Displaying #traffic-server/2015-10-27.log:

Tue Oct 27 01:32:38 2015  _klk_:Joined the channel
Tue Oct 27 01:52:05 2015  _klk_:Joined the channel
Tue Oct 27 02:13:09 2015  steveq:Joined the channel
Tue Oct 27 02:17:02 2015  gancho:Joined the channel
Tue Oct 27 02:25:19 2015  leprechau:Joined the channel
Tue Oct 27 02:31:13 2015  leprechau:Joined the channel
Tue Oct 27 02:36:18 2015  msekimura:Joined the channel
Tue Oct 27 02:57:16 2015  leprechau:Joined the channel
Tue Oct 27 03:33:49 2015  msekimura:Joined the channel
Tue Oct 27 03:51:24 2015  msekimura:Joined the channel
Tue Oct 27 04:03:41 2015  gancho:Joined the channel
Tue Oct 27 05:36:51 2015  boargod2:Joined the channel
Tue Oct 27 07:07:35 2015  bcall:Joined the channel
Tue Oct 27 07:09:05 2015  jpeach:Joined the channel
Tue Oct 27 07:15:10 2015  dcarlin:Joined the channel
Tue Oct 27 07:19:44 2015  briang:Joined the channel
Tue Oct 27 07:19:44 2015  briang:Joined the channel
Tue Oct 27 08:12:43 2015  mturk:Joined the channel
Tue Oct 27 08:12:43 2015  mturk:Joined the channel
Tue Oct 27 08:57:42 2015  Lethalman:Joined the channel
Tue Oct 27 11:13:00 2015  JSeymour:Joined the channel
Tue Oct 27 12:55:38 2015  gancho:Joined the channel
Tue Oct 27 13:07:16 2015  gancho:Joined the channel
Tue Oct 27 13:09:35 2015  shinrich1:Joined the channel
Tue Oct 27 13:10:21 2015  bahumbug:Joined the channel
Tue Oct 27 13:24:30 2015  es:Joined the channel
Tue Oct 27 13:27:41 2015  zwoop:Joined the channel
Tue Oct 27 13:42:01 2015  davet_:Joined the channel
Tue Oct 27 13:59:40 2015  esproul:Joined the channel
Tue Oct 27 14:01:16 2015  sudheerv:Joined the channel
Tue Oct 27 14:06:28 2015  _iwc:Joined the channel
Tue Oct 27 14:22:43 2015  amc:Joined the channel
Tue Oct 27 14:39:13 2015  zwoop:sudheerv shinrich1 bcall https://tools.ietf.org/html/draft-ruellan-http-accept-push-policy-00
Tue Oct 27 14:39:19 2015  zwoop:in case you guys are still looking at PUSH stuff ?
Tue Oct 27 14:40:12 2015  zwoop:I read some benchmarks / study, on the effectiveness on PUSH on a normal web site, it has little to no performance improvements, and requires some major requirements on both client and server side to not be worse (such as, sending a bloom filter like hash from client to server to indicate what items the client already have in cache)
Tue Oct 27 14:42:30 2015  sudheerv:Gah
Tue Oct 27 14:42:36 2015  sudheerv:zwoop: very interesting..
Tue Oct 27 14:42:49 2015  zwoop:trying to find that article, sec
Tue Oct 27 14:42:59 2015  zwoop:http://blog.kazuhooku.com/2015/10/performance-of-http2-push-and-server.html
Tue Oct 27 14:43:03 2015  zwoop:I can't say I'm surprised
Tue Oct 27 14:43:25 2015  zwoop:PUSH makes sense as a replacement for e.g. Comet and other server -> client side push scenarios, not for web page
Tue Oct 27 14:44:01 2015  sudheerv:i guess, there's one little catch here though
Tue Oct 27 14:44:22 2015  sudheerv:if the assets on your page are not on the same domain/ip
Tue Oct 27 14:44:38 2015  sudheerv:browser would need to establish new connections to those domain/ip to get the resources
Tue Oct 27 14:44:44 2015  sudheerv:PUSH can circumvent that..
