Time  Nick        Message
00:01 tumeninodes rubenwardy: Not sure if I should try those updates again? They're still just sitting there with a status of "importing" 0_o
00:05 rubenwardy  they would have failed, you need to do again
00:05 rubenwardy  the job would have been lost
00:05 tumeninodes ok, just checking. Thank you
11:01 CWz         what does SpatialIndex do?
11:09 sfan5       make AreaStore() faster
12:29 jas_        !mod
12:29 MinetestBot jas_: alphamg by Joz - https://forum.minetest.net/viewtopic.php?t=13074
12:29 jas_        what the fuck is that?
12:30 jas_        oh sorry, read it as alp hamg.  pretty cool lookin'
12:34 jas_        huh, the readme.md has lots of info and there's a gallery with a bunch of screenshots.  too bad that wasn't added to the thread.  neat
15:11 rubenwardy  Does anyone know how to work out the bottlenecks when running Minetest servers?
15:14 rubenwardy  Doesn't look like CPU is a problem   https://rwdy.uk/qJHNi.png
15:14 rubenwardy  I expect it's HDD or network, most likely the former
15:15 sfan5       generally: if the process is often stuck in disk sleep then it's the storage; if it's using 100% CPU then it could go faster with a better CPU
15:16 sfan5       if you see your network interface maxed out it's (obviously) the network
15:16 rubenwardy  ok
15:16 rubenwardy  is there a way to see why a thread is sleeping over time? Like: % disk sleep
15:17 rubenwardy  300K/s is used, not that much
15:17 rubenwardy  +B
15:18 VanessaE    don't forget luajit, and minimize disk access with things like leveldb for the map, or sqlite for the player files, or memory-caching anything outside MT that writes to disk a lot but doesn't need to (like rrdtool's databases)
15:18 rubenwardy  I use LuaJIT
15:18 rubenwardy  maybe leveldb would help
15:18 sfan5       watch -n1 'grep ^State /proc/$(pgrep minetestserver)/task/*/status'
15:18 sfan5       quick solution for what you asked for
15:20 sfan5       or system-wide: run `vmstat 1` and watch the "wa" column
15:21 rubenwardy  the output of task status shows one thread alternating between sleeping and running, and the rest sleeping
15:21 rubenwardy  I guess the rest will be emerge threads
15:21 rubenwardy  CTF doesn't have much mapgen
15:23 rubenwardy  I need to put some effort into learning this stuff
15:23 sfan5       the most likely bottleneck I can think of for your ctf server is lua exection (globalstep/abm mainly)
15:24 sfan5       unless your disk is so slow that it can't handle writing a few blocks every 10(?) seconds
15:24 rubenwardy  is Lua execution limited other than by CPU?
15:24 ANAND       CTF doesn't have any ABMs
15:24 rubenwardy  would a process that is constantly cache missing get 100% cpu?
15:25 ANAND       At least not intensive ones
15:25 rubenwardy  I've noticed there's a freeze when someone captures a flag
15:25 rubenwardy  also, my survival server has worse problems
15:25 rubenwardy  massive freezes
15:25 rubenwardy  I'll run the Lua profiler on it, see if it's pipeworks or mesecons
15:26 sfan5       1) No 2) I wouldn't worry about cache misses
15:26 rubenwardy  https://rwdy.uk/tc0OH.png
15:26 rubenwardy  it's gonna be HDD
15:26 rubenwardy  that's a lot
15:26 ANAND       99.23%
15:26 ANAND       Is that the problem?
15:26 rubenwardy  3MB per second with a write size of 16 bytes
15:26 rubenwardy  it's probably the logs, maybe
15:27 sfan5       linux does write caching afaik and any HDD should be able to handle 3 MB/s
15:27 rubenwardy  also, I noticed the other day that systemd duplicates my MT server logs
15:27 ANAND       For your survival server, there's also map saving
15:28 ANAND       Re: freezes upon flag capture, the ctf state is reset at the end of every match
15:28 rubenwardy  my friend suggests that logs should be buffered before flushing them
15:28 ANAND       Could that be an issue?
15:28 rubenwardy  on flag pick up
15:28 rubenwardy  not just capture
15:28 ANAND       Hmm
15:29 rubenwardy  please may someone join my survival server and idle?
