Displaying #traffic-server/2015-11-24.log:

Tue Nov 24 00:09:11 2015  bcall:amc: still all the ubuntu builds are broke
Tue Nov 24 00:09:27 2015  bcall:from:
Tue Nov 24 00:09:28 2015  bcall:../../proxy/InkAPITest.cc:334:48: error: ignoring return value of ‘ssize_t write(int, const void*, size_t)’, declared with attribute warn_unused_result [-Werror=unused-result]
Tue Nov 24 00:09:28 2015  bcall: ::write(vc->con.fd, "Bob's your uncle", 16);
Tue Nov 24 00:46:08 2015  jpeach:bcall: you need efc597969f6c0a55198a53e0f66d4cbe7b82d18b
Tue Nov 24 00:58:22 2015  _klk_:Joined the channel
Tue Nov 24 04:11:23 2015  es:Joined the channel
Tue Nov 24 04:22:59 2015  briang:jpeach, why isn't that code if (::write(...) < 0) { assert(!"write failed"); } or something?
Tue Nov 24 04:23:48 2015  briang:I know it's a test but still..any idea why they ignored it?
Tue Nov 24 04:25:42 2015  jpeach:ink_release_assert(::write(..) == nbytes) ?
Tue Nov 24 04:31:18 2015  briang:+1
Tue Nov 24 04:31:23 2015  briang:that makes so much more sense
Tue Nov 24 04:59:26 2015  dxu_:Joined the channel
Tue Nov 24 05:04:27 2015  dxu_:Joined the channel
Tue Nov 24 05:09:32 2015  dxu_:Joined the channel
Tue Nov 24 05:14:32 2015  dxu_:Joined the channel
Tue Nov 24 05:19:33 2015  dxu_:Joined the channel
Tue Nov 24 05:24:30 2015  dxu_:Joined the channel
Tue Nov 24 05:29:33 2015  dxu_:Joined the channel
Tue Nov 24 05:34:35 2015  dxu_:Joined the channel
Tue Nov 24 05:37:43 2015  gancho:Joined the channel
Tue Nov 24 05:44:05 2015  _klk_:Joined the channel
Tue Nov 24 05:44:28 2015  dxu__:Joined the channel
Tue Nov 24 06:42:07 2015  gancho:Joined the channel
Tue Nov 24 06:55:41 2015  _klk_:Joined the channel
Tue Nov 24 07:26:28 2015  dxu_:Joined the channel
Tue Nov 24 07:50:00 2015  dxu_:Joined the channel
Tue Nov 24 09:12:41 2015  Lethalman:Joined the channel
Tue Nov 24 10:11:08 2015  gotadachi:Joined the channel
Tue Nov 24 11:35:19 2015  niq:Joined the channel
Tue Nov 24 11:41:29 2015  bahumbug:Joined the channel
Tue Nov 24 11:51:27 2015  Zerpex_:Joined the channel
Tue Nov 24 11:55:51 2015  Zerpex_:Joined the channel
Tue Nov 24 12:25:23 2015  Lethalman:hi
Tue Nov 24 12:25:30 2015  Lethalman:how can I tell in a remap rule to never cache something?
Tue Nov 24 12:25:51 2015  Lethalman:I'm only thinking of cache_promote with 0 chance
Tue Nov 24 12:25:53 2015  Lethalman:is that the only way?
Tue Nov 24 12:26:10 2015  Lethalman:that is, url/ -> never cache, url/foo normal cache
Tue Nov 24 12:29:03 2015  felicity:Lethalman: use conf_remap.so and disable http.cache
Tue Nov 24 12:29:24 2015  Lethalman:felicity, aaaaaah
Tue Nov 24 12:29:32 2015  Lethalman:thanks! of course
Tue Nov 24 12:34:48 2015  Zerpex:Joined the channel
Tue Nov 24 12:41:17 2015  JSeymour:Joined the channel
Tue Nov 24 12:53:49 2015  Zerpex:Joined the channel
Tue Nov 24 12:54:23 2015  Zerpex:Joined the channel
Tue Nov 24 12:58:24 2015  Lethalman:mh I'm using TS 6.0, when I restart ats the disk cache is not loading entries
Tue Nov 24 12:58:27 2015  Lethalman:I mean, the disk cache is working
Tue Nov 24 12:58:47 2015  Lethalman:requests work, they get cached
Tue Nov 24 12:58:56 2015  Lethalman:traffic_top shows the number of entries
Tue Nov 24 12:59:20 2015  Lethalman:but when it gets restarted, the cache is like wiped out
Tue Nov 24 12:59:43 2015  Lethalman:0 entries, everything is a miss
Tue Nov 24 13:01:26 2015  Lethalman:is that normal?