Tue Oct 27 14:44:50 2015  sudheerv:or rather optimize that
Tue Oct 27 14:44:57 2015  sudheerv:zwoop: sort of like ycpi/ycs scenario
Tue Oct 27 14:45:14 2015  sudheerv:unsure though whether it's a generic problem for everyone or just us
Tue Oct 27 14:45:24 2015  zwoop:yeah, but with H2, you should just stop doing that :)
Tue Oct 27 14:45:28 2015  sudheerv::)
Tue Oct 27 14:45:29 2015  zwoop:i.e. no more domain sharding
Tue Oct 27 14:45:45 2015  zwoop:pretty sure Iliya Grigorik says that (domain sharding is an anti-pattern in H2)
Tue Oct 27 14:47:33 2015  zwoop:sudheerv https://docs.google.com/presentation/d/1r7QXGYOLCh4fcUq0jDdDwKJWNqWK1o4xMtYpKZCJYjM/present?slide=id.p19
Tue Oct 27 14:47:59 2015  sudheerv:zwoop: agree, no domain sharding with spdy/h2
Tue Oct 27 14:48:16 2015  sudheerv:the issue i'm referring to is slightly different to domain sharding
Tue Oct 27 14:48:28 2015  sudheerv:it's on purpose keeping certain assets isolated
Tue Oct 27 14:48:55 2015  sudheerv:but, i agree that it may not be a generic problem for everyone to worry/think about :)
Tue Oct 27 14:49:40 2015  zwoop:right
Tue Oct 27 14:49:48 2015  PSUdaemon:PUSH and streams
Tue Oct 27 14:49:59 2015  zwoop:you'd have an even worse problem though, I think
Tue Oct 27 14:50:05 2015  sudheerv:lol
Tue Oct 27 14:50:26 2015  sudheerv:PSUdaemon: why are you not wearing a cap today :)
Tue Oct 27 14:50:29 2015  zwoop:because if you don't communicate both the e.g. s.yimg.com and news.yahoo.com assets from client's cache to both YCS and YCPI, you would PUSH a lot more than needed, making things worse, not better.
Tue Oct 27 14:50:44 2015  sudheerv:yeah, it's not an easy problem for sure :)
Tue Oct 27 14:51:17 2015  PSUdaemon:sudheerv: ?
Tue Oct 27 14:52:29 2015  sudheerv:PSUdaemon: I tried to find your monicker and didnt' find it color coded :)
Tue Oct 27 14:53:56 2015  PSUdaemon:oh
Tue Oct 27 14:54:00 2015  PSUdaemon:my @ thing
Tue Oct 27 14:54:08 2015  PSUdaemon:Joined the channel
Tue Oct 27 14:54:13 2015  sudheerv:lol
Tue Oct 27 14:54:14 2015  sudheerv:yeah
Tue Oct 27 14:54:26 2015  PSUdaemon:so PUSH is good for CSS and stuff, right?
Tue Oct 27 14:54:30 2015  PSUdaemon:and JS
Tue Oct 27 14:54:44 2015  PSUdaemon:instead of inlining it all
Tue Oct 27 14:54:49 2015  PSUdaemon:so it can be cached better
Tue Oct 27 14:55:31 2015  sudheerv:PSUdaemon: yeah, i suppose the benchmarks that zwoop sent say that the gains are not significant enough
Tue Oct 27 14:55:58 2015  zwoop:PSUdaemon the issue though is, it's really difficult to know what is worthwhile PUSHing or not
Tue Oct 27 14:56:32 2015  zwoop:PSUdaemon like, how do you know if the upstream UA or server already has it or not? By the time that client gets a chance to close down the stream, you might have sent many KB's of data, for no good.
Tue Oct 27 14:58:09 2015  zwoop:https://tools.ietf.org/html/draft-ruellan-http-accept-push-policy-00 section 3.3.2 seems to try to address that
Tue Oct 27 15:18:47 2015  Amaryllis:with remap_required=0 and a parent defined for dest_domain=. in parent.config, what happens to CONNECT requests?
Tue Oct 27 15:42:17 2015  ibrezac:Joined the channel
Tue Oct 27 15:56:37 2015  blattj:Joined the channel
Tue Oct 27 16:04:01 2015  reveller1:Joined the channel
Tue Oct 27 16:06:27 2015  reveller2:Joined the channel
Tue Oct 27 16:07:19 2015  ibrezac:proxy.pac support is removed in 6 (and maybe earlier), is there an alternate way to server the proxy.pac directly from ats?
Tue Oct 27 16:08:07 2015  Amaryllis:ibrezac: from what i remember of that discussion, no, the consensus is you should run a web server on localhost to serve it
Tue Oct 27 16:08:14 2015  jpeach:ibrezac: I think you could use the healthcheck plugin to serve it
Tue Oct 27 16:11:44 2015  ibrezac:jpeach: does one such plugin exist?
