Time  Nick          Message
11:42 Bombo         helo
11:43 Bombo         i found the language setting in minetest 5.4.0, set it from nothing (english) to de (german) restarted it, but it still is englsh
11:43 Bombo         are the locales not included in the android version?
11:43 Bombo         did i mention amdroid? android ;)
11:44 sfan5         android currently does not come with language support
11:44 Bombo         no 5.4.1 version either
11:44 Bombo         ok sfan5
11:45 Bombo         probably nothing changed for android in the .1 i guess
11:47 sfan5         even if you had 5.4.1 it still wouldn't work, yea
12:12 Bombo         no i mean genera<lly speaking, not just locales
18:06 hisforeverkid Hi I need t find the mod that has cherry wood. Willow wood?
18:11 Krock         willow should be in moretrees
18:11 hisforeverkid tanks Krock
18:12 Krock         !next
18:12 MinetestBot   Another satisfied customer. Next!
18:16 calcul0n_     iirc cherry tree and another kind of willow are in cool_trees
18:17 calcul0n_     ho, he's gone
20:15 whosit        Hello. I'm trying to figure out the cause for periodical lag spike (0.6+ timestep), my research comes down to a mod calling player:get_meta():set_int() each 5 seconds. That somehow triggers some disk IO, which freezes server for a moment
20:16 whosit        any idea why that may happen only in this special each-5-seconds globalstep, and not in other places? how to fix it? where to look?
20:18 whosit        https://github.com/mt-mods/xp_redo/blob/master/stats.lua#L16 it comes down to this line
20:19 whosit        and seems to happen with sqlite and postgres backends on different machines
20:19 whosit        which is kinda crazy...
20:20 entuland      according to that code, that function should get called every time a player places or digs a node as well
20:20 entuland      are you seeing that spike in those occasions as well?
20:21 whosit        it's harder to see that case :\
20:21 entuland      ah, that code does that for all the connected players at that interval
20:22 whosit        yes, it gets worse the more players connected
20:22 entuland      makes sense
20:23 entuland      can't tell - checked if there is any thread about that mod and reports about such a performance hit?
20:23 whosit        not in the issues (closed or open) and not in the mod's thread
20:24 sfan5         mod storage is saved along with the server's map every 5.3 seconds (by default)
20:24 whosit        it uses player meta which by default is it's own db (players.sqlite) same thing?
20:25 sfan5         uh right
20:25 sfan5         but yes that's the same
20:27 whosit        that means, any constant change in players meta will cause same periodical lag spike? it's pretty significant
20:28 whosit        It's visible, that's why I'm trying to fix it
20:29 sfan5         map saving happens at the same interval so it's weird that it wouldn't cause a problem but player meta would
20:30 entuland      the amount of IO should be proportional to the amount of data altered, and that looks risible compared to the map as sfan5 pointed out - what's the size of your player.sqlite file, out of curiosity?
20:30 whosit        it may be an artefact of how I'm trying to reproduce it (starting server and connecting ~6 clients), but only enabling xp_redo causes same constant spike, not mobs_redo or mobkit mobs, etc.
20:31 whosit        it's just 6-7 test players on a clean server
20:31 entuland      definitely odd
20:31 sfan5         saving player meta rewrites the entire meta while saving a block wil only touch that block so that'd be a different
20:31 whosit        76k players.sqlite file
20:32 sfan5         if you configure profiler_print_interval to a number > 0 and enable verbose output it'll print the profiler
20:32 sfan5         there should be a "Server: map saving (sum)" entry which tells you how long map saving took
20:33 whosit        I did another thing:
20:33 whosit        moved players.sqlite file to ramfs directory and symlinked it, and spike was gone
20:34 whosit        "map saving" time will show how long it took to rewrite players.sqlite?
20:34 sfan5         map saving *and* player data saving
20:35 whosit        ok... I will try to measure it that way then. Still not sure how to better get rid of it :)
20:36 entuland      this seems important if it made a difference: "moved players.sqlite file to ramfs directory and symlinked it, and spike was gone"
20:37 whosit        and I see same spike on a server that's using postgres backend and has no players.sqlite
20:40 entuland      can't tell about postgres, but if the problem disappeared by changing the file location and symlinking it, since it should be transparent to minetest, that seems to point at something else interfering with that IO
20:42 whosit        it's not just location though, I have moved file to a memory filesystem - so no slow IO when writing it
20:42 whosit        (linux ramfs)
20:44 entuland      that's the thing, even writing the entire file at once every time the amount of data is risible and doesn't justify the entire gameserver freezing
20:45 sfan5         is the disk a HDD?
