I tried the usual commands, like top, ps, looked at the process tree in the taskbar, but I couldn’t find a way to understand where usage accounted as “System resources & caches” came from. Even after terminating all the apps and processes I had 6GB of memory accounted under such entry.
Given that the name suggests it’s some form of cache, how can I clear it?
The amount has been built up through usage of Falkon, I think watching YT Videos might have had an impact on that, but I have no way to prove as I didn’t check how the usage was changing. The system became unresponsive, I had to kill everything but the memory never freed up.
Why do you want to clear it? Haiku (like most modern OS) keeps a disk cache in RAM so file accesses are faster. This is better than letting that memory sit completely unused.
The cache is automatically shrunk when the memory is needed for something else. Clearing the cache when the memory is not needed would just make the system slower for no benefit.
The cleaning seems not to be happening correctly for me, I had to reboot because everything I tried to start failed with “Out of Memory” errors and there was no way to shrink that cache. There was nothing else running apart terminal (which I was using to try to find what was causing the memory usage) and all memory was reported as used under such entry in the sidebar
Hello. I have exactly the same issue that @javanx mention. When browse some pages with Falkon, the System resources and caches start to grow up, eating all the memory. The memory is not released even after closing Falkon browser. I am not sure, but looks like some webpages are more prone to causes this issue than others.
I will submit a ticket with more details (not sure where it belongs: Haiku trac or Haikuports bugtracker).
I mean as actually I’m on 64bit Haiku.
I have 32 GB RAM and since I installed this version some month ago … first time I went up more than 25 GB used memory. Nowas I closed Falkon
I ganed only about
max. 1.1 - 1.3 GB memory
Then I closed the Web Epiphany that is available on 32bit surely – which uses different engine and when I stopped that it also freed up some GB (max 3-4 GB - with 4 tabs with one heavy : Youtube, others were almost static by their content compared to YT), so more than Falkon stop freed up in memory.
Now I have Web+ running and used memory moves between 18092.0 MB … 18092.2 MB up and down- increasing stopped
Me also stopped earlier
2 Qmplay2 sessions
(one with a paused playing and
one with ready to play random youtube links to play)
1 waiting VLC as well for any downloaded videos.
They had not freed up much memory , but seems not occupied more as well.
I created a screenshot about desktop – opened the Memory usage in ProcessController of Deskbar’s “systray”,
and you can see the pinned
Memory usage in ActivityMonitor replicant as well.
As you can see the Haiku system resources occupy the most memory (on ritght - in the ProcessController details), but of course these can be allocated for an application as well …
as in htop Web+ (6300 MB) team and app_server (515 MB) team are the biggest ones.- by VIRT aka Virtual Memory usage.
From this - as I have not enough knowledge to investigate deeper - seems named " System resources and caches " the latter can be behind holded memory.
This way some caches should be the cause of increasing memory usage.
What the proper answer I let to answer else.
I’m curious if it happens on 32bit Haiku too ?
Noone above mentioned the actually used Haiku version.
EDIT :
fixed typos. I wrote cashes instead of caches - and had not recognized until now.
It’s very easy to reproduce, just run haiku in a vm and give it 1 or 2 GB of ram and use any browser, the memory is exhausted in less than 5 minutes of browsing and is not released when you close the browser, only a reboot will, sometimes loading a single website like reddit is enough to trigger it, because we have 2 problems here, a memory leak possibly somewehe in the networking and the fact that the kernel does no release the memory after closing the browser, FWIW i remember this networking memory leak started right after the gsoc when a student worked on the tcp protocol, i hope this helps narrow down the area of investigations.
The problem talked about in this thread seems to be about RAMFS memory usage, though. The ticket authors mention that when using BFS instead of RAMFS for the temporary files, BFS disk usage goes up and does not decrease until after checkfs. So something else is going on there.
What @javanx described above about using falkon is definitely a behavior that existed way before the RAMFS thing was implemented, and it’s not specific to falkon, and the RAMFS memory usage might just be yet another side effect of what ever causes the memory leak and what ever causes the kernel not to release the used memory after the apps are closed, the latter is not specific to browsers either it is just more obvious with the browsers.
