[00:18] baaaaaaah [00:19] No CREATE TABLE IF NOT EXISTS in 8.4 [00:19] Added in 9.1 :( [00:21] WGRANT, Y U NO CREATE TABLE IF NOT EXISTS [00:22] Heh [00:32] lifeless_: the temp journal flush is very slow because it calls a function for each row. Fixing it to do a bulk insert is still fairly slow, because of the bugsummaryjournal FKs and TRUNCATE being slower than DELETE for small tables like this. [00:33] wgrant: I'm questioning if my IBug leak fix is really QAable :-/ [00:33] StevenK: Did the test suite pass? [00:33] (yes, it did => qa-ok) [00:33] Or untestable [00:34] It's on qas, so both ec2 and buildbot did [00:34] Indeed. [00:48] wgrant: the flush shouldn't block journal inserts though [00:48] lifeless_: The flush is called as part of the trigger. [00:48] temp journal flushes to journal within the trigger [00:48] Then garbo rolls up the journal [00:50] ah, skimmed over the word temp [00:50] thanks for debugging === lifeless_ is now known as lifeless [02:15] not bad - 386 Exceptions [02:49] lifeless: 100 more than normal :( [02:49] hwdb crap, probably [02:50] also 50 from oops-tools thyself [02:50] touch 500.html -> -50 [02:50] Heh [02:50] Yeah [02:53] wgrant: what's the timeframe for the landing of the code to mirror data into the new access policy schema? [02:53] wallyworld_: I expect the trigger will land in a few hours (just waiting for a branch to be merged from stable to db-devel) [02:54] wallyworld_: But that doesn't include the sampledata updates. [02:54] (and we need to create AccessPolicies on prod and dev before the triggers will do much) [02:54] ok. i have a couple of options to do next. i'll wait for this stuff [02:54] just don't use sampledata [02:54] its deprecated anyhow [02:55] i wasn't going to use sample data [02:55] ah cool, I was misled :P [02:55] lifeless: We have to update the sampledata [02:55] Or none of the private bugs will be valid. [02:55] wgrant: how many tests use private bug sampledata? [02:55] Probably quite enough :( [02:56] might be worth a check [02:56] sampledata is also still critical for test performance, unfortunately. [02:56] lifeless: above, i was talking about waiting for the triggers to land, not sample data [02:57] wallyworld_: https://code.launchpad.net/~wgrant/launchpad/bug-mirror-access-policies/+merge/94722 is the MP, and it has tests which use normal factory methods. [02:58] thanks [02:59] wallyworld_: yes, I was misled by wgrant bringing in sampledata to the convo [02:59] np :-) [02:59] lifeless: Well, for most purposes this isn't useful without the sampledata updates. [02:59] Perhaps for wallyworld it is :) [03:01] perhaps [03:01] wgrant: I would love it if you check and see what the effort to delete the private bugs would be from the sampledata [03:02] lifeless: I have a sampledata vs. factory-ng battle experiment planned for next week. [03:04] you had me at battle [03:07] lifeless: Well, the current factory is reasonably awkward and extremely slow, so I'm going to attempt to improve it. [03:13] lifeless: there needs to be some tal for the basic page chrome. the other tal that's there was from an earlier branch where john just put in sample data as a placeholder. it's totally replaced at runtime. i still see some value in retaining that [03:38] wallyworld_: so, I'd love us to get rid of the duplication [03:38] wallyworld_: how is up to you :) [03:39] lifeless: Your DSL is still misbehaving [03:39] ? [03:41] StevenK: I sispect so [04:34] wgrant: O hai, Mr OCR. [04:35] Lies. [04:36] wut [04:36] how is it bad [04:36] I ran it over THE WHOLE TREE two months ago [04:37] Tell me about it [04:37] I only thought to because Aaron added a whole of bunch of 'from x import (\nsingle module,\n)' [04:45] Bah [04:45] One of them is mine :( [04:45] I also didn't notice the problematic codehosting import [04:45] In the diff [04:46] Most of those are now handled. [04:46] by a # FIRST comment. [04:46] Right. [04:46] * StevenK notes one is wrong and reverts it [04:46] Oh? [04:46] lib/lp/services/librarianserver/testing/server.py [04:47] What's wrong with it, beyond the superfluous canonical import? [04:47] It's been moved to the wrong place. [04:47] It shouldn't exist in the first place. [04:47] There's a few others around too. [04:47] I've listed in them in my review that I'm about to finish. [04:48] Right [04:48] But in general anything importing anything from canonical is wrong, unless it's using __file__ to find icing. [04:48] lib/lp/services/librarianserver/testing/server.py makes use of canonical.