutation (mostly future-proofing our logic), it's hard to avoid asking ourselves why we're all doing it the hard way. We should be able to click a block of code and choose "make sync" or "make functional" and vice versa. So that beginners could write batch files and macros with familiar iteration syntax and transpile it to safe and reliable functional code. And experts could write pure functional code in shorthand and export it as imperative code for others to verify.

This was always another dream of mine, since I've been waiting for big companies to do it since the mid 1990s but they can't be bothered, yet I'll likely spend the rest of my life building CRUD apps to make rent.

One would think it would be better to do interpose on `execve()` or even `fork()` rather than to simply do the waiting in an ELF `.init` section. After all, if the parent is spawning lots of child processes then that is a problem in itself. But yeah, this approach will presumably work most of the time.

Incidentally, as pure-Go and pure-Rust programs proliferate, `LD_PRELOAD` stops being useful.

Not only that, but it’s inherently racy (TOCTOU); a process could allocate a huge block of memory between the time the determination to allow the controlled program to start and the time that program is able to finish initializing is made.

A better solution is to use a proper memory-reservation scheduler, not hacks like this. Kubernetes has such a thing.

It's pretty naive, it cat's from /proc/mem and extracts MemTotal and MemAvailable: https://github.com/surban/memstop/blob/master/memstop.c#L40-....

In other words, that's not how virtual memory works. And I actually don't know how I would go and solve this problem because of the way commit, overcommit, and paging algorithms can intervene with the goal.

> When the program reaches its memory limit it will be throttled.

I thought it will be killed?

There is already memory.high (MemoryHigh= in systemd) you can set on cgroups which does something similar at the kernel level.

But its challenging to use correctly. As its easy to end up in a live lock situation where the process never frees memory but also never gets killed.

See all the details that Kubernetes had with introducing this

https://github.com/kubernetes/enhancements/issues/2570

I'd be more worried about the possibility of deadlock.

> there are security implication when using LD_PRELOAD

What do you mean?