[Sat September 4 2010] [11:04:34 am] mmcgrath: Hey, I've almost got a new pkgdb ready to deploy -- I think I'll finish getting it deployed in stg tonight and then plan on doing it on Monday. [Sat September 4 2010] [11:04:58 am] mmcgrath: There's one thing about it that I'm thinking of deferring until later in the week; probably need your help to figure out how we want to do it. [Sat September 4 2010] [11:05:34 am] mmcgrath: mbacovsk set up static icons -- currently we store the app icons in the db and have to query the db for every single icon. [Sat September 4 2010] [11:05:59 am] mmcgrath: The new code can be configured to put a url to somewhere on the filesystem instead. [Sat September 4 2010] [11:06:43 am] mmcgrath: I think the idea is that the repository sync script reads the icons in and writes them out to a directory structure on an nfs share. [Sat September 4 2010] [11:07:16 am] mmcgrath: Then we use that nfs share on the app servers and have apache on each app server serve it. [Sat September 4 2010] [11:07:57 am] mmcgrath: I think that the icons need to be available on each app server so that the pkgdb can tell whether an icon exists for the app or if it needs to fallback on the "no icon" image. [Sat September 4 2010] [11:08:12 am] ehh, that's ok I guess. We'll have to remove pkgdb from app5 and 6 [Sat September 4 2010] [11:08:18 am] since they don't have nfs access [Sat September 4 2010] [11:08:49 am] mmcgrath: You, mbacovsk, and I can go over if that's how it really works on Monday or later this week. [Sat September 4 2010] [11:09:02 am] are the database accesses causing issue? [Sat September 4 2010] [11:09:05 am] mmcgrath: And then we can get it setup and flip the configuration switch to enable it. [Sat September 4 2010] [11:09:15 am] mmcgrath: Well -- yes and no. [Sat September 4 2010] [11:09:47 am] mmcgrath: The db accesses are part of what made the postgres logs for queries to the front page so big recently [Sat September 4 2010] [11:10:56 am] mmcgrath: The fact that the icons are dynamically generated makes requests for people who have high latency / low bandwidth to the pkgdb less attractive -- since they can't cache the content. [Sat September 4 2010] [11:11:02 am] is it a query that happens in page creation, or when the browser requests /blah/image.jpg ? [Sat September 4 2010] [11:11:20 am] When the browser requests /blah/image [Sat September 4 2010] [11:11:37 am] we just need proper headers on them then, varnish and expires should be able to handle that well. [Sat September 4 2010] [11:12:34 am] mmcgrath: question brought to my attention by someone I was showing fedora to - the installer (well, after you reboot) talks about time servers (a fedora pool)... FI doesn't run those does it? (I hadn't heard of them..now I'm just curious) [Sat September 4 2010] [11:12:35 am] mmcgrath: Okay -- if you'd like to work it that way, we can sit down with mbacovsk and work out how to tune varnish for that. [Sat September 4 2010] [11:13:06 am] CodeBlock: correct, we don't, the public pool has some fedora branded though for some reason. [Sat September 4 2010] [11:13:18 am] abadger1999: and it may be easier to just throw those and have apache serve them anyway [Sat September 4 2010] [11:13:22 am] mmcgrath: Oh.. some people wnat to reuse the icon data for their own nefarious purposes (like pkgkit growing an "application installer") -- but I think all they need is a stable url. [Sat September 4 2010] [11:13:26 am] we'll just have to work through the details and find out [Sat September 4 2010] [11:13:37 am] hmm [Sat September 4 2010] [11:13:37 am] k [Sat September 4 2010] [11:14:04 am] mmcgrath: We'll need to talk with mbacovsk and seth about whether that should be doable with cached images.... I can't think of why not but I haven't thought as muchabout it. [Sat September 4 2010] [11:14:42 am] lmacken wanted the icons for fedoracommunity too -- I imagine the requirements would be the same. [Sat September 4 2010] [11:15:37 am] Anyhow -- that's what's coming this week for the pkgdb :-) [Sat September 4 2010] [11:15:38 am] k, I'll make sure to take some notes on what headers are needed. [Sat September 4 2010] [11:15:49 am] etagging would be useful too but probably more difficult for non-browser libraries. [Sat September 4 2010] [11:16:02 am] it's something we should look at though, I don't know if the urlgrabbers, libcurl's, etc have it. [Sat September 4 2010] [11:16:05 am] I suspect not. [Sat September 4 2010] [11:16:12 am] since