Time Nick Message
14:43 * cait waves
17:50 magnuse marcelr++ for bug 13438
17:50 huginn 04Bug http://bugs.koha-community.org/bugzilla3/show_bug.cgi?id=13438 normal, P5 - low, ---, m.de.rooy, Needs Signoff , Convert normarc framework plugins to new style (see also 10480)
19:40 bag hola
19:40 wahanui kia ora, bag
20:00 * cait waves
20:01 rangi morning
20:01 cait morning bag and rangi
20:04 tcohen hi
20:06 cait hi
20:06 wahanui hey, cait
20:06 cait :)
20:15 tcohen hi cait
20:15 rangi hi tcohen
20:16 tcohen hey rangi
20:16 rangi i've been experimenting with varnish
20:16 rangi in front of the opac
20:16 tcohen for caching stuff?
20:16 tcohen statc content?
20:16 rangi all of the js/css/images are cachecd
20:17 rangi and so is opac-main.pl if the user is not logged in
20:17 rangi it takes a lot of load off apache/plack
20:18 tcohen you mean it w¡got worse?
20:18 rangi nope
20:18 rangi it means theres a bunch less requests that hit apache
20:18 rangi varnish serves them from ram
20:18 tcohen ah, idiomatic issue
20:18 rangi you need nginx or something in front to do ssl termination
20:19 rangi it's not a lot faster (it is faster but not a lot) than without vagrant
20:19 tcohen rangi: is there still that google module for caching?
20:19 rangi but it scales much higher
20:20 rangi hmm?
20:20 tcohen https://developers.google.com/speed/pagespeed/module/
20:20 rangi basically vagrant respects the headers we put on stuff
20:21 rangi it's a lot less 'magic' than that google stuff
20:21 rangi and hence I actually trust it
20:21 rangi (ie it wont cache stuff it shouldnt)
20:21 tcohen we should set the right headers for static stuff anyway
20:22 rangi we do
20:22 rangi which means vagrant caches it
20:22 rangi because browsers/proxies don't always behave how they should
20:22 tcohen so the main advantage is that varnish caches pages
20:22 rangi but it also means, if I hit it, then you hit it
20:22 rangi varnish serves you the one from the cache, without bothering apache
20:23 rangi yes
20:23 tcohen interesting
20:23 wahanui interesting is sometimes good and sometimes bad
20:23 rangi you can just put it there, and do nothing
20:23 rangi and it works by obeying the headers
20:24 rangi or you can make it do more, it starts very conservative/safe which is a good default
20:24 rangi trying out pagespeed would be good too, but its much less simple to understand :)
20:24 cait hmmm
20:25 rangi caching is like security, it's good in layers
20:29 tcohen hehe
20:29 rangi 3.22.x is quite a bit faster with the caching work done, but if we are serving essentially static pages (like opac-main.pl before you login) we shouldn't even have to touch plack/apache if possible
20:29 tcohen the good thing about pagespeed is that it shrinks images, and that kind of stuff, apart from the cche pthing
20:30 rangi yep, we are doing that with mod_gzip etc already
20:30 rangi but it still would be good to try
20:30 rangi so for opac-main.pl if we did this
20:30 tcohen it does so matching the pages styles
20:31 rangi if user not logged in
20:31 rangi print $data->header(
20:31 rangi -expires => '+30m',
20:32 rangi then vagrant (and some browsers) would respect that, and not ask for the page again
20:32 rangi if you wanted to be really tricky
20:32 rangi can do
20:33 pastebot "rangi" at 127.0.0.1 pasted "caching stuff" (10 lines) at http://paste.koha-community.org/378
20:33 rangi before the script does anything else
20:33 rangi that says did the browser ask me for a last modified before
20:33 rangi if so, and its less that my timeout
20:34 rangi then hand them back a 304 (page not modified)
20:34 rangi and exit
20:34 rangi before it even runs the rest of the script
20:34 jcamins FWIW, I found that using nginx for static files was only about 10% slower than Varnish under relatively low load, but I believe when you start increasing the load to high you'd start seeing performance benefits.
20:34 rangi jcamins: yeah the trick is when you start caching the .pl files output where you can that you start getting the speed benefits too
20:35 rangi otherwise you are right, it mostly helps with scaling
20:37 tcohen rangi: i'll be looking forward to your results
20:37 tcohen i trust nginx caching capabilities, still
20:38 jcamins rangi: yeah, 304 support would be a pretty big win.
20:38 tcohen cover images should support 304
20:38 tcohen (and they don't)
20:40 rangi they do now
20:40 rangi well im testing it with one client
20:40 tcohen rangi: i mean local cover images
20:40 rangi [off] https://library.charteredaccountantsanz.com/whatsnew/#/all/ try that, then do a refresh
20:41 rangi opac-image.pl does the 304 thing there
20:42 tcohen not here!