__file__ [04:49] Ah, it's possible they all do. [04:49] But it seems unlikely. [04:49] If they do still need them, fix the formatter to know that canonical is local. [04:49] Reviewed, anyway. [04:49] I've already pushed the change that does [04:49] Hah [04:49] Look at how it uses canonical.__file__ [04:49] 'tis stupid. [04:49] Pls fix. [04:50] :q [04:50] To do what instead? [04:50] It uses canonical.__file__ so it can get the root of the tree. [04:50] Possibly use lp.__file__ instead. [04:51] Most other cases are probably similar. [04:51] Might as well wipe them all out now. [04:51] canonical.launchpad.__file__ is usually valid, though, as it's used to find icing. [05:10] wgrant: What do you think about http://pastebin.ubuntu.com/860097/ ? [05:14] StevenK: Looks reasonable. [05:16] wgrant: Recommendation for the use of canonical.__file__ in lib/lp/services/config/doc/canonical-config.txt? [05:17] StevenK: s/canonical/lp/g [05:21] steven@liquified:~/launchpad/lp-branches/format-imports-cleanup% bzr grep canonical.__file__ | wc -l [05:21] 0 [05:25] wgrant, StevenK: I think config.root gets you the root better [05:26] stub: It's testing config.root. [05:26] oh :) [05:26] I was going to say the same thing until I actually read the test. [05:39] zomg [05:39] it works [05:40] Which bit? [05:43] Bug searches without any joins at all. [05:43] (including for security checks, which were the big remaining pain) [05:45] wgrant: interesting [05:48] Still waiting for DF to finish regenerating the data, but the plan looks good. [05:55] whats the schema for that ? [05:57] wgrant: ^ [05:58] Include AccessPolicyArtifact.policy and AccessArtifactGrant.grantee in arrays on BugTaskFlat. Not feasible to index, but a whole lot faster in a scan than joining against BugSubscription. [05:58] Because we can calculate the policies and grantees to match against once at the start of the query. [05:58] interesting [05:58] I'm very glad you're spending time trying different htings [06:02] Hmm [06:03] Well, that's better than expected. [06:03] Seq Scan on bugtaskflat (cost=0.00..958352572.67 rows=193926 width=4) (actual time=0.020..970.982 rows=91164 loops=1) [06:03] Total runtime: 1026.971 ms [06:03] So even completely unindexed, as long as it's hot it only takes a second to do a full scan. [06:03] I think that is a reasonable next victory. [06:03] 'cause this table should be pretty damn hot. [06:04] GIN and GIST indexes support array operations. Landscape are using them. [06:04] stub: Yeah, but it's not too useful here. [06:04] Not sure if they are the operators you need though (IS IN etc.) [06:04] k [06:04] It's possible we can get something out of BitmapAnding [06:04] But in general this shouldn't be too bad. [06:04] Because the arrays are all small [06:05] Normally only one policy, and <5 grantees. [06:05] The index may provide significant benefit if you're unable to see most of the bugs, eg. in a private project. [06:05] But we'll see. [06:06] Still pulling up 91k rows, which is a lot even if half the estimate [06:06] (the above is a COUNT(*) of a default Ubuntu bug search, with bugs visible to me) [06:06] So it's meant to be 91k rows [06:06] ok [06:06] And better than what we already have for sure :) [06:06] Oh yes. [06:06] Oh [06:06] And it's still treating the CTEs as correlated. [06:06] Huh [06:06] It's not meant to do that... [06:07] Ah, no it's not, just the top-level. === stub1 is now known as stub [06:24] need more rams [06:26] Bah, forgot to include tags in this rebuild. [06:29] Hmm [06:29] But ideally need stats for tags [06:29] Which means forcing them into a tsvector AFAIK :( [06:32] Anyway, this will hopefully be fast enough for now, and we can easily garbo up a new fact table online later. [06:34] Hmm, under-selective FTI is slow, even with GIN. Possibly just that DF can't really fit the full indexes and tables in cache. [07:12] * wgrant declares the experiment a great success and shall propose it tomorrow. [07:12] awesome [07:13] wgrant: if you have a test script I can run it on asuka [07:17] lifeless: Are you around tomorrow? [07:17] a [07:17] yes [07:29] lifeless: Great, will throw stuff at you then. [07:36] 24 [07:36] ugh [08:51] StevenK: perhaps 'make' should format imports [08:51] That's a bad, bad precedent I think. [08:55] good morning [08:55] Lint could, though. [08:55] Morning adeuring. [08:59] wgrant: also, please to try doubling our data set with the flatbugtask perf tests [08:59] wgrant: [estimate behaviour if we sit on it for a while after deploying] [08:59] fuuuuu [08:59] but OK [08:59] still need to set up proper indexes and such tomorrow. [09:00] But even with no indices it's largely better than we have now. [09:00] Even on dogfood. [09:00] Which