they don't keep local caches [Sat September 4 2010] [11:16:12 am] [Sat September 4 2010] [11:17:03 am] * mmcgrath thinks the work I've been doing with caching would be a good FUDCon talk. [Sat September 4 2010] [11:17:44 am] yeah, that would definitely be an interesting technical talk. [Sat September 4 2010] [11:18:06 am] What can you tune with caching? How do you do it? [Sat September 4 2010] [11:18:57 am] yeah [Tue September 14 2010] [1:24:53 pm] Join mbacovsk has joined this channel (~mbacovsk@82.117.156.237). [Tue September 14 2010] [1:25:50 pm] abadger1999: mmcgrath: ping [Tue September 14 2010] [1:25:57 pm] mbacovsk: hey [Tue September 14 2010] [1:26:06 pm] mmcgrath: Greetings. [Tue September 14 2010] [1:26:16 pm] Join Karlik has joined this channel (~Karlik@fedora/Karlik). [Tue September 14 2010] [1:26:17 pm] mbacovsk: greetings :-) [Tue September 14 2010] [1:26:29 pm] hi :) [Tue September 14 2010] [1:27:06 pm] I'd like to discuss static icons [Tue September 14 2010] [1:27:15 pm] [Tue September 14 2010] [1:27:24 pm] mbacovsk: sure, hit us [Tue September 14 2010] [1:28:07 pm] :) the issue is web performance during display of application list [Tue September 14 2010] [1:28:22 pm] [Tue September 14 2010] [1:28:50 pm] The images were looked for in the db one by one which turned to be tooo slow [Tue September 14 2010] [1:29:54 pm] I prepared solution that can set the images to beeing served from disk as static content [Tue September 14 2010] [1:30:03 pm] the good ways to fix that are expires headers, etags and css sprites. [Tue September 14 2010] [1:30:14 pm] Quit nb has left this server (Read error: Operation timed out). [Tue September 14 2010] [1:30:15 pm] serving from disk won't be much faster then from the database. [Tue September 14 2010] [1:30:26 pm] if at all [Tue September 14 2010] [1:30:52 pm] you will save the db hit/search at least [Tue September 14 2010] [1:31:05 pm] yeah but that's not a very expensive operation [Tue September 14 2010] [1:31:13 pm] and sometimes could be faster then local disk depending on conditions [Tue September 14 2010] [1:31:16 pm] .headers https://admin.fedoraproject.org/pkgdb/appicon/show/Abe [Tue September 14 2010] [1:31:17 pm] mmcgrath: apptime: D=255996, content-length: 2497, set-cookie: tg-visit=231188235a6c3ef845deb00aeba21e672aea924d; Path=/; secure;, proxytime: D=257421, expires: Sun, 19 Sep 2010 20:31:15 GMT, connection: close, server: Apache/2.2.3 (Red Hat), appserver: app07.phx2.fedoraproject.org, proxyserver: proxy01.phx2.fedoraproject.org, cache-control: max-age=432000, date: Tue, 14 Sep 2010 20:31:15 GMT, content-type: (1 more message) [Tue September 14 2010] [1:31:18 pm] .more [Tue September 14 2010] [1:31:20 pm] mmcgrath: image/png [Tue September 14 2010] [1:31:25 pm] so lets see here. [Tue September 14 2010] [1:31:32 pm] we've got good expires headers there [Tue September 14 2010] [1:31:34 pm] no etag [Tue September 14 2010] [1:31:43 pm] mmcgrath: Well... actually db latency can be a big hit. [Tue September 14 2010] [1:31:52 pm] abadger1999: as can disk conteiton [Tue September 14 2010] [1:31:57 pm] but the truth is i'm not experienced in web server [erformance tuning on the server, so I'm open to anything else you can offer. [Tue September 14 2010] [1:32:14 pm] But if we cache the images then we're getting ahead of that. [Tue September 14 2010] [1:32:16 pm] mbacovsk: if you want to do this the 'right' way, I think css sprites are the way to do it. [Tue September 14 2010] [1:32:38 pm] Quit nhorman has left this server (Quit: Leaving). [Tue September 14 2010] [1:33:01 pm] Quit Sonar_Guy has left this server (Quit: Leaving). [Tue September 14 2010] [1:33:17 pm] mbacovsk: you have both solutions in place? run an ab on both [Tue September 14 2010] [1:33:21 pm] and post the results [Tue September 14 2010] [1:33:32 pm] from the webapp POV the bottleneck for me is the database, but that can be resolved much efficiently on the server [Tue September 14 2010] [1:33:40 pm] probably [Tue September 14 2010] [1:33:55 pm] do an ab -c1 -t 15 http://localhost/path/to/image [Tue September 14 2010] [1:33:55 pm] Quit MostafaDaneshvar has left this server (Quit: night). [Tue September 14 2010] [1:34:16 pm] even then though, the way to fix this is to remove the rtt between the browser and proxy server. [Tue September 14 2010] [1:34:23 pm] which is expires headers and css sprites. [Tue September 14 2010] [1:34:37 pm] but it'd be good to get some actual numbers to compare first [Tue September 14 