Tue Nov 24 13:10:56 2015  dxu_:Joined the channel
Tue Nov 24 13:28:50 2015  bahumbug:Joined the channel
Tue Nov 24 13:42:38 2015  blattj:Joined the channel
Tue Nov 24 14:01:38 2015  shinrich1:Joined the channel
Tue Nov 24 14:01:58 2015  ibrezac:Joined the channel
Tue Nov 24 14:18:05 2015  Lethalman:cannot understand what's going on
Tue Nov 24 14:21:39 2015  blattj:Joined the channel
Tue Nov 24 14:26:57 2015  reveller:Joined the channel
Tue Nov 24 14:32:20 2015  davet_:Joined the channel
Tue Nov 24 14:34:11 2015  tomr_frnd:Joined the channel
Tue Nov 24 15:09:31 2015  sudheerv:Lethalman: felicity: incidentally, there was an idea to deprecate http.cache at the ATS summit
Tue Nov 24 15:09:47 2015  Lethalman:sudheerv, :S
Tue Nov 24 15:09:53 2015  sudheerv:(fwiw, the consensus was to come up with some sort of alternative, though)
Tue Nov 24 15:09:56 2015  Lethalman:sudheerv, looks like I have quite a use case :P
Tue Nov 24 15:10:05 2015  sudheerv:(and this will all be for 7.0)
Tue Nov 24 15:10:11 2015  sudheerv:Lethalman: yeah, we do too
Tue Nov 24 15:10:21 2015  Lethalman:though it's very easy with a plugin
Tue Nov 24 15:10:24 2015  sudheerv:so, i'm pretty sure, there will be an alternative for sure
Tue Nov 24 15:10:28 2015  Lethalman:sudheerv, is that hard to maintain in the code?
Tue Nov 24 15:10:40 2015  esproul:Joined the channel
Tue Nov 24 15:10:45 2015  Lethalman:sudheerv, the experimental plugin should do the right job I guess
Tue Nov 24 15:11:06 2015  sudheerv:Lethalman: actually, there's some ugly dependency between the multiple cache settings
Tue Nov 24 15:11:08 2015  sudheerv:and API
Tue Nov 24 15:11:15 2015  sudheerv:i think this was an effort to do some clean up
Tue Nov 24 15:11:18 2015  Lethalman:ah ok
Tue Nov 24 15:11:39 2015  sudheerv:https://cwiki.apache.org/confluence/display/TS/Presentations+-+2015?preview=/56066455/61329168/Cacheability%20Control.pptx#Presentations-2015-2015FallSummit
Tue Nov 24 15:11:47 2015  sudheerv:Lethalman: that presentation talks about it
Tue Nov 24 15:12:25 2015  Lethalman:thanks
Tue Nov 24 15:12:49 2015  sudheerv:np
Tue Nov 24 15:43:07 2015  blattj:Joined the channel
Tue Nov 24 15:48:48 2015  gancho:Joined the channel
Tue Nov 24 15:49:43 2015  dxu_:Joined the channel
Tue Nov 24 15:53:15 2015  jpeach:hey kichan? ayt?
Tue Nov 24 16:26:34 2015  shinrich2:Joined the channel
Tue Nov 24 16:41:22 2015  felicity:has anyone seen crashes in stale-while-revalidate in 6.0? https://dpaste.de/FJoA
Tue Nov 24 16:43:43 2015  jpeach:TS-4009
Tue Nov 24 16:43:57 2015  jpeach:looks the same to me
Tue Nov 24 16:44:01 2015  jpeach:PSUdaemon: ^^^
Tue Nov 24 16:46:02 2015  felicity:thanks
Tue Nov 24 16:57:13 2015  PSUdaemon:jpeach: we should deprecate that plugin
Tue Nov 24 16:57:20 2015  PSUdaemon:sudheerv said he is going to put it in the core :)
Tue Nov 24 16:57:43 2015  sudheerv:yes, that's the plan :)
Tue Nov 24 17:02:09 2015  es:Joined the channel
Tue Nov 24 17:02:34 2015  bcall:jpeach: I have efc597969f6c0a55198a53e0f66d4cbe7b82d18b
Tue Nov 24 17:02:45 2015  jpeach:ack
Tue Nov 24 17:02:50 2015  bcall:jpeach: it is broke on jenkins
Tue Nov 24 17:02:58 2015  jpeach:bcall: wait, and it still fails to build?