can't fit the full table and indices in RAM [09:00] (well, it could if there was nothing else running, but there is) [09:01] yeah [09:01] If you can honestly state you're not worried about behaviour with our dataset doubled, I will take your word on it. [09:01] My point is that we need to make /some/ thought and assessment about it. [09:01] A lot of sort and tag/fti filter cases still degrade to an in-memory sort, so yeah, I'm a little worried. [09:02] But I've looked at it a bit. [09:02] fti often turns into a BitmapAnd, which drops order, so when the result is large it gets bad. [09:02] But it's still not more than 1.3s hot, even when it's sorting the whole set. [09:03] I'm going to look tomorrow morning at throwing tags into the first iteration. [09:03] But I don't have much hope because array stats aren't implemented. [09:03] And stats matter for Ubuntu's tags. [09:04] Mornin' [09:04] Although it may not be so bad to always do a full scan for tag searches if they're inline. [09:04] Mornin' mrevell [09:05] aloha [09:05] Still, it no longer feels like bug search performance is entirely hopeless. [09:10] aloha vera === almaisan-away is now known as al-maisan [11:24] morning === danhg_ is now known as danhg [12:37] StevenK, hey, do you know when you might have some time to finish the review of https://code.launchpad.net/~linaro-infrastructure/launchpad/workitems-migration-script/+merge/93883? I'm happy to bug somebody else if you won't have time for it soon [12:53] I've proposed the Work items UI if anyone is up for reviewing that: https://code.launchpad.net/~linaro-infrastructure/launchpad/workitems-widget/+merge/94790 :) === al-maisan is now known as almaisan-away === almaisan-away is now known as al-maisan [14:31] adeuring, abentley, rick_h_ -- I'm stuck on the phone with my bank. Can we hold standup until this completes please? [14:31] deryck: np [14:31] deryck: gotcha [14:31] deryck: sure thing. [14:32] thanks, guys. sorry. [14:43] lifeless: when you're about can you perhaps have a look over https://bugs.launchpad.net/launchpad/+bug/942523 please ? [14:43] <_mup_> Bug #942523: [Wish] Add a section Trophees in launchpad profile with the Trophee linked to Launchpad & Ubuntu Community < https://launchpad.net/bugs/942523 > [14:45] adeuring, abentley, rick_h_ -- sorry, almost done. let's hangout in 5. [14:46] deryck: ack [14:52] adeuring, abentley, rick_h_ -- https://plus.google.com/hangouts/extras/talk.google.com/orange-standup [15:18] jcsackett, do you have time to mumble [15:30] jtv: are you about? [15:30] czajkowski: I am [15:31] jtv: I dont fully understand the bug here. https://bugs.launchpad.net/launchpad/+bug/942645 [15:31] <_mup_> Bug #942645: Credit strings show up untranslated for kcm-gtk < https://launchpad.net/bugs/942645 > [15:31] czajkowski: you know what "translation credit strings" are? [15:32] jtv: yes [15:32] Launchpad pretends that it has translations for those, but actually those are generated on the fly. [15:33] ah ok [15:33] For some reason, in this case, the Launchpad UI lapses in its pretense that it has translations for these strings. [15:34] If “lapses” is the right word. [15:35] Argh. Look at the computed percentages here! That ought to be fixed: https://translations.launchpad.net/ubuntu/precise/+source/kcm-gtk/+pots/kcm-gtk/de/+details [15:36] czajkowski: if memory serves, we had some mechanism for preventing this — in which case this needs investigation for which unfortunately I do not have time. But it's also possible that this was never solved, in which case there's probably a duplicate bug. [15:37] jtv: nods no worries shall investigate further [15:37] thank you for the help [15:37] Thanks. === al-maisan is now known as almaisan-away === salgado is now known as salgado-lunch === danhg_ is now known as danhg [16:06] adeuring: I just noticed your lazr.jobrunner branch is lp:~adeuring/launchpad/lazr.jobrunner, not lp:~adeuring/lazr.jobrunner/foo. Intentional? [16:06] abentley: usual sloppyness... [16:07] adeuring: Oh, okay. [16:23] sinzui: changes to enum branch pushed. [16:23] fab [16:27] jcsackett, r=me [16:27] sinzui: thanks. [16:48] jcsackett: do you have a sec? [16:48] abentley: sure. [16:48] what's up? [16:48] You're a contributor to ec2 land, and I'm trying to harmonize it with lp-land. [16:49] jcsackett: A difference is that ec2 land updates the commit message. Do you find that useful? [16:49] abentley: you mean the bit about setting the [r=] tags? [16:49] or the commit message option you can pass it? [16:50] jcsackett: I mean that it sets the [r=] tags etc on the merge proposal. [16:50] abentley: i find that