[03:08:40] Krinkle: re https://gerrit.wikimedia.org/r/c/mediawiki/core/+/745638/1..3/includes/libs/rdbms/database/DatabaseMysqlBase.php#b1041 I see QUERY_CHANGE_NONE on both sides of the diff...still confused [16:36:48] AaronSchulz: hm.. you're right. I stared at that many times, I don't know why I couldn't see it. [16:36:59] sorry about that [17:29:08] Krinkle: 2 questions for you 1) best resource to point people to what want to look at xhgui for the first time? 2) is there a way to start generating flamegraphs for a specific API call on a semi regular basis (or perhaps just one off?) [20:14:12] Heh, Amir1 just pointed me to https://phabricator.wikimedia.org/T291015 xD [20:17:10] addshore: yeah, that answers both in the future [20:17:20] :D [20:17:43] if you are at all interested, i just rambeled by way though this one https://addshore.com/2022/02/profiling-a-wikibase-item-creation-on-test-wikidata-org/ [20:19:33] addshore: for xhgui, what I typically do is sort by total wall time which brings wfMain to the top, and then the other most spent functions below that. [20:21:04] I will say though that in practice it has become gradually less useful the more optimising work you do. For anything other than a high level overview it's not so useful imho because 1) it's just one request so anything small is influenced by random variance , 2) it's on an underpowered VM which further skews things, 3) it's with tideways overhead enabled which further penalizes small/deep functions. [20:24:08] but it's useful to e.g. know that a special page takes more than half of its time in db queries [20:24:08] (which it will probsbly report as 5 of 15s comparable to maybe 0.5s of 3s in prod) [20:24:08] we have a number of breakdowns at https://performance.wikimedia.org/arclamp/svgs/daily/ in addition to the overall entry point ones, such as EditAction and Pre/PostSend [20:24:08] we could add something like that for wikibase. [20:24:08] thsi would be the same as what you get on api.php flamegraph, but with more depth since it won't have filtered out nodes that are below the 1px threshold of the zoomed out graph [20:24:08] e.g. compare https://performance.wikimedia.org/arclamp/svgs/daily/2022-02-01.excimer-wall.index.svgz?s=EditAction&x=232.9&y=1397 to https://performance.wikimedia.org/arclamp/svgs/daily/2022-02-01.excimer-wall.all.fn-EditAction.svgz?x=10.0&y=1093 [20:24:11] both show the same data