Tue Nov 24 17:03:06 2015  bcall:https://ci.trafficserver.apache.org
Tue Nov 24 17:03:08 2015  bcall:yep
Tue Nov 24 17:03:17 2015  jpeach:I thought that was the commit amc landed to fix that problem?
Tue Nov 24 17:03:18 2015  bcall:(void) is not good enough
Tue Nov 24 17:03:25 2015  felicity:PSUdaemon: i'm sure people using it would appreciate it being fixed until it's implemented in core :)
Tue Nov 24 17:03:51 2015  jpeach:bcall: I guess briang's ink_release_assert then?
Tue Nov 24 17:04:16 2015  bcall:or maybe the code checkout is not being updated
Tue Nov 24 17:04:23 2015  PSUdaemon:felicity: sure. i don't have time to look at it this weekend though. is it verified working in 5.3?
Tue Nov 24 17:04:23 2015  es1:Joined the channel
Tue Nov 24 17:04:43 2015  felicity:we didn't use it in 5.3, but i remember one other person in here complaining it didn't work since they upgraded
Tue Nov 24 17:05:29 2015  amc:Yeah, looking at the console output, it doesn't have the (void) in it.
Tue Nov 24 17:06:22 2015  bcall:what project does the code get checked out on?
Tue Nov 24 17:07:01 2015  amc:Well it's odd though that the ::write() shows up, but two commits later the (void) doesn't.
Tue Nov 24 17:08:45 2015  bahumbug:Joined the channel
Tue Nov 24 17:09:05 2015  mturk:Joined the channel
Tue Nov 24 17:09:11 2015  bcall:there is a build going now - I will take a look after it is done
Tue Nov 24 17:09:58 2015  dxu_:Joined the channel
Tue Nov 24 17:22:59 2015  slx86:Joined the channel
Tue Nov 24 17:26:34 2015  biilmann:Joined the channel
Tue Nov 24 17:32:37 2015  blattj1:Joined the channel
Tue Nov 24 17:41:28 2015  kichan:jpeach: looking for me?
Tue Nov 24 17:55:14 2015  _klk_:Joined the channel
Tue Nov 24 18:02:47 2015  biilmann:Joined the channel
Tue Nov 24 18:07:43 2015  biilmann:Joined the channel
Tue Nov 24 18:11:41 2015  jpeach:kichan: I think i figured it out
Tue Nov 24 18:13:08 2015  jpeach:working on lua bindings to replace stats.config.xml and needed to construct a tree where nodes can be either tables or userdata
Tue Nov 24 18:25:22 2015  bfrance:Joined the channel
Tue Nov 24 18:27:37 2015  blattj:Joined the channel
Tue Nov 24 19:01:21 2015  dxu_:Joined the channel
Tue Nov 24 19:43:58 2015  es:Joined the channel
Tue Nov 24 19:48:08 2015  bcall:amc: it failed on the (void) also:
Tue Nov 24 19:48:09 2015  bcall:../../proxy/InkAPITest.cc:334:54: error: ignoring return value of ‘ssize_t write(int, const void*, size_t)’, declared with attribute warn_unused_result [-Werror=unused-result]
Tue Nov 24 19:48:09 2015  bcall: (void)::write(vc->con.fd, "Bob's your uncle", 16);
Tue Nov 24 19:48:19 2015  bcall:I updated the code to do ink_assert()
Tue Nov 24 19:49:32 2015  bcall:actually ink_release_assert
Tue Nov 24 19:52:40 2015  biilmann:Joined the channel
Tue Nov 24 19:56:39 2015  PSUdaemon:biilmann: you using influx?
Tue Nov 24 20:04:31 2015  amc:That's really annoying.
Tue Nov 24 20:10:41 2015  PSUdaemon:amc: sorry...
Tue Nov 24 20:10:45 2015  PSUdaemon:i'll nver do it again
Tue Nov 24 20:13:08 2015  amc:Yeah, I bet you say that to all the open source communities.
Tue Nov 24 20:14:17 2015  amc:I think it's a bit much for the compiler to complain that the code ignores a value cast to (void) as that's the primary point of doing that.
Tue Nov 24 20:14:44 2015  amc:I also don't see why it's only Ubuntu where this happens. I would think it's using the same gcc as other operating systems.