useful, yes, in that i would never remember to set those tags myself. [16:50] abentley: i thought lp-land does the same thing, though ... [16:51] jcsackett: You wouldn't have to set them yourself. [16:51] jcsackett: It would already use them for submitting the proposal. [16:51] jcsackett: It's just whether it rewrites them onto the merge proposal. [16:52] abentley: oh! no, i do not find that useful. i actually find it kind of odd that the MP gets tagged with that stuff outside the actual landing. [16:53] and if you have an error with an ec2 land and run it again, i have noticed you can end up with two sets of the [r=] tags. which is vexing. [16:53] jcsackett: Really? It was supposed to avoid duplication, but I guess it's buggy. [16:53] jcsackett: Okay, I'll leave that functionality out. [16:54] abentley: ok. [16:54] jcsackett: thanks. [16:54] abentley: you're welcome. === deryck is now known as deryck[lunch] === almaisan-away is now known as al-maisan === al-maisan is now known as almaisan-away [17:55] czajkowski: hi; crazy bug yes. [17:56] lifeless: one for you then :) [17:58] czajkowski: I'd just mark it low triaged [17:59] lifeless: ah so it may happen ? [17:59] czajkowski: if someone were to add support, with appropriate security, and maintain it, we'd (probably) have no problem doing it, though we wouldn't do it ourselves [17:59] probability 0 that someone will [18:00] mrevell as product manager can say 'no way never' if he likes :) [18:00] heh [18:01] I get to set the bar we need to reach when we do something, but for /what/ we do, thats up to gentle suasion! [18:01] grand job === deryck[lunch] is now known as deryck [18:13] czajkowski: you should never be concerned about triaging wrongly: the worst that can happen is someone can use the [formal] escalation process to request the thing be prioritised [18:13] czajkowski: if its wrong in a simpler manner (e.g. is a regression / isn't or whatever, one of your peers will probably just fix it up) [18:13] lifeless: true, but with that I jst wondered would there be further discussion on the topic given the amount of blogging about the topic and other mails [18:13] oh there probably will be [18:14] * lifeless sadfaces [18:23] would someone help us out and do a release of oops_datedir_repo please? [18:23] I don't want to have to depend on the branch in another oops module [18:24] sinzui: hey [18:25] hi lifeless [18:25] sinzui: I am sure you will know this [18:25] sinzui: why do I get messages 'foo has deactivated registry from bar' every day [18:25] james_w: sure [18:25] :) [18:25] sinzui: is it 'merge' or 'delete' ? [18:25] I was thinking about that this morning [18:26] It is delete [18:26] lifeless, thanks [18:26] I need to write code; this week has had no code writing and I'm getting so very very deprived [18:27] sinzui: so for delete, I'm guessing that that person wasn't involved at all, but the notification happens automagically [18:27] There is some step I want to avoid in the merge process when merging with ~registry. I do not know what case is. [18:27] lifeless, they do, but something is out of order... [18:27] sinzui: separately, there was a bug we found where an admin approved a membership but the notification claimed someone else did [18:27] sinzui: I wonder if it could be the same thing [18:28] We remove super and subteams when deleting. by the point of the merge...there should be no memberships that could trigger that email [18:28] well, a component of the same confusion [18:28] sinzui: there is an owner [18:28] sinzui: and presumably a creator. I wonder if we have fallback code [18:28] crazy cracy fallback code [18:29] lifeless, the person we see in the email is the one that initiated the merge job that performs the delete [18:29] sinzui: ok, so 'foo deleted team bar' is what the mail should say (and it should perhaps go to the old members only) [18:29] lifeless, No [18:30] * lifeless doth not understand [18:30] There is a step missing. Only the owner user doing the action should get an email...but that is not the email we are seeing [18:30] I think the act of merge/delete queue *2* jobs [18:31] 1st job queues the membership changes which is the email we are seeing. [18:31] sinzui: wouldn't it be confusing for the members to have the team disappear? That they were part of ? [18:32] 2nd job queue the merge. The merge happens first, so that the email go out and report ~registry was removed when the team was never a member [18:32] sinzui: ah, I see, you're saying that yes the members get a notification (perhaps 'removed from team' which is a little misleading); and separately we purge