20:42 tcohen ah, status=200 but chrome says "From cache"
20:44 rangi yep, what does firefox says
20:44 rangi they both behave a bit differently i found
20:46 rangi you can test with curl too
20:46 rangi [off] curl -I --header 'If-Modified-Since: Mon, 04 Apr 2016' https://library.charteredaccountantsanz.com/cgi-bin/koha/opac-image.pl\?biblionumber\=85364\&thumbnail\=1
20:49 rangi i love curl
21:02 bag yeah we spent sometime looking at varnish too
21:04 * bag thinks it’s time for a martini ;) (I’ve got a brisket on the smoker since 7am (7 hours ago) ) should be ready in another 3~4 hours :)
21:05 cait hm
21:05 cait did someone try html5media by chance?
21:05 cait i have problems to get it show up
21:06 rangi not recently
21:07 * cait tries some old sample records from the original bug
21:15 * tcohen feels like serving a chilled jaeger
21:17 * tcohen joins bag
21:18 * cait has a ginger ale
21:19 tcohen rangi: so the 304 for opac-image.pl is varnish's responsability
21:19 tcohen ?
21:19 rangi nope
21:19 rangi that code i pasted before
21:19 tcohen ah
21:19 rangi http://paste.koha-community.org/378
21:19 rangi varnish will of course respect that too
21:20 rangi you have to make sure that it is setting a last modified too
21:21 rangi -Last_Modified => strftime('%a, %d %b %Y %H:%M:%S GMT', gmtime),
21:21 rangi so that if it 200s it sets a last modified, so the browser can ask for that
21:21 rangi make sense?
21:25 tcohen it does
21:26 tcohen i'm intrigued because I see a lot of time is spent on local images
21:26 rangi cool
21:27 rangi it still hits the perl, so has to do overhead of loading modules etc (if cgi) but doesn't have to fetch the image from the db, and return it etc
21:28 rangi so if you are running cgi, its not going to be much faster for a single request
21:28 rangi but if someone refreshes the page, thats 20 hits .. that will all be a little faster
21:29 rangi and if you have something like varnish caching those images .. for a day or 2
21:29 rangi then thats a massive improvement
21:29 rangi once one person has seen them, they are cached for everyone
21:30 rangi (until they login that is)
21:30 rangi altho, probably could ignore the cookie on images .. and even return them for logged in users
21:30 rangi something like
21:32 rangi if (req.http.Content-Type ~ "image") {
21:32 rangi unset.req.http.cookie;
21:32 rangi }
21:32 rangi (in the varnish config)
21:33 rangi then varnish should cache it
21:33 rangi ill try it after work tonight
21:39 tcohen maybe we can teach apache/nginx some stuff are static content
21:40 rangi hmm only the stuff that actually is :)
21:41 rangi but yes it makes no sense to have plack hand out any static stuff
21:58 tcohen rangi: interesting that no local-images == high latency due to the check
21:59 rangi yep
22:04 tcohen i bet checking local image existence in opac-search.pl and putting the URL on the template, would save at least a couple seconds to the end user
22:06 rangi hmm?
22:07 rangi maybe
22:07 rangi or make the link be client side via js
22:08 tcohen it is
22:08 rangi then it wont slow the page load down
22:08 rangi its a static link right? not an ajax one?
22:08 rangi (you could make it fall back to static if people had js turned off too)
22:09 rangi then it wont slow the page down
22:10 rangi can do both of course but then opac-search.pl itself would be slower
22:12 rangi do we link to the image even if local images are turned off?
22:12 rangi cos doing that check in the template would be pretty easy eh?
22:13 rangi i suspect that most people run with localimages off
22:14 rangi yeah about 60% have the pref off, so checking that before putting the link in the template (if we dont alredy) would save 60% of the libraries a bunch of time
22:14 rangi (60% of hea that is)
22:15 tcohen yeah, that's why i put my libraries there
22:15 tcohen :-d
22:15 tcohen HEA++
22:15 tcohen rangi, i leave a question for you
22:15 tcohen how is plack related to files in /tmp ?
22:15 tcohen it seems that some permission issues i've been sufffering running plack+3.22 in production are related to that
22:16 rangi its not
22:16 rangi but the fastmmap stuff is
22:16 rangi (oh and sessions if you have sessions in tmp not mysql or memcache)
22:17 rangi i have a site running plack both staff and opac and there is nothing in /tmp
22:17 rangi https://obs.library.workbuffer.org/
22:18 rangi running 3.22
23:13 talljoy good day
23:13 talljoy dani how goes round rock?
23:14 talljoy dang it wrong channel again. lol
23:14 talljoy good day #koha
23:32 cait hm'