Tue Nov 24 20:28:21 2015  biilmann:Joined the channel
Tue Nov 24 20:29:20 2015  jrushford:Joined the channel
Tue Nov 24 20:33:21 2015  biilmann:Joined the channel
Tue Nov 24 20:38:54 2015  _klk_:Joined the channel
Tue Nov 24 21:03:14 2015  biilmann:Joined the channel
Tue Nov 24 21:06:10 2015  dxu_:Joined the channel
Tue Nov 24 21:59:09 2015  ushachar:Joined the channel
Tue Nov 24 21:59:09 2015  ushachar:Joined the channel
Tue Nov 24 22:20:39 2015  blattj:Joined the channel
Tue Nov 24 22:25:34 2015  jacksontj:is there a way to log whether the request was SPDY/HTTP2 ?
Tue Nov 24 22:25:46 2015  jacksontj:internally we make them all http/1.1, but it would be nice to log them
Tue Nov 24 22:26:57 2015  shinrich2:Yes, es added that recently.
Tue Nov 24 22:27:38 2015  jacksontj:k, but not in 5.2 then :(
Tue Nov 24 22:27:54 2015  shinrich2:No, you'd have to cherry pick it back.
Tue Nov 24 22:28:30 2015  jacksontj:another question for you, in HTTP/1.1 I have a rule in HRR that adds connection close on response headers if I want to take the box OOR
Tue Nov 24 22:28:44 2015  jacksontj:in SPDY/HTTP2 there is no such thing-- the equivalent is goaway frames
Tue Nov 24 22:28:56 2015  jacksontj:any idea how I'd get the same behavior?
Tue Nov 24 22:31:20 2015  shinrich2:Hmm. I don't know off hand. bcall would probably know.
Tue Nov 24 22:31:38 2015  jacksontj:i imagine I'll need to add something to HRR
Tue Nov 24 22:31:39 2015  jacksontj::/
Tue Nov 24 22:32:02 2015  jacksontj:which is fine, I just don't know how to determine if the requsest was spdy/h2 and then how to send goaways ;)
Tue Nov 24 22:33:44 2015  es:jacksontj: you can log the %<pitag> to see that
Tue Nov 24 22:33:57 2015  es:and that might be in 5.2. let me check
Tue Nov 24 22:34:10 2015  jacksontj:pitag is there
Tue Nov 24 22:34:16 2015  jacksontj:so it'll say spdy or something?
Tue Nov 24 22:34:43 2015  sudheerv:you may set proxy.config.spdy.no_activity_timeout_in to a very low value (1)
Tue Nov 24 22:34:55 2015  jacksontj:well, the problem is they are active ;)
Tue Nov 24 22:34:58 2015  sudheerv:that's the closest you can do to no keep-alive behavior with spdy
Tue Nov 24 22:35:03 2015  es:jacksontj: in 5.2 you'll get the spdy version
Tue Nov 24 22:35:06 2015  es:for pitag
Tue Nov 24 22:35:10 2015  jacksontj:3?
Tue Nov 24 22:35:10 2015  es:nothing for H2
Tue Nov 24 22:35:11 2015  jacksontj:or SPDY3
Tue Nov 24 22:35:17 2015  es:spdy3 or spdy3.1
Tue Nov 24 22:35:21 2015  jacksontj:cool
Tue Nov 24 22:35:33 2015  sudheerv:setting proxy.config.spdy.no_activity_timeout_in to 1 will slowly drain the connections
Tue Nov 24 22:35:36 2015  sudheerv:what do you mean they are active?
