!9IQChSjwSHXPPWTa:lix.systems

Lix

962 Members
Lix user channel. Feel free to discuss on-topic issues here and give each other help. For matrix.to links to the rest of the Lix channels, see: https://wiki.lix.systems/books/lix-organisation/page/matrix-rooms259 Servers

Load older messages


SenderMessageTime
9 Jul 2025
@grimmauld:grapevine.grimmauld.deGrimmauld (any/all)yep17:52:02
@grimmauld:grapevine.grimmauld.deGrimmauld (any/all)does help too. Time to benchmark!18:00:05
@grimmauld:grapevine.grimmauld.deGrimmauld (any/all)i do think its now the sqlite write thats blocking, because we are opening a new db connection for each line of nej18:00:53
@grimmauld:grapevine.grimmauld.deGrimmauld (any/all) that feels slow, but i'll need to do more testing 18:01:32
@k900:0upti.meK900No we're not?18:01:33
@grimmauld:grapevine.grimmauld.deGrimmauld (any/all)at least its not lix 2.92 now18:01:38
@grimmauld:grapevine.grimmauld.deGrimmauld (any/all)ah wait i am stupid, oops18:02:03
@k900:0upti.meK900 for_each_init basically creates a connection per worker thread 18:02:12
@k900:0upti.meK900Or rather per batch TECHNICALLY18:02:16
@k900:0upti.meK900I think if you want more cheap perf win you can also do batch inserts for references18:03:36
@k900:0upti.meK900There's also probably something with prepared statements18:04:25
@k900:0upti.meK900And possibly also something to be said for doing everything in a transaction?18:06:29
@k900:0upti.meK900 @Grimmauld (any/all) can you try the thing I just pushed 18:20:36
@k900:0upti.meK900OK that does not work18:41:05
@grimmauld:grapevine.grimmauld.deGrimmauld (any/all)with the old changes i got a full run in 9m48s18:41:06
@k900:0upti.meK900I forgot sqlite is single writer18:41:25
@grimmauld:grapevine.grimmauld.deGrimmauld (any/all)10min is fine, 2.91 (without those changes) was 12min18:42:02
@grimmauld:grapevine.grimmauld.deGrimmauld (any/all)* 10min on 2.93 with buffer is fine, 2.91 (without those changes) was 12min18:42:19
@k900:0upti.meK900But what if duckdb18:42:20
@grimmauld:grapevine.grimmauld.deGrimmauld (any/all)will python be able to read it as easily?18:42:37
@k900:0upti.meK900Yes18:42:41
@grimmauld:grapevine.grimmauld.deGrimmauld (any/all)then maybe18:42:47
@k900:0upti.meK900Let me cook18:42:53
@grimmauld:grapevine.grimmauld.deGrimmauld (any/all)🧑‍🍳18:43:02
@k900:0upti.meK900OK I can get 4:3019:06:12
@k900:0upti.meK900Let's see how much better that is compared to sqlite I guess19:10:39
@k900:0upti.meK900* Let's see how much better duckdb is compared to sqlite I guess19:15:54
@k900:0upti.meK900The answer, as it turns out, is it's actually just worse19:16:08
@k900:0upti.meK900lol19:16:11
@grimmauld:grapevine.grimmauld.deGrimmauld (any/all)and i am once again reminded just how slow my laptop is, despite having been quite capable at the time of buying it...19:16:50

Show newer messages


Back to Room ListRoom Version: 10