Why does mpd --systemd consume more memory than mpd?

[UtkarshVerma]

When I launch mpd normally in my shell, it consumes about 48 MiB RAM, but if launch it using systemd --user start mpd, then the memory consumption is 75 MiB¹.

Just out of curiosity, what causes this memory increase?

Footnotes

1. This is the same if I invoke mpd --systemd in the shell. ↩

[MaxKellermann]

That shouldn't really be a difference at all, but if you say 48 MB, this is so extremely high it sounds bogus. What exact value are you looking at? AS, RSS, PSS, ...?
If you're really interested in MPD memory usage, you should use valgrind --tool=massif to launch MPD and check the output file. Compare with or without systemd mode.

[UtkarshVerma]

I got this from btop and it is consistent with what I get from free. I tried running mastif as follows:

$ valgrind --tool=massif --massif-out-file=non-systemd.log mpd --no-daemon
$ valgrind --tool=massif --massif-out-file systemd.log mpd --systemd

I got this for systemd:

#-----------
snapshot=41
#-----------
time=419299156
mem_heap_B=3672191
mem_heap_extra_B=328913
mem_stacks_B=0
heap_tree=empty

and this for non-systemd:

#-----------
snapshot=41
#-----------
time=606400274
mem_heap_B=4273480
mem_heap_extra_B=345256
mem_stacks_B=0
heap_tree=empty

These are way below 48 MB as you had surmised. Regarding why we get different memory consumptions in btop, I think it is because mpd --systemd is blocking and doesn't spawn a subshell which affects how btop sees the memory consumption.

Anyhow, I still don't know why it reports 48 MB being used.

[MaxKellermann]

That's both just 4 MB, plus/minus a bit (Massif gives you many more details than just the sum, if you're interested in seeing where the (small) difference really comes from). Maybe you don't understand what the number in btop really means (me neither - I've never used or heard of btop). Memory usage can mean a lot of different things, and Massif shows you what MPD has explicitly alloted. But there's also shared memory (for code from shared libraries), file mappings, the page cache, (kernel) socket buffers etc. - which is real, but hard to attribute to MPD because these are managed mostly by the kernel and it's the kernel who decides to allocate the memory, not MPD.

The truth is that MPD is really only responsible for 4 MB of memory usage. That's still a lot - probably most of that spent for MPD's in-memory music database: MPD keeps all your music metadata in RAM. That seems like a waste of RAM, but hey, 4 MB is ridiculously small these days. Your browser wastes a thousand times more than that.

btw. there's no such thing as "spawning a subshell", however. What you maybe mean is the daemonization, an obsolete feature for old-school operating systems that don't have a service manager. That has nothing to do with a shell; it's a double-fork, with no shell involved. Daemonization makes no practical difference for memory usage.