Tue Nov 24 22:35:37 2015  jacksontj:yea, so I just need some way in plugin land to say "ats send a goaway frame"
Tue Nov 24 22:35:44 2015  sudheerv:even with connection close, it's the same effect
Tue Nov 24 22:35:50 2015  jacksontj:well, kinda
Tue Nov 24 22:35:57 2015  jacksontj:with connection close you say "please close this and make another"
Tue Nov 24 22:35:58 2015  sudheerv:no, it's the same
Tue Nov 24 22:36:03 2015  jacksontj:in SPDY conneciton headers aren't propogated
Tue Nov 24 22:36:06 2015  sudheerv:with proxy.config.spdy.no_activity_timeout_in=1 you are doing something similar
Tue Nov 24 22:36:12 2015  sudheerv:but, you are allowing 1 sec window
Tue Nov 24 22:36:13 2015  jacksontj:GOAWAY is the same, i just have no way of doing that right now
Tue Nov 24 22:36:19 2015  sudheerv:set proxy.config.spdy.no_activity_timeout_in=1
Tue Nov 24 22:36:27 2015  sudheerv:that will send a goaway if the client doesn't send a new req within 1 sec
Tue Nov 24 22:36:30 2015  jacksontj:well, setting the no_activity timeout low like that would make the bleed case work-- but i'd have to change it to take the box oor
Tue Nov 24 22:36:36 2015  jacksontj:i guess I could do that-- traffic_line it
Tue Nov 24 22:36:40 2015  sudheerv:yes
Tue Nov 24 22:36:43 2015  es:jacksontj: http2 pitag support is in 6.0 and newer
Tue Nov 24 22:37:14 2015  sudheerv:i wish we allowed a special value to basically disable no-activity timeout
Tue Nov 24 22:37:23 2015  sudheerv:that could be useful for use cases like this
Tue Nov 24 22:38:18 2015  sudheerv:i have a few scenarios too where i need such an instant no-activity timeout
Tue Nov 24 22:38:23 2015  sudheerv:i think i;ll do that :)
Tue Nov 24 22:39:31 2015  jpeach:fwiw when we first added spdy we had an object called protostack that tracked the protocol stack of txn
Tue Nov 24 22:40:05 2015  sudheerv:i think for this specific problem, knowing whether or not spdy is not the primary issue
Tue Nov 24 22:40:09 2015  sudheerv:sending a goaway is
Tue Nov 24 22:40:23 2015  jacksontj:yea, I need some mechanism in the plugin to send that goaway
Tue Nov 24 22:40:29 2015  jacksontj:in http/1.1 its connection close
Tue Nov 24 22:40:35 2015  jacksontj:in spdy/h2 its different-- and not exposed
Tue Nov 24 22:40:45 2015  sudheerv:afaik, there's no such mechanism right now
Tue Nov 24 22:40:48 2015  jacksontj:I'm not sure we want to say all connection close headers == goaway
Tue Nov 24 22:40:55 2015  sudheerv:the closest that is possible is the no-activiity timeout
Tue Nov 24 22:41:00 2015  jacksontj:that seems really heavy handed-- so it seems like there is a missing API ;)
Tue Nov 24 22:41:24 2015  sudheerv:yeah, i agree it could be useful
Tue Nov 24 22:42:01 2015  sudheerv:btw, goaway is a spdy/h2 level control frame
Tue Nov 24 22:42:10 2015  jacksontj:correct
Tue Nov 24 22:42:13 2015  sudheerv:it is not the same as Connection header. which is http
Tue Nov 24 22:42:22 2015  sudheerv:in fact, spdy/h2 don't allow using Connection header, i think
Tue Nov 24 22:42:31 2015  jacksontj:i realize, for the particular use case I have-- they are the ways in the 2 protocols to achieve the goal I have
Tue Nov 24 22:42:40 2015  jacksontj:correct, connection headers don't exist in spdy/h2
Tue Nov 24 22:43:10 2015  sudheerv:btw, setting the timer is not really heavy-handed
Tue Nov 24 22:43:17 2015  sudheerv:you can override the timer with a plugin as well
Tue Nov 24 22:43:30 2015  sudheerv:however, it's still not equivalent of connection:close
Tue Nov 24 22:43:33 2015  jacksontj:oh, interesting idea...
Tue Nov 24 22:43:34 2015  sudheerv:since, that is almost instant
Tue Nov 24 22:43:38 2015  jacksontj:i could make my HRR set that value
Tue Nov 24 22:43:39 2015  jacksontj:Hmm
Tue Nov 24 22:43:42 2015  sudheerv:while, the timeout still leaves a 1 sec window
Tue Nov 24 22:43:52 2015  sudheerv:so, if a client makes another req within the 1 sec, it keeps the connection
Tue Nov 24 22:43:55 2015  jacksontj:assuming that options is overrideable
Tue Nov 24 22:44:02 2015  sudheerv:(it should be)
Tue Nov 24 22:44:21 2015  sudheerv:if you had a really busy connection though, it'd remain active for a while
Tue Nov 24 22:44:21 2015  jacksontj:yea, looks like it
Tue Nov 24 22:44:25 2015  jacksontj:thats okay
Tue Nov 24 22:44:31 2015  jacksontj:let me mess with this as a HRR rule :)
Tue Nov 24 22:44:38 2015  sudheerv:cool
Tue Nov 24 22:47:42 2015  bahumbug:Joined the channel
Tue Nov 24 23:03:21 2015  blattj1:Joined the channel
Tue Nov 24 23:16:01 2015  blattj:Joined the channel
Tue Nov 24 23:30:04 2015  es1:Joined the channel

Comments