2010] [1:35:05 pm] mmcgrath: I was thinking of that but the are loads of images that have to be 'spritefied' periodicaly, i'm not sure if that is the bet way [Tue September 14 2010] [1:35:44 pm] Quit jskarvad has left this server (Quit: Quit). [Tue September 14 2010] [1:37:24 pm] well, best depends on who you're looking at. End user or developer. Ideally it'd be somewhere in the middle. But the way we're doing things. [Tue September 14 2010] [1:37:28 pm] https://admin.fedoraproject.org/pkgdb/apps/name/list/a* [Tue September 14 2010] [1:37:31 pm] that's 166 images. [Tue September 14 2010] [1:37:34 pm] that's 166 requests. [Tue September 14 2010] [1:37:45 pm] with a sprite you could taurn that into 1 or 2 requests [Tue September 14 2010] [1:38:13 pm] Part ninjazjb has left this channel. [Tue September 14 2010] [1:38:38 pm] now, we could enable varnish for those images. that would help save some time [Tue September 14 2010] [1:39:40 pm] * mmcgrath tries in staging [Tue September 14 2010] [1:42:14 pm] Join charlymanja has joined this channel (~manjarrc@201.102.96.137). [Tue September 14 2010] [1:42:40 pm] mbacovsk: abadger1999: keep in mind with local storage, you're really talking nfs storage unless you're going to store the icons in the rpm or something [Tue September 14 2010] [1:43:00 pm] mmcgrath: Note, there's other people wanting to use the images... if we spriteify them, all those consumers will need to know how to process that as well. [Tue September 14 2010] [1:43:10 pm] [Tue September 14 2010] [1:43:19 pm] we might be able to make that part of the build process [Tue September 14 2010] [1:43:24 pm] combine the images into a sprite [Tue September 14 2010] [1:43:26 pm] * mmcgrath doesn't know [Tue September 14 2010] [1:43:34 pm] I've never actually done this, I just know the theory on how it works :) [Tue September 14 2010] [1:43:55 pm] mmcgrath: I think I originally thought of "precaching" them on the proxies similar to how the static website was setup for release days. [Tue September 14 2010] [1:44:02 pm] nfs was your idea :-) [Tue September 14 2010] [1:44:21 pm] we can precache too but I don't think it'll help that much [Tue September 14 2010] [1:44:24 pm] * mmcgrath is testing that now :) [Tue September 14 2010] [1:45:20 pm] mmcgrath: the thing is that there is 5000+ app icons, that means 3400pxx3400px + mapping. That sounds viable, I can look into that [Tue September 14 2010] [1:45:40 pm] you'd probably want to break them into different sheets [Tue September 14 2010] [1:45:42 pm] like 5 or 6 [Tue September 14 2010] [1:45:50 pm] mbacovsk: How does that work for bodhi and app-install? [Tue September 14 2010] [1:45:51 pm] mbacovsk: any luck on the ab? [Tue September 14 2010] [1:46:21 pm] the sprite has to be generated by pkgdb-sunc-yum though [Tue September 14 2010] [1:46:24 pm] you could also just not display the icons in list view [Tue September 14 2010] [1:46:27 pm] Quit ggruener has left this server (Remote host closed the connection). [Tue September 14 2010] [1:46:45 pm] mmcgrath: And on the front page.... [Tue September 14 2010] [1:47:01 pm] yeah, pretty much wherever you'd list them [Tue September 14 2010] [1:47:09 pm] :) [Tue September 14 2010] [1:47:24 pm] you'd just show them in detail view when it's just one icon at a time [Tue September 14 2010] [1:48:24 pm] Quit giarc has left this server (Remote host closed the connection). [Tue September 14 2010] [1:48:26 pm] Still wouldn't help bodhi, app-install... [Tue September 14 2010] [1:49:01 pm] do they do list icon views? [Tue September 14 2010] [1:49:47 pm] They want to -- They were waiting on static icons. [Tue September 14 2010] [1:51:16 pm] I think that a url like this should be fine for bodhi's use case but I'm not sure: https://admin.fedoraproject.org/pkgdb/appicon/show/Terminator [Tue September 14 2010] [1:51:19 pm] lmacken: ping [Tue September 14 2010] [1:51:39 pm] abadger1999: yo [Tue September 14 2010] [1:51:55 pm] yeah, that url would be fine [Tue September 14 2010] [1:52:06 pm] lmacken: So mmcgrath, mbacovsk and I are talking about the best way to make retrieving icons work. [Tue September 14 2010] [1:52:23 pm] lmacken: One thing about that url -- it's an application url, not a package icon. [Tue September 14 2010] [1:52:33 pm] err... package url. [Tue September 14 2010] [1:52:42 pm] gotcha [Tue September 14 2010] [1:52:47 pm] Is that okay? [Tue September 14 2010] [1:53:04 pm] well... I'm not sure how I would determine which updates in bodhi had an icon or