00:33:52 Unstable branch on cbro.berotato.org updated to: 0.34-a0-1989-gd38d1f0726 (34) 04:32:51 Experimental (bcrawl) branch on underhound.eu updated to: 0.23-a0-5261-gd9800d219b 05:23:28 Unstable branch on crawl.akrasiac.org updated to: 0.34-a0-1989-gd38d1f0 (34) 06:58:07 -!- Noisytoot_ is now known as Noisytoot 10:15:06 Kewl (L12 GnWn) ASSERT(wpn && (wpn->base_type == OBJ_MISSILES || is_range_weapon(*wpn))) in 'ranged-attack.cc' at line 37 failed. (Orc:2) 10:20:23 !crashlog Kewl GnWn xl=12 14:04:50 <09g​ammafunk> further investigation into the Sequell I/O issue has revealed that the fundamental problem seems volume block storage. Namely that it's only good for that advertised high throughput when reading with large block sizes or when doing large numbers of asyncronous io 14:06:07 <09g​ammafunk> I can get significantly better performance for postgresql's 8kb block size plus synchronous IO if I use the local virtual disk, so my new plan is to switch to a plan with enough local storage to accomodate Sequell's db, which it looks like we'll be able to do 14:07:25 <09g​ammafunk> postgresql 18 does support asynchronous I/O but from what I've read, this isn't used for index scans, which is what we're mostly doing. However we could tune postgres to not use indexes for larger row counts, potentially 14:07:48 <09g​ammafunk> Either way it seems like we're going to need a combination of approaches to get performance where it needs to be within a reasonable price range 14:53:57 Lusmu (L16 ReEE) ASSERT(_valid()) in 'ray.cc' at line 229 failed. (Swamp:4) 16:14:07 <09g​ammafunk> !tstats 16 t0.33-games 16:14:49 <04C​erebot> Stats after 16 days (t0.33-games): 2870 players, 890 runers, 421 winners, 1428 wins, 89582 games, winrate 1.59%, total player time 3y+258d+12:23:44. 16:15:16 <09g​ammafunk> this query works and in a decent timeframe, which is one of the most important type of tournament queries 16:16:00 <09g​ammafunk> I've moved postgresql's data dir over to a resized instance with about 25% of space to spare, which should be enough to cover sequell data for a long time (it grows slowly) 16:19:41 <09g​ammafunk> moving data off volume block storage to local instance storage got us about a 2x performance improvement. There are plenty of longer milestone queries that still won't be fast enough, but hopefully the very important ones can work 16:19:53 <09g​ammafunk> if any devs have queries they feel they need that still aren't working, let me know 16:21:01 <09g​ammafunk> sequell tuning will be an ongoing thing, and there are things we can do like trying postgresql 18 with its asyncio i/o (somewhat dubious in that we're doing index scans, which can't be async, but we could try tuning), we could try sharding the data based on cv, and we can try other index strategies to some degree 16:22:07 <09g​ammafunk> for now though I'm mostly focused on having it useable for the most important queries (including tournament) 16:41:59 Unstable branch on underhound.eu updated to: 0.34-a0-1989-gd38d1f0726 (34) 19:21:54 Turquoise (L24 KoBr) ASSERT(can_pass_through_feat(env.grid(newpos))) in 'player.cc' at line 653 failed. (Slime:2) 19:26:54 !crashlog Turquoise KoBr xl=24 place=slime:2 19:26:56 1. Turquoise, XL24 KoBr, T:51870 (milestone): https://underhound.eu/crawl/morgue/Turquoise/crash-Turquoise-20260111-022153.txt 19:31:39 <08o​____0> That was marrionette on a kobold fleshcrafter btw 19:32:21 <08o​____0> Oh and grand finale into the wall creature again heh 19:35:17 <08o​____0> oh I replied to the wrong one, that Lusmu's was the kobold fleshcrafter marionette one 19:36:16 I didn't even look at that one, there's so many ways to trip that ASSERT… 19:37:20 <08o​____0> I was setting it up with a debugger before the power went out lol 23:25:33 03DracoOmega02 07* 0.34-a0-1990-g17a993e2a6: Update the changelog with recent changes 10(71 seconds ago, 1 file, 36+ 5-) 13https://github.com/crawl/crawl/commit/17a993e2a6a5