mozilla :: #testpilot

20 Apr 2017
15:13Caspy7Alright, I've gotten the Pulse bar notification for reddit *multiple times* now. Perhaps 6 or more. And it's officially annoying and irritating
15:13Caspy7how can we stop this?
15:15Caspy7I've filled it out multiple times now and no matter if I fill it out or X it out it scrolls the content to the top of the page
15:15Caspy7this is also very annoying and pretty anti user
15:17chuckCaspy7: it prompts you every 30 hours; likely just a coincidence that it was on Reddit multiple times.
15:18chuckEach separate report is helpful on its own; it includes information about the browser's performance at the time.
15:18Caspy7chuck: perhaps, but do you see that this is an issue? Beyond the annoyance to me, this information is not very valuable
15:18Caspy7the same site repeatedly?
15:18chuckNot just the same site repeatedly, but how the same site is doing at that specific time.
15:19chuckWe gather metrics about, for example, how many HTTP requests that page made, and how long it took to render the page.
15:19chuckWith enough reports, we'll be able to notice patterns that can inform future performance work.
15:19Caspy7all I feel is badgered. Making it pop up for different sites seems more understanding
15:19chuckSo it's not that we care so much about how Reddit is doing; it's how Reddit is doing for you, at that moment.
15:20chuckI'm sorry you feel badgered.
15:20Caspy7I'm also expressing this so you know that there are probably others who are feeling similarly for the same reasons
15:21Caspy7I'm very sleepy atm and will ponder a UX suggestion
15:21Caspy7that is, I'm not in a place to give one, but maybe later
15:21chuckI hear you. But that's the core functionality of the experiment, and it's important for the integrity of the collected data that the sample we collect is representative of our users. That is, if you spend 40% of your time on Reddit, that 40% of the time you're prompted, it is for Reddit.
15:21Caspy7to get the same info while having the user to be understanding
15:22Caspy7part of that UX would likely acknowledge that it has been rated before
15:23Caspy7otherwise I start to feel that it's broke, because I don't think it should do the same site 6x (whether I'm right or not - it seems broke)
15:23chuckHappy to take any suggestions you have here:
16:45clouserwremove-react-router is running at
17:01milesfyi: I am relud for the next ten days
17:39chuckNot sure if others' experiences match my own, but it feels a lot faster to me with than without.
18:37fzzzychuck: did you test with router on
18:38chuckI used prod because I had testpilot.env on dev, but they all use Cloudfront so it shouldn't be very different.
18:39chuckTesting dev now, feels more or less the same.
18:39chuckThe random pieces of UI rendering on different timelines without router is a total nonstarter to me. It feels terrible.
18:39chuckEspecially when it happens on every page load.
18:41fzzzyyeah that's really unfortunate, because we don't know if the user will have the add-on or not when rendering at build time
18:44fzzzyalso i notice that the build time rendered components don't match what the client renders and react complains, so that will slow things down a lot
18:46fzzzythat could be fixed
18:48fzzzythat actually doesn't happen to me when i run it locally so maybe something went wrong in npm run static
18:49fzzzyI'll spend a bit of time seeing how much work upgrading to the latest react-router is
19:00_6a68I think it would be good to generate HAR files rather than talk about gut feel with respect to perceived performance
19:01_6a68that would get us on a path to avoiding future performance regressions through automated testing, regardless of whether we remove react-router or not
19:01fzzzyand we definitely need to make sure react isn't complaining about not being able to reuse the build time generated nodes
19:01fzzzythe ux of loading is definitely clunky right now but could be polished
19:04fzzzythe layout is just too different between "we don't know if you have the addon" and "you don't have the addon"
19:16lorchardYeah, I think our initial little loading spinner kind of papers over a bunch of things that happen in the first frame or two after page load
19:18fzzzyi actually had a version where the loading spinner was there until we knew if the browser had the addon, i could just put it back
19:22fzzzythe cloudfront-hosted site has http links to all the subresources, prod has https, something isn't quite right
19:23lorchardWell, that cloudfront distribution isn't behind https
19:24fzzzyah, right
19:24fzzzyi don't like that browsers removed the protocol part from the url bar
19:24lorchardThat'll break a few things, but probably not the basic rendering stuff
19:25fzzzyno. but I'm wondering why react can't reuse the build-time generated nodes when it can when i run it locally
19:25lorchardTempted to gin up an Elastic Beanstalk deployer for dev stacks, but man I don't want to poke at that stuff again
19:26fzzzyI poked at beanstalk a lot for pageshot, i might be able to tolerate poking it again
19:26lorchardI had Test Pilot all set up on it back when we were on Docker, not too hard to spin up new instances under different host names. Got totally lazy for the static file switchover
19:26lorchardWell, Docker + Django that is
21:50fzzzylorchard: nice
21:52fzzzylorchard or anybody: is there a way i can test the prod build locally? i think building as dev is massively inflating the assets
21:53fzzzyI guess just npm run static and then point an http server at some directory?
21:53lorchardI'm out the door, but check what happens in bin/circleci/
21:54lorchardthere's also the gulp dist task under frontend/tasks/dist
21:55lorchardthe `npm run static` command is I think what we run to build for prod and gulp dist runs a filewatcher & server along with dist-build
21:58lorchardoh yeah, and this might be reasonably current
21 Apr 2017
No messages
Last message: 8 days and 4 hours ago