mozilla :: #perf

15 Mar 2017
07:49YoricHey.
07:49YoricI have introduced a telemetry probe (keyed, count) to find out the encoding of external scripts.
07:50YoricWhat's the best way to simply visualize the proportion of results?
07:50Yorici.e. I'm not interested in all the details, only in knowing that e.g. 45% of all external scripts are in utf-8.
07:51Yoric(it's DOM_SCRIPT_SRC_ENCODING)
08:12Yoricrvitillo|workweek: ^ (if you find time during your work week)
10:41YoricDexter: ping
10:41DexterYoric, pong o/
10:42YoricHi, how are you?
10:42YoricDexter: I'm trying to get a nice visualization of DOM_SCRIPT_SRC_ENCODING (keyed, count). I'm only interested in the proportion of each encoding. Is there a good way to do that?
10:43YoricA pie chart would work nicely, or just a set of percentages.
10:43DexterI'm doing fine, thanks! And it's always a pleasure to hear from you ;)
10:43DexterLet me see
10:45DexterI guess that a pie chart would work, but the effectiveness of that pretty much depends on how many keys you have
10:46DexterYou could also consider bar charts
10:46Dexter(both of them should be available on re:dash !)
10:52YoricDexter: I expect I'll have 3 or 4 big ones and a few smaller ones.
10:53YoricI'm interested essentially in the big ones.
10:54* Yoric is reading about re:dash.
10:54DexterYoric, then you could have both a pie chart with the major one + an "other" chunk (that includes the smallest ones)
10:54Yoricpie chart or bar chart would be fine
10:54Dexteror a bar chart with all the keys
10:54Dextercool :)
10:55Dexteras for re:dash.. -> http://sql.telemetry.mozilla.org/
10:57Dexterlet me find a few helpful links
10:58YoricDexter: If I want to fetch a histogram, what source is that?
10:59DexterYoric, the most recent Telemetry onboarding is helpful to know about the various things -> https://air.mozilla.org/roberto-vitillo-telemetry-and-other-data-sources/
10:59YoricThanks.
11:00DexterYoric, https://github.com/mozilla/telemetry-batch-view/blob/master/docs/choosing_a_dataset.md
11:00DexterThat explains what is each dataset for
11:00Dexterlongitudinal vs main summary vs others
11:01Dexterif you're only interested in aggregates, https://mzl.la/2ms6wzx
11:01Dexter(only shows 4 keys unfortunately)
11:03YoricDexter: Well, I'm interested in global proportions, so this view gives me stuff that's not really readable.
11:03* Yoric is trying to find the histogram in a longitudinal.
11:08DexterYoric, depending on when the histogram is landed, it might not be there
11:08YoricDexter: How do I even find an histogram in the list?
11:09YoricThere are thousands of them and the search field doesn't seem to work.
11:09DexterLet me check
11:13DexterYoric, these datasets gets generated weekly
11:13YoricAh, so I may be a few days early.
11:13Dexteryeah
11:13YoricI'll try again in a few days.
11:13DexterCool
11:13DexterIf you absolutely need answers now, you could try Spark
11:13Dexterbut it gets.. messy
11:13YoricNah, no emergency.
11:13YoricThanks.
11:14DexterYou're welcome!
12:52igoldanjamaher|afk: hi, joel?
12:52igoldanjmaher|afk: are you online?
13:34igoldanrwood: hi, Robert
13:35igoldanrwood: I've prepared my build environment
13:35rwoodigoldan: good day
13:35igoldanrwood: and now I'm looking over improvements on perfherder, like Joel asked me
13:35igoldanrwood: currently, I'm looking over https://treeherder.mozilla.org/perf.html#/alerts?id=5325
13:36igoldanrwood: I think I can link this to bug https://bugzilla.mozilla.org/show_bug.cgi?id=1339089
13:36igoldanrwood: could you confirm if I'm right?
