!XQQVyIbcAcHFvzmcTl:nixos.org

NixOS GNOME

398 Members
A room for maintainers of GNOME & GNOME-Related desktop environments (xfce, cinnamon, pantheon...)92 Servers

Load older messages


SenderMessageTime
29 Mar 2025
@grimmauld:grimmauld.deGrimmauld (moving to @grimmauld:grapevine.grimmauld.de)oh god, libsoup_2_4 has over 3k rebuilds going into gst_plugins_bad via libnice08:33:23
@jtojnar:matrix.orgJan Tojnarhttps://github.com/NixOS/nixpkgs/issues/36089710:52:52
@emilazy:matrix.orgemily
In reply to @k900:0upti.me
So why do we have two ABI versions and how do packages decide which one to use?
I think we have three?
12:40:16
@emilazy:matrix.orgemilythe ISO has two of them12:40:23
@emilazy:matrix.orgemilyah that was mentioned12:42:00
@andrew15_5:matrix.orgAndrew

Not very soon, but that is because I couldn't get the laptop to not heat up after second lock in during the day (I think some notification wake it up every time), so it either dies or I turn it off (otherwise it switches to default settings, which is insane behavior).

But during all these days nothing has changed and the RAM usage gets bigger with time. Though the issue is that I still have around 22% RAM left and a ton of swap (and idle CPU), and now any task becomes super sluggish and the OS starts to freeze. Just a few moments ago I was just editing a file in Neovim and boom...now the GUI is frozen and only music is playing like nothing has happened. (The uptime is about 3.3 days.) So I had to hard power off again. Previously when this happened I also had about 75% RAM used, but then it all froze during loading of a web page in Firefox.

As a result, there are absolutely no signs of why the system just suffocates over time, as all regular indicators show that it should work just fine. Other than ever growing RAM usage, yet there is still plenty of it left when the GUI freezes indefinitely. Since the music is still playing, I think that it's definitely something to do with gdm/mutter.

And I'm on 6.13.7, actually.

Cc K900

18:08:28
@k900:0upti.meK900If you update to latest nixos-unstable, we should have memory allocation accounting enabled by default now18:09:09
@k900:0upti.meK900 So you can use /proc/allocinfo to find out what's using the memory in the kernel 18:09:23
@andrew15_5:matrix.orgAndrewCan you explain what it shows?18:10:08
@andrew15_5:matrix.orgAndrewa PID of each process and how much RAM it uses?18:10:38
@k900:0upti.meK900https://docs.kernel.org/mm/allocation-profiling.html18:10:45
@andrew15_5:matrix.orgAndrew* a PID of each process and how much RAM it uses (in kernel)?18:10:47
@andrew15_5:matrix.orgAndrewthanks18:18:05
@andrew15_5:matrix.orgAndrewI think I will make a logger for it, so that I can see changes over time.18:22:30
@andrew15_5:matrix.orgAndrew

This shows 15.02 GiB while only around 200 MiB is free:

sudo cat /proc/allocinfo | tail -n +3 | awk '{print $1}' | tr '\n' + | xargs cpc | numfmt --to iec-i --format '%.2f'

Does this mean that it shows cumulative sum of used+cached memory?

19:27:41
@k900:0upti.meK900It shows specifically kernel allocations 19:28:48
@k900:0upti.meK900Not userspace processes 19:28:51
@k900:0upti.meK900But memory used within the kernel 19:28:56
@andrew15_5:matrix.orgAndrewWhich means that I indeed can actually see what is using how much..19:28:58
@andrew15_5:matrix.orgAndrewBut I only have 15 GiB of RAM.19:29:21
@k900:0upti.meK900Are there any outliers there? Maybe lines that take up most of the memory? 19:29:22
@andrew15_5:matrix.orgAndrew* But I only have 15 GiB of RAM. Which means no memory left for userspace stuff19:29:46
@andrew15_5:matrix.orgAndrew

Over 10 MiB:

     10.58Mi      243 mm/util.c:174 func:kvmemdup
     11.07Mi     2840 kernel/bpf/syscall.c:396 func:__bpf_map_area_alloc
     11.08Mi   161263 security/security.c:755 func:lsm_inode_alloc
     11.13Mi     2848 mm/percpu-vm.c:95 func:pcpu_alloc_pages
     12.31Mi     2619 kernel/fork.c:180 func:alloc_task_struct_node
     13.88Mi    82681 kernel/fork.c:487 func:vm_area_dup
     16.01Mi     4097 drivers/net/wireless/intel/iwlwifi/pcie/rx.c:391 [iwlwifi] func:iwl_pcie_rx_alloc_page
     24.25Mi     6208 mm/page_ext.c:271 func:alloc_page_ext
     26.00Mi       16 fs/btrfs/zstd.c:366 [btrfs] func:zstd_alloc_workspace
     38.67Mi   211151 fs/dcache.c:1664 func:__d_alloc
     41.29Mi    13185 kernel/fork.c:314 func:alloc_thread_stack_node
     41.57Mi    11530 mm/execmem.c:44 func:execmem_vmalloc
     50.61Mi     2245 mm/slub.c:2425 func:alloc_slab_page
     69.11Mi    17690 arch/x86/mm/pgtable.c:44 func:pte_alloc_one
     69.83Mi    17874 drivers/iommu/dma-iommu.c:920 func:__iommu_dma_alloc_pages
     93.71Mi   168250 lib/xarray.c:377 func:xas_alloc
    104.27Mi    26692 mm/filemap.c:1969 func:__filemap_get_folio
    107.77Mi    27587 mm/shmem.c:1863 func:shmem_alloc_folio
    126.64Mi   130691 fs/btrfs/inode.c:7694 [btrfs] func:btrfs_alloc_inode
    362.05Mi    92683 fs/btrfs/compression.c:477 [btrfs] func:add_ra_bio_pages
    546.53Mi    43777 mm/slub.c:2423 func:alloc_slab_page
    669.41Mi   171368 fs/btrfs/extent_io.c:631 [btrfs] func:btrfs_alloc_page_array
    795.85Mi   203736 mm/memory.c:1062 func:folio_prealloc
      1.47Gi     1270 drivers/gpu/drm/ttm/ttm_pool.c:97 [ttm] func:ttm_pool_alloc_page
      4.70Gi  1230002 mm/memory.c:1060 func:folio_prealloc
      5.11Gi  1338120 mm/readahead.c:186 func:ractl_alloc_folio
19:31:06
@andrew15_5:matrix.orgAndrewWeird that now cache is very big from the start (or after a few minutes).19:32:39
@k900:0upti.meK9005GB of readahead looks a bit too much, but that should be released as necessary 19:33:47
@k900:0upti.meK900I guess you just need to wait for whatever is happening to happen again 19:34:05
@k900:0upti.meK900And look at the data then19:34:09
@andrew15_5:matrix.orgAndrewThe only problem program I know is yandex-disk, that sits now at 1.1 GiB RAM. Very annoying, but I have too much stuff there that I can't transfer. It didn't do this before.19:35:14
@andrew15_5:matrix.orgAndrewIt also uses at least 1 (maybe more) full CPU core when indexing, which can slow down the system.19:36:09
@andrew15_5:matrix.orgAndrewI set a 1 min loop for logging that whole file. I won't wait until my GUI dies again, which means I hopefully will have something to tell in 1 or 2 day.19:46:25

Show newer messages


Back to Room ListRoom Version: 6