[00:07] <StevenK> The bugtask subscriber vocab is built in place
[01:03] <StevenK> wgrant: mawson is updated, my patch cowboyed and JS unshelved, but it only renders timeouts
[01:04] <StevenK> Ah, /auditor finally works
[01:05] <lifeless> zomg'
[01:05] <StevenK> lifeless: Ah, the auditor project page on DF
[01:05] <lifeless> oh
[01:20] <StevenK> wgrant: Right, so the exact matches are great for DF
[01:20] <StevenK> Very fast
[01:20] <wgrant> StevenK: I'd certainly hope so!
[01:20] <StevenK> Maybe we want to restrict the MP vocab
[01:21] <wgrant> One thing I thought of is that you probably want to do a case-insensitive match
[01:21] <wgrant> On Branch.name
[01:21] <StevenK> wgrant: So searching for 'devel' on proposing a LP branch returns too many matches
[01:21] <StevenK> But I'd honestly expect to usually paste in a URL
[01:22] <wgrant> StevenK: You're conditioned to paste a URL because search is so useless today
[01:22] <wgrant> It usually times out, and when it doesn't time out it doesn't return useful results unless you're very accurate
[01:23] <StevenK> Maybe we want to search on unique_name too
[01:23] <StevenK> But I discounted that
[01:24] <wgrant> Prefix search on unique_name makes sense if there's a / in the term
[01:24] <wgrant> Otherwise I don't think it's valuable
[01:27] <StevenK> wgrant: Thoughts on using BranchRestrictedOnProduct for BMP's vocab?
[01:28] <wgrant> StevenK: Remember that BMPs work for more than products
[01:28] <StevenK> wgrant: Prefix search? Then searching for launchpad/devel won't work
[01:30] <wgrant> StevenK: That's true.
[01:30] <wgrant> But mmm
[01:30] <StevenK> contains_string is probably too heavy handed for unique_name
[01:31] <wgrant> Yes, but it's all we have without schema changes
[01:34] <wgrant> And I don't think it's going to be toooo long for a trigram to be effective
[01:34] <wgrant> But you should try.
[01:36] <StevenK> You mean like http://pastebin.ubuntu.com/1641700/ :-)
[01:37] <wgrant> StevenK: Right, that sort of thing
[01:37] <wgrant> Ideally a unique_name match would be component-wise rather than character-wise, but we can't really do that atm
[01:38] <StevenK> Right
[01:38] <StevenK> Let me patch that onto DF
[01:38] <StevenK> wgrant: Oh, shall we TGRM up name and unique_name ?
[01:39] <wgrant> StevenK: Try with and without
[01:39] <wgrant> Max length of unique_name today is 141
[01:39] <wgrant> Which isn't great, but not fatal
[01:40] <StevenK> What's that one?
[01:40] <wgrant>  ~openerp-commiter/openobject-addons/trunk-configuration_wizard_improvement-atp-improve-general-apply-button-color-in-configuration-wizard-pja |         141
[01:40] <StevenK> That's digusting
[01:40] <StevenK> *disgusting
[01:41] <wgrant> Then
[01:41] <wgrant>  ~78luphr0rnk2nuqimstywepozxn9kl19tqh0tx66b5dki1xxsh5m-launchpad-a811i2i3ytqlsztthjth0svbccw8inm65tmkqp9sarr553jq53in/syncany/delimiters       |         135
[01:41] <wgrant>  ~78luphr0rnk2nuqimstywepozxn9kl19tqh0tx66b5dki1xxs-phn5hho65-a811i2i3ytqlsztthjth0svbccw8inm65tmkqp9sarr553jq5/sysbench/sysbench-pg           |         131
[01:41] <wgrant> Then more openerp
[01:41] <StevenK> I can't complain about the 135 or 131, those are effectively our fault
[01:42] <wgrant> Yep
[01:42] <wgrant> https://code.launchpad.net/~chipaca/ubuntuone-client/dobey-needs-a-nice-self-contained-dbus-signal-to-let-the-applet-know-when-content-queue-changes
[01:45] <StevenK> Apparently, the searches are all ~ 2 seconds
[01:46] <wgrant> ie. awful
[01:46] <wgrant> get a plan
[01:49] <StevenK> wgrant: http://pastebin.ubuntu.com/1641721/
[01:49] <wgrant> StevenK: Heh
[02:02] <wgrant> StevenK: FYI I prefer to name cowboyed indices and tables temp_* so they can be easily found and removed
[02:06] <StevenK> It's still 440 ms with the TRGM
[02:36] <StevenK> wgrant: Would gist rather than gin help?
[02:36] <wgrant> StevenK: No
[02:36] <wgrant> GiST vs GIN is mostly a write vs read tradeoff
[02:36] <wgrant> GIN is slower at writes, but faster at reads
[02:36] <wgrant> So we want GIN
[02:38] <StevenK> But then we're stuffed even with the index
[02:38] <wgrant> Not if it's a relatively consistent 4s
[02:38] <wgrant> It's not create
[02:38] <wgrant> great
[02:38] <wgrant> But it's better than 2s consistent.
[02:38] <wgrant> And will be faster on prod
[02:38] <wgrant> StevenK: Is this unique_name or name?
[02:39] <StevenK> The former
[02:39] <wgrant> Which is rare
[02:39] <wgrant> though most of the length will be in name too, see how it performs
[02:39] <StevenK> That index is just creating now
[02:42] <StevenK> TRGM for name is 300ms
[02:42] <wgrant> Reasonable.
[02:43] <StevenK> I was expecting more on the order of 100
[02:43] <StevenK> Not 300 or 400
[02:43] <wgrant> Not with a long large GIN index
[02:44] <StevenK> What's the \d magic to see the index size?
[02:44] <wgrant> \di+
[02:45] <StevenK> Holy crap, they're 200MiB all up
[02:47] <StevenK> wgrant: So, push the changes for this branch, and create a new branch for two shiny TRGMs?
[02:48] <wgrant> I think so
[03:14]  * StevenK stabs the branch scanner and reaches for carob
[03:33] <StevenK> wgrant: https://code.launchpad.net/~stevenk/launchpad/branch-trgm-indicies/+merge/148082
[03:33] <StevenK> wgrant: And new-branch-search is updated
[08:59] <adeuring> good morning