13:37igoldanrwood: this is a simple one, I guess
13:37igoldanrwood: but it's a start
13:37igoldanrwood: :)
13:37rwoodigoldan: I'm in the middle of filing a perf regression bug, will get back to you in a few min :)
13:37igoldanrwood: cool, thanks
13:40rwoodigoldan: i'll cc you on the bug so you can see an example
13:41igoldanrwood: ok
13:50rwoodigoldan: hey, ok cool checking out your links now
13:51rwoodfyi the bug I cc'd you on is for this alert: https://treeherder.mozilla.org/perf.html#/alerts?id=5378
13:51rwoodI did a bunch of retriggers last night that verified the regression
13:54rwoodigoldan: I see there are a bunch of retriggers for the improvement for the alert you're looking at (5325)
13:54rwooddid you do those?
13:55igoldanrwood: no
13:55igoldanrwood: I didn't yet run any retrigger
13:55rwoodigoldan: ok, j-maher must have
13:56rwoodwell no need to retrigger as they're already done, so yes lots of data points and you're correct it looks like bug 1339089, awesome
13:57rwoodigoldan: I believe jmaher|afk was looking at one yesterday but at that time we couldn't mark them as acknowledged b/c of a bug (it's fixed) so he was probably trying to acknowledge this one
13:58igoldanrwood: and then I should change from 'untriaged' to 'Link to bug', pointing to bug 1339089?
13:58igoldanrwood: or create a new bug, informing of the improvement?
13:58rwoodigoldan: yessir :)
13:59rwoodthen select the alert and chose 'acknowledge'
13:59igoldanrwood: which step? the second one? with the new bug?
13:59rwoodigoldan: first link to bug, and choose 1339089
14:00igoldanrwood: ok
14:00igoldanrwood: thanks
14:00igoldanrwood: :)
14:00rwoodigoldan: then when you're finished that, select the alert check box and set it to 'acknowledged'
14:02igoldanrwood: did it :)
14:02rwoodigoldan: ok I see that's done awesome
14:02igoldanrwood: ohoo!
14:02rwoodnext: the fun part, you get to tell the developer the good news
14:02igoldanrwood: gotta get them all :D
14:02rwoodigoldan: so you get to add a note to the bug with the good news, one sec... i'll see if I can find an example
14:03igoldanrwood: nice, Joel told me about this kind thing :)
14:03rwoodigoldan: yeah this is the fun side, delivering good news hah
14:04rwoodgive me one sec...
14:04igoldanrwood: no problem
14:04rwoodigoldan: ok here's an example for you: https://bugzilla.mozilla.org/show_bug.cgi?id=1332036#c8
14:05rwoodif you could please add a comment to bug 1339089 similar to that, but with the correct details instead for your improvement
14:05igoldanrwood: sure
14:09rwoodigoldan: looks good
14:10rwoodigoldan: congrats on your first improvement sheriffed :) normally you would have to do the retriggers but they were done for you this time haha
14:10igoldanrwood: I thanked for the improvement: https://bugzilla.mozilla.org/show_bug.cgi?id=1339089#c14
14:10rwoodigoldan: yep I had a look thanks for that
14:13igoldanrwood: what must I do to retrigger tests for a suspect revision?
14:13igoldanrwood: could you highlight the main steps?
14:13igoldanrwood: just for reference
14:13igoldanrwood: I'll look closely into Joel's videos
14:14igoldanrwood: but your answer would confirm my assumptions
14:15rwoodigoldan: so first step is select the alert as displayed on the graph in perfherder, and on the pop-up select the 'job' link
14:15rwoodThen look at the list of commits in that push (top left)...
14:16rwoodIf you're lucky there's only one commit, or a couple of commits associated with the same bug
14:16rwoodif that's the case then the job can be retriggered (will get to that...)
14:17rwoodif there's a list of multiple commits, i.e. it's an uplift/merge commit, then you need to bisect and find out which commit is the suspect.. that is a story for another time/when the time comes
14:17rwoodIf there's a single commit, then you'll see the job on treeherer is already selected after you clicked the 'job' link...