not. [Tue September 14 2010] [1:53:09 pm] [Tue September 14 2010] [1:53:33 pm] although, I could do what I did in fedoracommunity with the hackergochis... some javascript that caught the load error, and replaced it with a default one. so that is fine [Tue September 14 2010] [1:53:48 pm] abadger1999: that url actually is icon url [Tue September 14 2010] [1:53:51 pm] Join nb has joined this channel (~nb@fedora/GSWoT.Member.nb). [Tue September 14 2010] [1:54:04 pm] lmacken: So for you, something that had a query a package name, get back an icon or set of icons for the package would work better? [Tue September 14 2010] [1:54:19 pm] lmacken: Oh right, and you probably want this for community as well? [Tue September 14 2010] [1:54:31 pm] the other option is /pkgdb/static/appicon/Terminator/Terminator.png [Tue September 14 2010] [1:54:50 pm] mbacovsk: Right.. I mean -- the key value is an application name rather than a source package name. [Tue September 14 2010] [1:54:56 pm] abadger1999: yeah, I want to take a package name, and get an icon :) [Tue September 14 2010] [1:55:18 pm] ideally not through some API... through a predictable URL [Tue September 14 2010] [1:56:24 pm] lmacken: Okay -- so if we optimized this by using CSS sprites (basically, one image with a whole bunch of icons and css to tell where to cut the image apart), that would make your life difficult? [Tue September 14 2010] [1:56:56 pm] abadger1999: maybe... wouldn't that be a freaking huge file? [Tue September 14 2010] [1:57:25 pm] lmacken: [13:45:20] mmcgrath: the thing is that there is 5000+ app icons, that means 3400pxx3400px + mapping. That sounds viable, I can look into that [Tue September 14 2010] [1:57:30 pm] [13:45:40] you'd probably want to break them into different sheets [Tue September 14 2010] [1:57:36 pm] like 5 or 6 [Tue September 14 2010] [1:57:46 pm] that seems like a lot of work for little gain? [Tue September 14 2010] [1:58:01 pm] It cuts down on the number of http requests. [Tue September 14 2010] [1:58:03 pm] it's a lot of gain [Tue September 14 2010] [1:58:08 pm] when you're doing lists [Tue September 14 2010] [1:58:08 pm] why not just serve them in a static directory at the proxy layer? [Tue September 14 2010] [1:58:09 pm] mmcgrath: I have the ab results [Tue September 14 2010] [1:58:15 pm] mbacovsk: fpaste it! [Tue September 14 2010] [1:59:12 pm] Quit spoleeba has left this server (Remote host closed the connection). [Tue September 14 2010] [2:00:05 pm] mmcgrath: static http://fpaste.org/M9iO/, from db http://fpaste.org/PX8z/ [Tue September 14 2010] [2:00:24 pm] mmcgrath: just note that my ntb is not latest HW [Tue September 14 2010] [2:00:39 pm] that's ok, it's the comparison. [Tue September 14 2010] [2:02:20 pm] mbacovsk: something didn't work there [Tue September 14 2010] [2:02:24 pm] mmcgrath: I found out that the resulting image is not the same [Tue September 14 2010] [2:02:29 pm] yeah [Tue September 14 2010] [2:02:33 pm] I was just about to say that :) [Tue September 14 2010] [2:03:14 pm] mmcgrath: mmt [Tue September 14 2010] [2:03:32 pm] one of the tests was 4M the other was 20K [Tue September 14 2010] [2:03:41 pm] Quit ricky has left this server (Ping timeout: 255 seconds). [Tue September 14 2010] [2:04:38 pm] Join oget_zzz has joined this channel (~oget@192.12.88.51). [Tue September 14 2010] [2:04:41 pm] mmcgrath: Hmm... if I read those the difference in the images is 4.5K, 3.4K [Tue September 14 2010] [2:04:44 pm] Quit oget_ has left this server (Ping timeout: 276 seconds). [Tue September 14 2010] [2:05:10 pm] 4500720 bytes vs 21972 bytes [Tue September 14 2010] [2:05:16 pm] oh wait [Tue September 14 2010] [2:05:20 pm] The static version completed 940 requests compared to the cherrypy version completing 6. [Tue September 14 2010] [2:05:22 pm] * mmcgrath is looking at the total transferred :) [Tue September 14 2010] [2:05:28 pm] That's where your numbers are differing. [Tue September 14 2010] [2:05:33 pm] yeah [Tue September 14 2010] [2:05:38 pm] mbacovsk: yeah re-do it with the same image :) [Tue September 14 2010] [2:08:21 pm] mmcgrath: from db http://fpaste.org/AF0a/ [Tue September 14 2010] [2:08:34 pm] Quit bhinson has left this server (Quit: Leaving). [Tue September 14 2010] [2:08:50 pm] mmcgrath: static http://fpaste.org/yoWk/ [Tue September 14 2010] [2:09:16 pm] why are the document lengths still different? [Tue September 14 2010] [2:09:54 pm] mmcgrath: 1848 for both [Tue September 14 2010] [2:10:13 pm] but no of requests differs [Tue September 14 2010] [2:10:20 pm] * mbacovsk wonder why [Tue September 14 2010] [2:10:25 pm] abadger1999: ah, yes I had an old tab open [Tue September 14 2010] [2:10:39 pm] mbacovsk: something's very wrong with your setup, we should test this in staging. [Tue September 14 2010] [2:11:01 pm] we're certainly not seeing 1.8 seconds / image request :) [Tue September 14 2010] [2:11:11 pm] mmcgrath: Note that when we've benchmarked in the past, acpache does blow cherrypy out of the water. [Tue September 14 2010] [2:11:50 pm] Quit nb has left this server (Ping timeout: 276 seconds). [Tue September 14 2010] [2:11:53 pm] Quit harish has left this server (Ping timeout: 265 seconds). [Tue September 14 2010] [2:12:03 pm] it does [Tue September 14 2010] [2:12:09 pm] really none of those tests matter though. [Tue September 14 2010] [2:12:13 pm] Yeah, staging is a better env for figuring out what's happening. [Tue September 14 2010] [2:12:17 pm] since we're not really considering serving from the app servers anyway [Tue September 14 2010] [2:14:02 pm] I'm trying to get varnish to ignore the cookie stuff. [Tue September 14 2010] [2:14:02 pm] mmcgrath: when the static image is served cherrypy is left out it is served just by apache. The version from DB has to instantiate cherrypy, db connection, etc. Maybe the result would be more telling if I had just one worker in apache [Tue September 14 2010] [2:16:49 pm] i can tell we're frozen when the volume on fedora-infra goes up :) [Tue September 14 2010] [2:17:02 pm] jokajak: isn't that the truth :) [Tue September 14 2010] [2:17:06 pm] mmcgrath: The static icons are designed so that you can easily switch from generating from db to static and back. So we could deploy it on staging and do some tests. Then we can decide if it is worth to keep it or not.okay? [Tue September 14 2010] [2:17:19 pm] Join ricky has joined this channel (~ricky@fedora/ricky). [Tue September 14 2010] [2:20:15 pm] Quit p3ck has left this server (Read error: Operation timed out). [Tue September 14 2010] [2:23:26 pm] Join harish has joined this channel (~harish@nat/redhat/x-hutyhdlqctnifnjk). [Tue September 14 2010] [2:23:28 pm] Quit maploin has left this server (Quit: ERC Version 5.3 (IRC client for Emacs)). [Tue September 14 2010] [2:25:00 pm] Part d3vnull has left this channel ("exit(1);"). [Tue September 14 2010] [2:26:26 pm] Nick stickster is now known as stickster_afk. [Tue September 14 2010] [2:28:10 pm] abadger1999: mbacovsk: ok, I think I've got it [Tue September 14 2010] [2:30:28 pm] abadger1999: mbacovsk: ok, take a look. [Tue September 14 2010] [2:30:36 pm] https://admin.stg.fedoraproject.org/pkgdb/apps/name/list/a* [Tue September 14 2010] [2:30:42 pm] do a shift refresh to round trip to the database [Tue September 14 2010] [2:30:49 pm] do a regular refresh to hit the proxy server only. [Tue September 14 2010] [2:30:57 pm] you can tail the access log on app1 to verify [Tue September 14 2010] [2:31:01 pm] err app01.stg [Tue September 14 2010] [2:32:34 pm] it's still at 8 seconds for me. [Tue September 14 2010] [2:32:41 pm] at its fastest [Tue September 14 2010] [2:32:56 pm] which is twice as fast and is something we could theoretically do tomorrow [Tue September 14 2010] [2:33:03 pm] walters: ive made a target for f-14 but please just do the rawhide work in rawhide [Tue September 14 2010] [2:34:53 pm] mbacovsk: abadger1999: so what do you think from there? [Tue September 14 2010] [2:47:49 pm] Interesting -- my response times are pretty varied -- 14s - 25s cached. 20s - 36s uncached. [Tue September 14 2010] [2:48:04 pm] mbacovsk: How does that work for you? [Tue September 14 2010] [2:48:11 pm] abadger1999: yeah, it's the rtt times. [Tue September 14 2010] [2:48:47 pm] it's 200 total gets. [Tue September 14 2010] [2:48:50 pm] well 210 [Tue September 14 2010] [2:50:32 pm] abadger1999: check this out [Tue September 14 2010] [2:50:34 pm] go to about:config [Tue September 14 2010] [2:50:47 pm] search for network.http.max-persistent-connections-per-server [Tue September 14 2010] [2:50:48 pm] set it to 25. [Tue September 14 2010] [2:50:59 pm] then try again cached [Tue September 14 2010] [2:51:21 pm] how do you measure the time per request in FF? is it standart feature? [Tue September 14 2010] [2:51:29 pm] mbacovsk: yslow [Tue September 14 2010] [2:51:42 pm] it's *amazing* for these types of problems [Tue September 14 2010] [2:52:01 pm] mmcgrath: That's cheating :-) but yeah, persistent connections can really speed things