Conversation
Notices
-
My VPS can't pull #guix any more. It times out after 1h of no text output (but 100% guile CPU). Guile 2.2 together with an expanding package catalog have simply made the pull process too big for this poor machine.
-
Sorry, not CPU as such. 100% wait. For swap.
-
@dt I guess. Better is worse.
The VPS has 1 GiB of RAM and Guile uses something like 750 MiB of it. Near-zero buffers, near-zero cached pages.
Not sure where the mem goes, the biggest processes resident in memory are tmux, mosh-server and transmission-daemon and they take like 2M res each. Actually, basically those and systemd are the *only* processes apart from kernel processes, which in top have 0 bytes res. So I guess the kernel needs 250M to itself.
-
Oh hey, I have zram installed, so of course 25% of RAM is that. Remains to be seen if removing zram makes it better or worse.
-
With zram it reached about 75% before giving up. Now trying with zram-config.service stopped.
-
Now past 76%, so in this case, no zram was the right decision. Guile is using 900 MiB of mem, buffers and caches are empty, and swap is at 850 MiB. I think swap was at < 300 MiB before, so most of that is probably Guile as well.
-
Still, can't run i2p or freenet on this machine while pulling, so if I'm keeping guix on the machine, I can't pull in a cron job any more.
-
Actually maybe 76.3% is where it was stuck before too. Now it's at 100% wait.
-
Still 76.3%, now 1 GiB swap. There's something especially nasty happening at this particular point, apparently.
-
@dirb I think zswap probably captures the best of both worlds.