User:Rhelmer/Socorro notes on process type
From MozillaWiki
09:56 < rhelmer> ted: I am trying to figure out how browser/plugin hang pairs are sent exactly. Are there any docs on it, or any part of the code I should look at ? 09:56 < rhelmer> ted: trying to figure out exactly what is sent and when 09:56 < rhelmer> it kind of looks like process_type is set to for the browser and 'plugin' for plugins, with normal crashes having process_type 'content' 09:57 < rhelmer> all the reports are mixed up in the same table so I need ways of identifying which is which :) 09:57 <@ted> when you say "normal" 09:57 <@ted> i don't think you know what you mean 09:57 < rhelmer> heh 09:58 < rhelmer> ted: no I think you're right 09:58 <@ted> before we had OOP anything, all crashes were of one type: browser 09:58 < rhelmer> ok 09:58 <@ted> but they didn't specify that anywhere, because there was only one type 09:58 < rhelmer> sure 09:58 <@ted> when we added OOPP, we started sending process_type plugin 09:58 <@ted> and lack of process_type is assumed 'browser' 09:58 < KaiRo> ted: ok, I'll file a bug for it then - would you think we should omit it or prepend it? 09:58 -!- Usul_phone is now known as Usul 09:58 <@ted> KaiRo: probably just omit 09:59 <@ted> OOP content (like in mobile firefox) sends process_type 'content' when the content process crashes 09:59 < rhelmer> ted: that makes more sense 10:00 <@ted> okay 10:00 <@ted> when a plugin hangs, we send two reports: one for the plugin system board on processor01 in production 09:44 < KaiRo> ted: I just looked at https://crash-stats.mozilla.com/report/list?signature=RtlpAdjustHeapLookasideDepth because I saw some increase in that signature, and it looks like the second frame is a third-party DLL in most cases - should we skiplist that signature? if so, irrelevant or prepend? 09:47 <@ted> KaiRo: that's interesting 09:47 <@ted> sounds like that top frame is a kernel API used to implement malloc 09:47 -!- Bebe [bebe@601F3B17.33662590.A5830293.IP] has quit [Quit: Leaving] 09:48 <@ted> that particular crash looks like it's an LSP :-/ 09:49 < rhelmer> hey peterbe 09:49 < rhelmer> wb 09:49 < rhelmer> peterbe: just responded to your email 09:50 < rhelmer> peterbe: I want to make sure we're all on the same page, let's take it to the public mailing list soonishly 09:54 < KaiRo> ted: I have seen Rtlp* signatures in a number of cases somewhere and I wonder if we should do skiplist stuff on those - if that's mostly LSPs that's not nice, but we know they tend to be a pain as we know 09:54 <@ted> KaiRo: sounds like it's the right decision, having a kernel API as a signature is basically never correct 09:56 < rhelmer> ted: I am trying to figure out how browser/plugin hang pairs are sent exactly. Are there any docs on it, or any part of the code I should look at ? 09:56 < rhelmer> ted: trying to figure out exactly what is sent and when 09:56 < rhelmer> it kind of looks like process_type is set to for the browser and 'plugin' for plugins, with normal crashes having process_type 'content' 09:57 < rhelmer> all the reports are mixed up in the same table so I need ways of identifying which is which :) 09:57 <@ted> when you say "normal" 09:57 <@ted> i don't think you know what you mean 09:57 < rhelmer> heh 09:58 < rhelmer> ted: no I think you're right 09:58 <@ted> before we had OOP anything, all crashes were of one type: browser 09:58 < rhelmer> ok 09:58 <@ted> but they didn't specify that anywhere, because there was only one type 09:58 < rhelmer> sure 09:58 <@ted> when we added OOPP, we started sending process_type plugin 09:58 <@ted> and lack of process_type is assumed 'browser' 09:58 < KaiRo> ted: ok, I'll file a bug for it then - would you think we should omit it or prepend it? 09:58 -!- Usul_phone is now known as Usul 09:58 <@ted> KaiRo: probably just omit 09:59 <@ted> OOP content (like in mobile firefox) sends process_type 'content' when the content process crashes 09:59 < rhelmer> ted: that makes more sense 10:00 <@ted> okay 10:00 <@ted> when a plugin hangs, we send two reports: one for the plugin http://wiki.mozilla.org/Breakpad | Meetings are Wednesdays at 11am PT, 92 300 process and one for the browser process 10:00 -!- Usul is now known as Usul_away 10:00 <@ted> they should be identical to what we'd send if each of those processes had crashed, with the addition of the HangID 10:01 < KaiRo> we just don't have reports for content crashes in Socorro yet, but that's a different topic :) 10:01 < rhelmer> KaiRo: gtk though thanks :) 10:02 < rhelmer> ted: ok my main confusion was why the process_types in the reports table were (, 'plugin', 'content'), now I understand what those are 10:02 < rhelmer> ted: I am just doing a self-join on the reports table to build this report, but I needed a way to identify browser.process_type versus plugin.process_type, for the report 10:03 < rhelmer> (well to be able to disambiguate them at all in the join really) 10:03 < rhelmer> ted: thanks for the walkthrough, I don't suppose this is written down anywhere ? :) 10:03 <@ted> um 10:03 <@ted> probably not 10:03 < rhelmer> I might stick it up on my wiki just to have it as reference 10:03 <@ted> please do