up. [Tue September 14 2010] [2:52:16 pm] :) [Tue September 14 2010] [2:53:58 pm] Quit pwbarnes has left this server (Remote host closed the connection). [Tue September 14 2010] [2:56:08 pm] Quit nul0op has left this server (Quit: Leaving). [Tue September 14 2010] [2:58:55 pm] * abadger1999 just uses firebug. [Tue September 14 2010] [2:59:04 pm] Quit josemm has left this server (Quit: josemm). [Tue September 14 2010] [2:59:06 pm] SOmeone told me that arora has a great timing interface though :-) [Tue September 14 2010] [3:01:28 pm] abadger1999: I've heard that too :) but was not sure how to do shift refresh in arora. I'll try that [Tue September 14 2010] [3:02:06 pm] jokajak: I'd just create a new package for it if we wanted to do it. [Tue September 14 2010] [3:02:10 pm] the webkit inspector is cool. :) install your fav webkit browser and webkitgtk-devel and then right click and 'inspect element' [Tue September 14 2010] [3:02:13 pm] Quit walters has left this server (Ping timeout: 245 seconds). [Tue September 14 2010] [3:02:27 pm] nirik: I suspect that's what aora is using. [Tue September 14 2010] [3:02:34 pm] yep. should be. [Tue September 14 2010] [3:03:08 pm] midori or ubzi or rekonq etc, all should have the same thing. [Tue September 14 2010] [3:03:11 pm] It's got graphs by time and type (image/stylesheet/etc) and size... lots of interesting things. [Tue September 14 2010] [3:03:23 pm] * abadger1999 should try some of those others. [Tue September 14 2010] [3:05:40 pm] Quit jerboaa has left this server (Quit: Leaving). [Tue September 14 2010] [3:05:43 pm] Join harish has joined this channel (~harish@nat/redhat/x-lbabwsmcvlrxqcmt). [Tue September 14 2010] [3:06:21 pm] it also has a suggestions for making loads faster tab [Tue September 14 2010] [3:06:38 pm] Quit nucleo has left this server (Ping timeout: 276 seconds). [Tue September 14 2010] [3:08:29 pm] nirik: Hmm... Where do I find that tab/what's it labeled? [Tue September 14 2010] [3:08:31 pm] is there an easy way to download a scratch build by task id? [Tue September 14 2010] [3:08:55 pm] abadger1999: it's in the inspector... [Tue September 14 2010] [3:09:19 pm] notting: http://kojipkgs.fedoraproject.org/scratch/username/task_tasknumber [Tue September 14 2010] [3:09:26 pm] notting: I use a script that I believe mbonnet wrote. [Tue September 14 2010] [3:10:09 pm] Join bpepple|lt has joined this channel (~bpepple|l@cpe-184-57-0-60.columbus.res.rr.com). [Tue September 14 2010] [3:10:09 pm] mmcgrath, db02 collectd still doesn't show fas connections [Tue September 14 2010] [3:10:13 pm] abadger1999: 'audits' tab [Tue September 14 2010] [3:12:46 pm] Nick jsmith is now known as jsmith-away. [Tue September 14 2010] [3:12:46 pm] nirik: Hmm... my arora doesn't have that. /me installs rekonq [Tue September 14 2010] [3:13:03 pm] abadger1999: what fedora? [Tue September 14 2010] [3:13:10 pm] nirik: f13 [Tue September 14 2010] [3:13:51 pm] notting: oh, what bodhi issues did you hit today/yesterday ? [Tue September 14 2010] [3:14:09 pm] abadger1999: do you have webkitgtk-devel installed? [Tue September 14 2010] [3:14:31 pm] Quit inode0 has left this server (Quit: Leaving.). [Tue September 14 2010] [3:15:15 pm] nirik: nope. /me installs [Tue September 14 2010] [3:15:24 pm] Nick mbonnet is now known as mbonnet_. [Tue September 14 2010] [3:15:40 pm] abadger1999: I am trying to change that. The inspector should be in the main package, IMHO [Tue September 14 2010] [3:17:17 pm] lmacken: absolutely, positively, refused to believe that https://admin.fedoraproject.org/updates/simple-scan-2.31.91-1.fc14 was actually tagged as a candidate/update [Tue September 14 2010] [3:18:18 pm] Quit mizmo has left this server (Ping timeout: 240 seconds). [Tue September 14 2010] [3:18:23 pm] notting: I had to frob that one by hand the other day, I wonder if I broke it [Tue September 14 2010] [3:18:54 pm] nirik: oh, brilliant. that crashed firefox [Tue September 14 2010] [3:19:25 pm] which? [Tue September 14 2010] [3:19:52 pm] opening crap in that dir. obvs unrelated to kojipkgs apache indexes [Tue September 14 2010] [3:20:40 pm] Quit harish has left this server (Ping timeout: 272 seconds). [Tue September 14 2010] [3:20:51 pm] Quit braoru has left this server (Quit: This computer has gone to sleep). [Tue September 14 2010] [3:21:49 pm] notting: ok, that update should hit stable with the next push [Tue September 14 2010] [3:28:25 pm] Quit dmlloyd_road has left this server (Ping timeout: 252 seconds). [Tue September 