[08:01:36] dpifke: Hi! How it's going with the WPT server? Let me know if there's anything I can do to help! I remember last time when I setup the agent talking to the server, the docs aren't good at that part, and I spent a lot of time trial and error, so I can do that again so you don't need to + update the documentation so its easy the next time (or if we wanna run multiple agents). [13:05:09] http://dl.ifip.org/db/conf/cnsm/cnsm2020/1570662329.pdf [13:12:13] FYI, just wanted to share in the context of T253160 that we enabled this yesterday - there were a few minor swings, but overall nothing crazy [13:12:14] T253160: Wall-clock Excimer profiling in production - https://phabricator.wikimedia.org/T253160 [15:45:12] Krinkle: putting the blog post together today I realised that what you asked about had been answered by the original study already: the distribution of PLT is the same for people who answered the survey and those who didn't. that's the part that matters, the survey corpus is representative of the wider corpus. this suggests that people who didn't answer because they didn't see it don't influence the overall distribution [15:45:57] so there's no skew introduced in the survey response corpus introduced by people with high PLT who've already scrolled past the survey by the time it appears [15:48:13] while that phenomenon is real, it's not big enough to change the PLT distribution [17:11:07] mszabo: cool! Thanks for that. Any particular motivation for that? (I still like it as well, just curious) [17:12:28] we've noticed that our slowest requests were typically slow due to I/O latency and cpu time was much lower than the actual WT of the request, so we wanted to check if those function calls are missing from the graphs because those slow reqs are actually too few to make a difference, or if they're missing because sampling based on cpu time overlooks them [17:12:45] but so far it seems like it's the former, so they were just outliers [17:13:12] Okay [17:13:26] Are you doing both or switched? [17:13:36] We switched so now it's pure WT [17:14:12] we could probably bring back cpu time once we have the time to clean up the UI and stop using that ghastly GCS browser for accessing the graphs [18:03:13] is there any sort of tracking anywhere for JS enabled vs no JS enabled page views? [18:39:49] addshore: in general no, but it's easy to measure ad-hoc as needed. We've done it a couple times. [18:41:21] It's important to balance though that there's a lot of nuance here. Eg on slow connections one can get perfectly fine page views in a standard browser and never have Js arrive in time before navigating away. Or any number of other failures, background tabs, browse plugin breaking stuff, etc [18:42:07] So you'll want a more tuned question to answer