General programming discussion • Re: 'Best' language for user Services

You have a paging problem, right from the start it would seem. For contrast here's my PC (not paged yet)..

Code:

foo@sdu:~$ top -b -n 1 | headtop - 20:32:12 up 18 min,  2 users,  load average: 0.17, 0.73, 0.66Tasks: 469 total,   1 running, 468 sleeping,   0 stopped,   0 zombie%Cpu(s):  0.3 us,  1.0 sy,  0.0 ni, 98.6 id,  0.0 wa,  0.0 hi,  0.0 si,  0.0 stMiB Mem :  64203.5 total,  47152.6 free,   6752.9 used,  10298.0 buff/cacheMiB Swap:   1024.0 total,   1024.0 free,      0.0 used.  56598.0 avail Mem     PID USER      PR  NI    VIRT    RES    SHR S  %CPU  %MEM     TIME+ COMMAND   4815 foo       20   0   13388   4284   3408 R  11.8   0.0   0:00.03 top    539 root      20   0       0      0      0 S   5.9   0.0   0:01.49 nvidia-+   3474 foo       20   0 1788404 250628 184864 S   5.9   0.4   0:23.66 steamwe+

..ie: MiB Swap: 0.0 used.

Here's an rpi5 I use to do a lot of compiles. It has paged..

Code:

foo@pi23:~ $ top -b -n 1 | headtop - 20:33:35 up 66 days,  4:20,  1 user,  load average: 0.13, 0.13, 0.19Tasks: 186 total,   1 running, 185 sleeping,   0 stopped,   0 zombie%Cpu(s):  2.6 us,  2.6 sy,  0.0 ni, 94.9 id,  0.0 wa,  0.0 hi,  0.0 si,  0.0 stMiB Mem :   7811.3 total,    110.0 free,    328.7 used,   7372.7 buff/cacheMiB Swap:  16384.0 total,  13925.6 free,   2458.4 used.   7351.9 avail Mem     PID USER      PR  NI    VIRT    RES    SHR S  %CPU  %MEM     TIME+ COMMAND2927075 foo       20   0    9968   3324   2712 R   5.6   0.0   0:00.06 top      1 root      20   0  165364   8704   6264 S   0.0   0.1   1:34.14 systemd      2 root      20   0       0      0      0 S   0.0   0.0   0:10.60 kthreadd

..ie: MiB Swap: 2458.4 used.

Consider your lines..

Code:

MiB Swap:    200.0 total,      0.2 free,    199.8 used.    238.3 avail MemMiB Swap:    200.0 total,      0.1 free,    199.9 used.    184.9 avail MemMiB Swap:    200.0 total,      0.0 free,    200.0 used.     98.4 avail MemMiB Swap:    200.0 total,      0.1 free,    199.9 used.     76.8 avail Mem

Something has immediately gobbled up all your pagefile and over time there is less and less of it (avail). This means the system will be "thrashing" as apps fight each other for swap space thus the kernel will be spending a disproportionate amount of time paging things in & out. In the worse scenario parts of the kernel itself may be subject to paging.

Unless you're prepared to change the rpi you're stuck with 1Gb ram so your only option is to increase the pagefile. I'd start with making it 1Gb and see how that goes.

I started by increasing the swap to 500 Mb, which initially left almost 50% of the increased swap space shown as 'free'. After about 12 hours, however, the free swap had dropped to 90 Mb. The rest of the numbers still look much 'better'. I will continue to monitor.

I'm guessing that 200Mb swap is the stock dphys one? I'm not too familiar with "/etc/dphys-swapfile" in that if I need to mess with paging, I turn dphys off and use a raw old fashioned "mkswap" type partition. I *think* you'd change "CONF_SWAPSIZE=1024" then reboot but verify that.

Yes, the 200 Mb was the default setting for dphys, and this is changed as you indicate.

What you're after is a situation where about half the swap gets used then usage doesn't increase. If "avail" stills decreases over time all you've done is push the problem into the future. I envisage two causes: (a) an app is specifically requesting virtual memory and not handing it back. (b) file system operations.

I'd guess at (b) because sqlite3 is going to be constantly accessed so the kernel is going to favour keeping that in ram. Ordinarily the kernel allocates all unused ram to "buff/cache" and shrinks it as apps demand more memory. I'm only guessing at sqlite3 accesses because I'd think "buff/cache" would be smaller if something wasn't constantly accessing the filesystem. Something is causing paging rather than shrinking "buff/cache" is my reasoning. One way to figure that out would be to use (say) mariadb over a remote connection to a mysql server: way overkill unless you're into full on nerd mode!

Your guess is a good one! Almost all my services access one or both of two local sqlite databases. Only one service writes to each database, but most of the rest execute one or two SELECTs every 5 minutes. I do have a constantly running NAS on the network to which I could move the databases, but I would really rather not!

Reality has a more sensible answer. If you can keep the rpi running for a week without slowing down then reboot it in the early hours sun/mon. You're gonna have to do this anyway in order to back everything up because if it's paging constantly, it'll be killing your sdcard: even ssd's don't fare well.

This (weekly reboots) is essentially what I have been doing manually, and I already back everything important up to the NAS on a nightly basis. The problem with automating the reboots is that my services interact in fairly complex ways and don't always start up cleanly - I really need to be around to clean things up if necessary after a reboot.

Statistics: Posted by pfletch101 — Mon Jun 09, 2025 2:35 pm

---