14 2010] [3:29:03 pm] does anyone have att's dsl? [Tue September 14 2010] [3:30:50 pm] skvidal, ping. Func on puppet01 is dead [Tue September 14 2010] [3:32:12 pm] abadger1999: any success with arora and audits tab? [Tue September 14 2010] [3:32:35 pm] mbacovsk: Nope, I still didn't find the audits tab in either arora or rekonq. [Tue September 14 2010] [3:33:03 pm] weird. [Tue September 14 2010] [3:33:19 pm] I know it's there in midori. ;) [Tue September 14 2010] [3:33:33 pm] I wonder if I need webkitpart-devel? [Tue September 14 2010] [3:35:39 pm] here same with arora, even webkitgtk-devel didn't help. midori works as expected though [Tue September 14 2010] [3:42:27 pm] Quit plarsen has left this server (Disconnected by services). [Tue September 14 2010] [3:44:19 pm] Join caillon has joined this channel (~caillon@redhat/caillon). [Tue September 14 2010] [3:44:40 pm] humm... perhaps thats the qt webkit that has another package for the inspector? [Tue September 14 2010] [3:45:39 pm] yeah, might be. (looking at midori now) [Tue September 14 2010] [3:45:41 pm] Join pwbarnes has joined this channel (~nman64@fedora/nman64). [Tue September 14 2010] [3:46:15 pm] Nick pwbarnes is now known as pwbarnes_away. [Tue September 14 2010] [3:46:25 pm] dgilmore, no I have qwest [Tue September 14 2010] [3:46:57 pm] nirik: Are audits and console both enabled when you install the devel package? [Tue September 14 2010] [3:47:13 pm] yes. [Tue September 14 2010] [3:47:32 pm] Well, both of those are missing in arora and rekonq. [Tue September 14 2010] [3:47:32 pm] well, the entire inspector is. [Tue September 14 2010] [3:47:48 pm] I had the inspector without installing webkitgtk-devel [Tue September 14 2010] [3:47:54 pm] So they must be somehwat separate. [Tue September 14 2010] [3:48:03 pm] They look the same, though [Tue September 14 2010] [3:48:13 pm] yeah, must be qt vs gtk webkit [Tue September 14 2010] [3:49:01 pm] and I think the qt one lags behind the gtk one, so it might be just a newer version. [Tue September 14 2010] [3:49:18 pm] Could be. [Tue September 14 2010] [3:49:57 pm] abadger1999: mmcgrath: biggest eaters are for sure dijit stylesheets and images. The results are interesting: FF cached 20s, FF uncached 30s, Arora (first load) 30s, Arora (reload) 20s, midori (no matter what) 40s [Tue September 14 2010] [3:51:01 pm] stylesheets for dijit are something I want to do something about -- I think that dojo has a "build" process for turning those into a single stylesheet. [Tue September 14 2010] [3:51:17 pm] I just haven't had the time to look at javascript for quite a while. [Tue September 14 2010] [3:51:52 pm] Join charlymanja1 has joined this channel (~manjarrc@201.102.96.137). [Tue September 14 2010] [3:53:16 pm] * nirik notes for f14 he has a webkitgtk build with the JIT re-enabled that makes midori about 5x faster on javascript stuff. [Tue September 14 2010] [3:55:00 pm] midori also has a disk cache extention, but it's not enabled by default. [Tue September 14 2010] [3:55:44 pm] Quit charlymanja has left this server (Ping timeout: 255 seconds). [Tue September 14 2010] [3:57:19 pm] Join noriko has joined this channel (~noriko@d58-106-141-64.rdl801.qld.optusnet.com.au). [Tue September 14 2010] [3:57:48 pm] Join inode0 has joined this channel (~inode0@fedora/inode0). [Tue September 14 2010] [3:57:48 pm] smooge: i know someone on att and saw they had public ipv6 addressing [Tue September 14 2010] [3:57:57 pm] i was wondering if they are doing it across the board [Tue September 14 2010] [3:58:16 pm] Join nb has joined this channel (~nb@fedora/GSWoT.Member.nb). [Tue September 14 2010] [3:58:18 pm] :( [Tue September 14 2010] [3:58:22 pm] my server is down [Tue September 14 2010] [3:58:41 pm] Quit Karlik has left this server (Quit: Leaving). [Tue September 14 2010] [4:04:00 pm] nb: :( [Tue September 14 2010] [4:04:59 pm] dgilmore, ah that would be sweet. qwest/comcast is our bi-opoly so am limited to 1990's technolgoy [Tue September 14 2010] [4:05:16 pm] Join PaulW has joined this channel (~paul@CPE001310360dac-CM001e6b0f7310.cpe.net.cable.rogers.com). [Tue September 14 2010] [4:05:31 pm] Quit daMaestro has left this server (Ping timeout: 240 seconds). [Tue September 14 2010] [4:07:04 pm] Join daMaestro has joined this channel (~jon@fedora/damaestro). [Tue September 14 2010] [4:11:07 pm] Quit kzak has left this server (Ping timeout: 240 seconds). [Tue September 14 2010] [4:11:33 pm] Join nb_ has joined this channel (~nb@fedora/GSWoT.Member.nb). [Tue