it [18:34] lifeless, I looked at the staging memberships of the people reported in the email. They are the team owners and the team is not related to ~registry. So the email really is about ~fnord was deactivated from ~pting [18:35] sinzui: unless registry gets put in fpting first [18:35] lifeless, this is the *very odd* part merge will not take place is any memberships exist. I assume team participation are indeed complete, so the emails should be impossible to send [18:37] james_w: done [18:37] thanks lifeless [18:37] james_w: can you do me a favour and close off any fix-committed bugs? [18:38] zero open bugs [18:38] done! [18:39] plus moving LP to pymongo's bson and getting rid of the compatibility code is probably a good idea, given that the tests in datedir and amqp aren't checking against both implementations [18:39] one inconsistency I've found is that pymongo refuses to serialize things it doesn't know about [18:40] so you can't just put arbitrary objects in the oops [18:40] I don't know what the expected behaviour is there [18:41] that would be nice [18:41] we had a nasty bug where we tossed something bson didn't understand in [18:41] and it just skipped it [18:41] then oops-tools barfed [18:41] ah [18:42] skipped is clearly the wrong thing to do [18:42] I assumed it did repr or something [18:49] sinzui: so, I think we should move all of _merge into the job [18:50] sinzui: there could be a lot of work there, may need multiple transactions to avoid concurrency issues, and there is no UI except for 'yay I did it' or 'boo cannot be done' [18:52] sinzui: (and, I think we should notify about deletes, but thats an orthogonal discussion) [18:52] the options seem to be to repr every value, or try and bsonify every value individually and repr if it fails [18:53] (we're trying to add task *args and **kwargs to the oops, and we don't know the types up front) [18:53] james_w: I thought you said pymongo errors ? [18:53] it does [18:53] and I don't want it to [18:53] james_w: I think that is great [18:53] I want to put the data in without caring about the type in this case [18:54] safe_unicode them all ? [18:54] I suspect that (with the exception of collections) is probably going to DWIM for you [18:55] where does safe_unicode live? [18:55] createhooks [18:57] good idea, /me switches === almaisan-away is now known as al-maisan [19:22] lifeless: ping [19:22] hi [19:22] howdy, I'm trying to figure out your comment in https://code.launchpad.net/~rharding/launchpad/watch_jsbuild/+merge/94379 [19:22] about both the make target and buidlout template? [19:23] can you elaborate for the buildout impaired what I've done wrong/correct way to fix? [19:23] lifeless, hi. I'll get in line behind rick_h_, but would love to get your review of a preload attrs for buglistings branch. [19:24] hah, my faster typing pays off [19:25] I let you win. ;) [19:25] I feel a monty python skit involving a black knight coming on [19:29] deryck: url ? [19:30] (OTP :P) [19:30] lifeless, https://code.launchpad.net/~deryck/launchpad/preload-tags-for-buglistings-901122/+merge/95023 (and thanks!) [19:31] rick_h_: I made that comment on the initial diff [19:31] lifeless: ok, cool [19:31] rick_h_: it looks sane now, but perhaps s/foo.py.in/foo.in/ [19:31] lifeless: ok [19:32] (e.g. I don't think the language is relevant for an executable interface) [19:32] right, ok nop [19:32] np that is === al-maisan is now known as almaisan-away [21:42] deryck: reviewed [21:43] with 2 comments (second coming in a sec) === matsubara is now known as matsubara-afk [21:45] lifeless, ok, thanks, man. [21:52] sinzui: i will be a few minutes late to standup. [21:53] okay [21:53] hi people [21:53] is launchpad still using windmill? [21:56] nup [21:56] died in a fire [21:57] someone should put the smackdown on windmill on -tech [21:59] lifeless, so I've got the last branch up for review, but you don't have to review it now. I'll add the scaling test in that final branch. [21:59] lifeless, but if you're curious: https://code.launchpad.net/~deryck/launchpad/adapt-badges-listing-item-901122/+merge/95063 [22:00] thumper, we still have windmill tests lying around, with the hope they'll get ported to our new yuixhr framework. [22:00] thumper, but they haven't been running for some time. [22:01] if I ever need LOC credit for something I'm working on, I plan to delete them then. ;) [22:03] do what is the new magic? [22:04] html5-browser [22:06] yeah, just plain ol' yui run headless with html5-browser. [22:06] I'm out guys. catch you all later. [22:51] wgrant, https://code.launchpad.net/~sinzui/launchpad/mailman-archive-0