15:29 rubenwardy  I'm using mobile data currently
15:30 sfan5       look at your htop screenshot I'd do one thing first: disable swap
15:30 sfan5       or if you want to keep it enabled set swappiness so low that it doesn't swap anything during normal operation
15:30 sfan5       and the other thing is press F2, go to columns and pull in IO_READ_RATE and IO_WRITE_RATE
15:30 rubenwardy  done
15:31 sfan5       to see which processes do the most I/O
15:31 rubenwardy  ahhh
15:31 rubenwardy  I was trying to figure out how to iotop
15:31 rubenwardy  htop is really nice
15:31 rubenwardy  swapoff is taking ages
15:32 sfan5       it shouldn't take ages to read 190MB from disk back into RAM
15:32 sfan5       though swapoff can sometimes act stupidly
15:32 rubenwardy  it's been a minute so far
15:33 rubenwardy  and done
15:34 rubenwardy  argh, why does the terminal emulator capture key presses
15:35 rubenwardy  journald is writing almost a mb per second
15:35 rubenwardy  I should disable systemd from capturing logs from minetest - as I run them as services
15:36 rubenwardy  also, how do I disable sendmail?
15:36 sofar       why do you log that much? lol
15:36 sfan5       systemctl disable --now sendmail
15:36 sofar       I have mine log to journal and it never caused an issue
15:36 rubenwardy  ah, it's a server
15:37 rubenwardy  I should probably follow the journel to see why it's logging so much
15:37 rubenwardy  and maybe consider only logging to journal
15:37 rubenwardy  because I also use MT's logging
15:38 rubenwardy  huh, it's not writing so much now
15:38 sofar       I generally just prefer it in the journal, easy to see system coherence for one
15:38 rubenwardy  I find reading the MT logs a pain in the ass
15:38 rubenwardy  with my current system
15:41 sfan5       how much is "not so much"
15:41 sfan5       check iostat
15:41 rubenwardy  journeld is now writing 80KB
15:42 rubenwardy  https://rwdy.uk/ZP7qk.png
15:42 rubenwardy  maybe sendmail was spamming it?
15:42 VanessaE    that reminds me, rubenwardy: mount with noatime
15:42 rubenwardy  basically, I had scripts which were attempting to use sendmail
15:42 VanessaE    that'll cut down a tad on disk activity alos.
15:42 VanessaE    also*
15:42 rubenwardy  but long ago I removed the email server from my server
15:42 rubenwardy  so there were loads of sendmail processes
15:45 rubenwardy  the first entry is the same, I guess loading iostat messes it up
15:47 rubenwardy  I think it's worth disabling MT's logging, and instead using journalctl
15:48 rubenwardy  It could also be good to have tiered logging, I guess
15:49 rubenwardy  make it so it only logs moderator actions and chat
15:49 rubenwardy  permanently, that is
15:49 sfan5       you dont need logs of people punching eachother
15:49 sfan5       you dont need logs of dug, placed or clicked nodes
15:49 rubenwardy  exactly
15:50 rubenwardy  having the last hour's logs is good for crashes
15:50 rubenwardy  but beyond that you only need chat commands, chat, and PMs
15:57 rubenwardy  MT's disk usage is much lower without logging
15:57 rubenwardy  is MT logging async
15:58 rubenwardy  it must be async because it's multithreaded, but is it async from the main thread?
15:58 rubenwardy  *thread-safe
15:58 rubenwardy  argh, terminology
16:00 sfan5       https://github.com/minetest/minetest/blob/master/src/log.h#L172
16:00 sfan5       data it put into the stream, the stdlib decides when to flush stuff
16:01 sfan5       though chances are it will flush on every line written
16:03 rubenwardy  argh, CLion sucks
16:20 sofar       rubenwardy: so you're saying it's more efficient to log to the journal? heh
16:20 rubenwardy  it's more efficient to only log once
16:20 sofar       sure
16:20 rubenwardy  and I expect that MT logging to the std::cerr is faster than MT logging to a file itself
16:20 sofar       but if debug.log is fflush()ed every line
17:00 rubenwardy  Thanks for the assistence, my server is running much better
17:00 rubenwardy  cross fingers
17:00 rubenwardy  I was too quick to blame kimsufi
17:13 VanessaE    not really
17:13 VanessaE    KS's servers do kinda suck, like I said.
17:13 sfan5       don't they have SSD servers?
17:13 VanessaE    I don't recall.
17:14 VanessaE    but f.ex:  my mapper script took something like 5 hours to run on the server I had there.  on the one I have now, 40ish mins.
17:15 VanessaE    I can only blame it not on the fact that the box I had then was spinning rust, but that they were especially slow disks for that tech
17:17 VanessaE    actually more like 20 mins, not 40.
18:06 sfan5       today in minetest_game/farming: "Wet Dry Soil"
18:06 sfan5       Schrödinger's soil
18:07 rubenwardy  lol what
18:13 Krock       LGTM