September 14 2010] [4:11:50 pm] abadger1999: mmcgrath: so what is the plan? I can look on dojo SDK to optimize the dojo requests tmrw, I'm not sure if we decided to try the sprites [Tue September 14 2010] [4:12:12 pm] mbacovsk: Does static icons give us more of a gain than varnish? [Tue September 14 2010] [4:12:13 pm] Quit nb has left this server (Quit: Leaving). [Tue September 14 2010] [4:12:13 pm] Nick nb_ is now known as nb. [Tue September 14 2010] [4:12:55 pm] mbacovsk: I think I'd rather have bugs fixed and get us moving towards an 0.6.x release with sqlalchemy migrate and a better db schema first... optimizations second. [Tue September 14 2010] [4:13:20 pm] But if things are still unbearably slow where you are then dojo requests can be looked at. [Tue September 14 2010] [4:14:04 pm] Quit ricky has left this server (Ping timeout: 252 seconds). [Tue September 14 2010] [4:14:34 pm] Quit notting has left this server (Quit: Ex-Chat). [Tue September 14 2010] [4:20:47 pm] abadger1999: I have no comparision of varnish and static icons but I believe it will be similar. According to experiments with stg it seems that number of requests is now more of an issue. Also it seems that varnish caching is ignored(?) by midori. [Tue September 14 2010] [4:21:19 pm] yeah -- nirik do you know if that might make sense? [Tue September 14 2010] [4:21:37 pm] maybe setting headers better would help [Tue September 14 2010] [4:22:20 pm] huh...not that i know of... [Tue September 14 2010] [4:23:03 pm] * nirik can look later... [Tue September 14 2010] [4:24:27 pm] abadger1999: mmcgrath: I assume that apps screenshots could be handled the same way as icons, right? [Tue September 14 2010] [4:24:33 pm] yep [Tue September 14 2010] [4:28:12 pm] Join jtanner has joined this channel (~jtanner@12.189.100.66). [Tue September 14 2010] [4:40:38 pm] anyone around that knows anything about pacemaker? [Tue September 14 2010] [4:42:17 pm] abadger1999: I'll try to follow the webkit audits and fix the headers. I'll go on with the collection/branch and other stuff. [Tue September 14 2010] [4:43:11 pm] mbacovsk: Thanks [Tue September 14 2010] [4:43:57 pm] warthog9: sorry, we're still mostly just on heartbeat where we use it [Tue September 14 2010] [4:44:56 pm] mmcgrath: yeah, I figured since I'm doing up something new and shiny I should move to what's considered "better" [Tue September 14 2010] [4:45:09 pm] :) [Tue September 14 2010] [4:45:11 pm] mmcgrath: btw, forgot to tell you (or forgot if I told you) I got the yubikey [Tue September 14 2010] [4:45:27 pm] but I'm stuck trying to deal with a file system mount ordering problem, and I ohhhh so love the fact that it gives you *NO* useful debugging output beyond "it's broken" [Tue September 14 2010] [4:45:36 pm] :) [Tue September 14 2010] [4:45:50 pm] Oxf13: good to hear, we'll be staging the yubikey stuff soon, if you want to test it in devel let me know in a bit [Tue September 14 2010] [4:45:55 pm] * mmcgrath has a cranky boy to try to settle down first [Tue September 14 2010] [4:46:14 pm] mmcgrath: tazer? [Tue September 14 2010] [4:46:31 pm] warthog9: that's way cheaper then what I had in mind :) [Tue September 14 2010] [4:47:06 pm] abadger1999:can we for now keep the static icons in the code (inactive) and use varnish to do the job? We will be able to benchamrk with/without it later and decide if it will be removed from codebase [Tue September 14 2010] [4:47:21 pm] mbacovsk: Yes, that's fine. [Tue September 14 2010] [4:49:12 pm] abadger1999: okay. Will there be sa-migrate stuff in next 0.5.x release or 0.6.x is going to be next? [Tue September 14 2010] [4:49:56 pm] mbacovsk: We'll continue to do 0.5.x releases if need be to fix bugs until 0.6.x is done. [Tue September 14 2010] [4:50:54 pm] boy this thing flashes a lot [Tue September 14 2010] [4:51:16 pm] mbacovsk: 0.6.x can have db schema subtractions and such that can't go into 0.5.x [Tue September 14 2010] [4:51:45 pm] mbacovsk: We can put the initial sa-migrate stuff into 0.5.x as long as there's no subtractions to the schema. [Tue September 14 2010] [4:54:42 pm] abadger1999: they are not. Should I try to merge to 0.5.x. then? [Tue September 14 2010] [4:54:54 pm] mbacovsk: Yep, go ahead. [Tue September 14 2010] [4:55:19 pm] mbacovsk: Shouldn't be too hard, the branches are almost identical right now. [Tue September 14 2010] [4:55:37 pm] abadger1999: great [Tue September 14 2010] [4:56:40 pm] * mbacovsk goes to bed [Tue September 14 2010] [4:58:23 pm] Thanks