Tue Oct 27 16:13:38 2015  jpeach:ibrezac: https://github.com/apache/trafficserver/tree/master/plugins/experimental/healthchecks
Tue Oct 27 16:13:54 2015  jpeach:zwoop: why is healthchecks still experimental? and undocumented?
Tue Oct 27 16:15:07 2015  ibrezac:Amaryllis: thanks. I understand, although this is not convenient for us.
Tue Oct 27 16:16:14 2015  Amaryllis:is healthchecks now? i don't remember seeing that before
Tue Oct 27 16:16:54 2015  jpeach:yeh healthchecks can serve a proxy.pac
Tue Oct 27 16:18:27 2015  msekimura:Joined the channel
Tue Oct 27 16:18:44 2015  ibrezac:thanks jpeach, I see it
Tue Oct 27 16:18:54 2015  ibrezac:as long is it is less than 16k :)
Tue Oct 27 16:20:02 2015  jpeach:you might needs to minify the JS ;)
Tue Oct 27 16:20:19 2015  ibrezac:not an issue atm :)
Tue Oct 27 16:30:11 2015  reveller2:zwoop: was healthchecks another Koop plugin? Would you like me do a PR to move it out of experimentaland put together a doc page for it?
Tue Oct 27 16:35:47 2015  bahumbug:Joined the channel
Tue Oct 27 16:40:23 2015  psp1:Joined the channel
Tue Oct 27 16:40:31 2015  jpeach:reveller: that would be great .. it is widely used in production
Tue Oct 27 16:42:37 2015  _klk_:Joined the channel
Tue Oct 27 16:43:37 2015  ibrezac:jpeach: how do people typically use this plugin? Some process that manages content/existance of these files?
Tue Oct 27 16:44:37 2015  jpeach:typically you have some global traffic management system that directs traffic, so you whack the healthcheck file to drain traffic or release it
Tue Oct 27 16:45:42 2015  gancho:Joined the channel
Tue Oct 27 16:57:18 2015  blattj:Joined the channel
Tue Oct 27 17:05:35 2015  ibrezac:jpeach: I see, that's assuming you have something in front of ats
Tue Oct 27 17:07:12 2015  jpeach:not necessarily ... if you look at a system like Traffic Control, then the healtcheck plugin is used to feed state into the global traffic management system, same mechanics for when you move traffic around using DNS
Tue Oct 27 17:09:25 2015  Amaryllis:hmm, we did this in lua, but maybe we should move to healthcheck
Tue Oct 27 17:10:34 2015  jpeach:if you have something working, there's probably not much to gain by switching
Tue Oct 27 17:17:20 2015  Amaryllis:true, although i like the idea of being able to remove the server from rotation by deleting the healthcheck file, instead of having to stop ATS or reconfigure the LB
Tue Oct 27 17:21:26 2015  zwoop:reveller no, I wrote that
Tue Oct 27 17:22:26 2015  jpeach:Amaryllis: lua code should be reloadable, right?
Tue Oct 27 17:22:57 2015  Amaryllis:jpeach: the lua script is managed by puppet, so it can't be edited locally
Tue Oct 27 17:23:11 2015  Amaryllis:i could have lua check for a file, but if i'm doing that anyway, it seems easiest to just switch to healthcheck
Tue Oct 27 17:23:17 2015  jpeach:ah
Tue Oct 27 18:03:09 2015  blattj:Joined the channel
Tue Oct 27 18:07:46 2015  blattj1:Joined the channel
Tue Oct 27 18:08:28 2015  blattj:Joined the channel
Tue Oct 27 18:44:04 2015  blattj:Joined the channel
Tue Oct 27 19:22:22 2015  jpeach:has anyone used the teamd bonding driver http://libteam.org ?
Tue Oct 27 20:42:56 2015  bahumbug:Joined the channel
Tue Oct 27 20:42:56 2015  bahumbug:Joined the channel
Tue Oct 27 20:44:48 2015  davet:Joined the channel
Tue Oct 27 20:48:50 2015  davet_:Joined the channel
Tue Oct 27 21:03:28 2015  psp:Joined the channel
Tue Oct 27 21:05:55 2015  blattj:Joined the channel
Tue Oct 27 22:05:20 2015  blattj:Joined the channel
Tue Oct 27 22:58:09 2015  blattj:Joined the channel
Tue Oct 27 22:59:12 2015  blattj:Joined the channel
Tue Oct 27 23:18:15 2015  Amaryllis:hm... with remap_required=0, requests to https://domain try to establish an https connection to the parent?
Tue Oct 27 23:23:23 2015  Amaryllis:ah, "regex_map https://(.*) http://$1" seems to fix that

Comments