14:18rwoodso on the bottom left panel you'll see a circular arrow icon, that is a retrigger button, so you'd click that once to retrigger once... you need at least 6 points total minimum so 5 retriggers
14:18rwoodbtw you need to be logged into treeherder first in order to retrigger
14:24rwoodigoldan: for improvements we don't generally spend time bisecting, it takes alot of resources, so that's usually just for regressions, unless maybe it's a huge improvement and we really want to find out which commit
14:24igoldanrwood: it's good you told me that
14:25rwoodigoldan: yeah but for improvements like the one you just looked at, that had a couple of commits with the same but it's good to do
14:25rwood*same bug
14:28rwoodigoldan: when it comes time for bisecting it's best I describe that on vidyo/skype
14:28igoldanrwood: sure
15:05igoldanrwood: is it ok if I repeat random Talos tests? one each-and-there, to track on Perfherder, graphs
15:06igoldanrwood: to better familiarize with the treeherder interface, options
15:07rwoodigoldan: prefer you don't actually as that takes resources to run, probably best to only retrigger tests when you are investigating
15:07igoldanrwood: ok
15:08rwoodigoldan: any questions about the treeherder interface feel free to ask (or email me) anytime
15:12igoldanrwood: in what situations do we retrigger jobs using treeherder instead of Try?
15:12igoldanrwood: also, where can I see the Try results for my retriggered jobs?
15:13igoldanrwood: on treeherder, I can view mozilla-inbound, autoland and mozilla-beta
15:14rwoodigoldan: so when we need to bisect commits that's when you push to / run jobs on try
15:15rwoodigoldan: when you push to try you can see your push on treeherder too, but the try repo: https://treeherder.mozilla.org/#/jobs?repo=try
15:15rwoodyou also receive an email with the try link
15:16rwood(on treeherder you can use the Repo pull down to change)
15:47igoldanrwood: I wish you a good day
15:47igoldanjmaher: talk to you tomorrow :)
15:47igoldanrwood: thanks for your time and for the good answers
15:48jmaherigoldan, bye for now
15:48jmaherrwood|afk: thanks for helping him
16:27rwoodjmaher: np :)
16:52rwoodwlach|afk: thank you, re: Bug 1347602
16:58wlachrwood: np, I just looked at the alerts last night and figured that the two cases of a tresize regression had to be related
16:59rwoodwlach: good call
19:01rwoodjmaher: ping
19:01jmaherrwood: pong
19:01rwoodjmaher: regarding https://treeherder.mozilla.org/perf.html#/alerts?id=5340
19:02rwoodthis is an older alert (feb 24) I see someone retriggered already awhile back.. .my question is...
19:02rwoodhttps://treeherder.mozilla.org/perf.html#/graphs?timerange=2592000&series=%5Bautoland,edaec66500db21d37602c99daa61ac983f21a6ac,1,1%5D
19:02rwoodis it even valid now? there was another alert after that one, that was since fixzed
19:07jmaherrwood: that is so old we couldn't do much with it
19:08rwoodjmaher: ok, yeah that's what I was wondering, thanks I'll mark invalid
19:08jmaherrwood: often if it is >2 weeks old and the test has had other alerts, I leave it alone- in some cases I have seen improvements fix the regression so there is less need for documenting it
19:08jmaherrwood: I would just acknowledge it
19:08rwoodjmaher: ok good point, it's not really invalid
19:08rwoodthanks!
19:13jmaherrwood: just took a look at alerts, looks as if most are cleaned up, I assume the reds are downstreams or related to existing bugs (i.e. moar data == new alerts)
19:14rwoodjmaher: yep, I have retriggers going for two of them that are not downtream but I believe the others are
19:16jmaherrwood: cool
19:17jmaherthanks for doing that
19:18rwoodnp!
16 Mar 2017
No messages
   
Last message: 161 days and 12 hours ago