20:46 whosit        on my test machine, yes, on the postgres backed one - idk
20:46 whosit        lag seems about the same on both
20:48 whosit        I notice this periodical spike only with xp_redo enabled, not other mods - that's the strangest thing. But it may be explainable by constant metadata change...
20:50 entuland      I would try and narrow down that thing by doing a small mod that only touches the meta at that same interval
20:51 entuland      it may as well be something done by hooks.lua
20:51 entuland      which is the line run right after touching the meta
20:52 entuland      that hook goes on to run all the functions hooked by register_hook
20:53 whosit        tried commenting it out - makes no difference
20:53 entuland      good
20:53 entuland      commenting the meta touch out gets rid of the issue I guess?
20:53 whosit        yes, set_int() removes the lag spike
20:54 whosit        *commenting it out
20:54 entuland      ok
20:55 entuland      maybe the profiler data will give more clues
21:06 whosit        https://pastebin.com/TZn8Q1k8 <- map saving from debug.txt
21:09 sfan5         writing some stuff to disk certainly shouldn't take 500ms
21:10 sfan5         you might also want to look at this setting https://github.com/minetest/minetest/blob/stable-5/minetest.conf.example#L1512-L1514
21:15 whosit        setting sqlite_synchronous = 0 (OFF) gets rid of lag spike (map saving is up to 8ms with 6 clients)
21:17 sfan5         hope you read the warning about corruption in the docs
21:18 whosit        yes.
21:21 whosit        setting it to 1 (NORMAL) gives same 5-second-period ~500ms spikes
21:22 whosit        default is 2 I assume... so normal option works same as full sync
21:30 entuland      crabs... to times that my desktop rebooted without any error or reason whatsoever today
21:31 entuland      s/to/two
21:33 whosit        :( never happened before?
21:36 whosit        so, what should I do? make a minimal mod to reproduce the spike?
21:50 MTDiscord     <entuland> No idea sorry, out of power to even try googling
21:51 MTDiscord     <entuland> At least seems like it was power grid hiccups here and not my PC going nuts
21:53 MTDiscord     <entuland> The fact that sqlite misbehaves in one mode on the normal disk, stops misbehaving changing that mode, and stops also going to ramfs, the issue could be more tied to the FS / sqlite manager / driver or whatever than being on minetest itself
21:55 MTDiscord     <entuland> Although the fact that the issue happens on postgres as well would not match (unless for some reason the culprit affects posgres as well somehow)
22:00 MTDiscord     <Jordach> you have a dead sector on disk
22:14 whosit        1. the spike issue happened on another machine 2. I re-created server from scratch using /mods list, and issue appeared on my local machine too
22:14 whosit        only common thing - xp_redo mod
22:16 whosit        server is using completely different config, I don't even have direct access to it to see how postgres is set up, etc.
22:17 whosit        but we did make sure that spike issue disappears from server too, when xp_redo is disabled
22:22 whosit        real server should have bigger player db, but if it's using postgres and values change only for ~10 test clients, that still does not make sense
22:29 AntumDeluge   The API shows me how to unregister items (craftitem & node), but I don't see any method for unregistering ores. Can it be done? Or, is it necessary to unregister ores if I am replacing an ore-bearing node?
22:34 calcul0n_     AntumDeluge, ores are just nodes like others, i guess you do it the same way
22:44 AntumDeluge   calcul0n_, interesting. I called "core.unregister_item(item_name)", but I find that it is still stored in "core.registered_ores".
22:52 calcul0n_     ho, i forgot about register_ore(), i don't know what happens if you unregister the corresponding node then
22:53 AntumDeluge   Maybe I just need to do "core.registered_ores.ore_hash = nil". Or maybe I just leave it alone as maybe it isn't necessary to unregister ores.
22:58 calcul0n_     i'd try the second first but that's a shot in the dark :)