[00:11:57] https://www.catchpoint.com/webpagetest-joins-catchpoint [00:13:01] (Patrick Meenan, Engineering Fellow at Catchpoint) [00:57:18] AaronSchulz: q at https://gerrit.wikimedia.org/r/c/mediawiki/core/+/627387 [07:41:45] I tried the new Safari on iOS 14, it has Paint Timing behind a flag and you can turn it on (I din't remember if it had it older version on iOS, I never tried). However it seems running current OS X doesn't work with webdriver to the new iOS, so I cannot test it. [15:42:24] Krinkle, phedenskog , gilles : is there an easy way for me to see what user agents we might be oversampling for navigation timing? [15:44:45] nuria: that would be in production Mediawiki config, but as far as I remember we’ve only over sampled countries or wikis, no UAs [15:45:20] gilles: taht could also skew UAS though [15:45:35] Yes it can [15:45:55] Over sampled records are marked as such, though, and can be easily filtered out [15:53:57] gilles: awesome , on the schema itself? [15:54:09] gilles: cause we need to filter them fro our queries [15:54:22] (which queries?) [15:54:28] also: https://grafana.wikimedia.org/d/000000218/navigation-timing-by-browser?viewPanel=6&orgId=1&var-source=navtiming2&var-metric=loadEventEnd&var-browserFamily=All&var-browserVersion=all&var-m=p75 [15:54:33] oversampled: https://grafana.wikimedia.org/d/000000218/navigation-timing-by-browser?viewPanel=6&orgId=1&var-source=navtiming2_oversample&var-metric=loadEventEnd&var-browserFamily=All&var-browserVersion=all&var-m=p75 [15:54:50] about 60% of the incoming beacons are from oversampled clients [15:55:04] 700 is_oversample:True, 300: false [15:58:49] Krinkle: see https://superset.wikimedia.org/superset/dashboard/73/ [15:58:56] Krinkle: this is a measure of change in UA [15:59:01] Krinkle: see OS [16:02:17] I don't expect EL intake to be representative of real user device distribution. We have page view data for that more easily accessible, right? I'd expect all sorts of campaigns to be localised to certain countries, browsers, devices or user groups. [16:02:55] Im probably missing something here :) but naively, I'd expect this not to filter out oversampled data, if it is meant to show where/how EL is being used. [16:04:25] Krinkle: on meeting, will answer later [16:04:40] ah, this isn't showing browser usage. interesting. it's purely about entropy. Cool. [16:05:22] I don't know of this is opt-in for certain schemas/topics only, but yeah, to include navtiming there youd' want to ignore oversampled :) [17:21:06] Krinkle: to exclude oversampled in our selects (to confirm) isOversample ==true is sufficient, correct? [17:22:02] yes [17:22:42] nuria: or, if processing old data as well, then by excluding there 'isOversample' exists and is false (non-existant or true means not oversampled) [17:23:03] the property is technically still optional [18:24:22] Krinkle: just added https://gerrit.wikimedia.org/r/c/analytics/refinery/+/628169/ [18:24:34] Krinkle: let me see if we are casting to false absence [18:25:48] I guess the column always exists at this point even if the individiual row has no data [18:25:53] so would be null? [18:25:55] maybe != true would work [19:00:59] AaronSchulz: few min late, be there in 3 [22:01:53] hi there, I have a mediawiki instance in which the "$cur" at the beginning of the splitPath() function of RelPath (line 44/45 of the current git master) throws an error that slows MW down heavily because $cur seems to be empty/uninitialized and therefore the access to $cur[0] fails. Would you consider that a bug or rather a misconfiguration (or both)?