Displaying #traffic-server/2015-10-14.log:

Wed Oct 14 00:01:37 2015  gancho:Joined the channel
Wed Oct 14 02:17:08 2015  _klk_:Joined the channel
Wed Oct 14 03:39:45 2015  jpeach:ASFBot: help
Wed Oct 14 03:40:25 2015  jpeach:Humbedooh: can I get ASFBot to close a pull request?
Wed Oct 14 04:02:40 2015  es:Joined the channel
Wed Oct 14 05:03:04 2015  psp:Joined the channel
Wed Oct 14 06:34:06 2015  Humbedooh:jpeach: yes, provided it's merge into master
Wed Oct 14 08:30:05 2015  Lethalman:Joined the channel
Wed Oct 14 10:37:46 2015  Amaryllis:is there a reason ATS 5.3.1 would not cache redirect responses by default?
Wed Oct 14 10:44:57 2015  Amaryllis:ah, because the response body is empty and there's no content-length
Wed Oct 14 11:48:01 2015  JSeymour:Joined the channel
Wed Oct 14 11:56:39 2015  bahumbug:Joined the channel
Wed Oct 14 13:26:55 2015  shinrich1:Joined the channel
Wed Oct 14 13:27:14 2015  shinrich_:Joined the channel
Wed Oct 14 13:48:19 2015  esproul:Joined the channel
Wed Oct 14 14:24:12 2015  davet_:Joined the channel
Wed Oct 14 14:44:43 2015  reveller1:Joined the channel
Wed Oct 14 14:51:20 2015  reveller:Joined the channel
Wed Oct 14 14:56:12 2015  PSUdaemon:Amaryllis: you can change a setting to cache empty docs
Wed Oct 14 15:00:17 2015  zwoop:PSUdaemon I thought we changed that default ?
Wed Oct 14 15:00:33 2015  zwoop:RECT_CONFIG, "proxy.config.http.cache.allow_empty_doc", RECD_INT, "1"
Wed Oct 14 15:01:28 2015  sudheerv:PSUdaemon: zwoop: a response with headers but no body (Content-Lenth = 0) is not considered an "empty" doc, no?
Wed Oct 14 15:01:44 2015  zwoop:it is
Wed Oct 14 15:01:49 2015  sudheerv:huh, really?
Wed Oct 14 15:01:52 2015  zwoop:yes
Wed Oct 14 15:01:54 2015  zwoop:pretty sure
Wed Oct 14 15:01:57 2015  sudheerv:i thought headers are part of the *doc*?
Wed Oct 14 15:02:01 2015  Amaryllis:yes, the setting was already enabled, the problem was the original wasn't sending content-length
Wed Oct 14 15:02:04 2015  zwoop:sudheerv no
Wed Oct 14 15:02:11 2015  sudheerv:hmm..interesting..
Wed Oct 14 15:02:23 2015  zwoop:sudheerv I'm fairly certain, test it
Wed Oct 14 15:02:35 2015  sudheerv:cool..we have this bug where we sometimes end up caching 0 length body
Wed Oct 14 15:02:38 2015  PSUdaemon:i agree with zwoop
Wed Oct 14 15:02:45 2015  PSUdaemon:it's all about the body
Wed Oct 14 15:02:56 2015  zwoop:Amaryllis yeah, that setting needs a Content-Length for it to be able to know that the body really is empty, and not just the connection broke
Wed Oct 14 15:02:57 2015  sudheerv:cool, but, that setting isn't overridable, i guess?
Wed Oct 14 15:03:02 2015  zwoop:sudheerv That's why ^
Wed Oct 14 15:03:23 2015  sudheerv:zwoop: ok, got it
Wed Oct 14 15:03:28 2015  zwoop:sudheerv without the CL: header, we could accidentally cache a broken request with no body
Wed Oct 14 15:04:16 2015  sudheerv:(ok, i guess the CL requirement could probably be enhanced to consider Chunked encoding case)
Wed Oct 14 15:04:16 2015  zwoop:I'm sure it could be improved, such that if we truly know that the connection did not prematurely close and we really did not expect any body at all, we could cache without a CL: 0
Wed Oct 14 15:04:29 2015  sudheerv:agree, ^^
Wed Oct 14 15:04:42 2015  zwoop:who would send a Chunked encoding with no body :)
Wed Oct 14 15:04:44 2015  zwoop:(but, anything is possible)
Wed Oct 14 15:04:46 2015  sudheerv:lol..
Wed Oct 14 15:05:05 2015  sudheerv:zwoop: it might be possible in spdy
Wed Oct 14 15:05:11 2015  sudheerv:or h2 for that matter
Wed Oct 14 15:05:22 2015  sudheerv:there's no C-L in spdy/h2, no?
Wed Oct 14 15:05:25 2015  sudheerv:you rely on the FIN flag
Wed Oct 14 15:05:28 2015  zwoop:right
Wed Oct 14 15:05:33 2015  zwoop:but does it need Chunked ?
Wed Oct 14 15:05:41 2015  zwoop:the frames themselves are "chunks"
Wed Oct 14 15:05:44 2015  sudheerv:yeah, true
Wed Oct 14 15:05:50 2015  sudheerv:it's sort of like Chunked encoding
Wed Oct 14 15:05:56 2015  zwoop:yeah, it's part of the spec
Wed Oct 14 15:05:57 2015  sudheerv:but, with frames
Wed Oct 14 15:06:29 2015  sudheerv:but, yeah - you are right - we could enhance to consider a case where we absolutely know it's not a broken connection, but, empty body with no C-L
Wed Oct 14 15:06:34 2015  sudheerv:and cache that
Wed Oct 14 15:06:56 2015  sudheerv:zwoop: PSUdaemon: any reason not to make the allow_empty_doc overridable?
Wed Oct 14 15:07:23 2015  sudheerv:or it's not possible, at all?
Wed Oct 14 15:07:26 2015  sudheerv:checking..
Wed Oct 14 15:14:11 2015  zwoop:sudheerv yeah, I'd be +1 on that. I looked at this before, and it was "complex" in that the cache really wanted to know the content length up front :)
Wed Oct 14 15:14:32 2015  zwoop:sudheerv it might be because it's cache related, and we don't communicate the HttpSM into the CacheSM
Wed Oct 14 15:14:36 2015  PSUdaemon:sudheerv: it's in cachesm not httpsm
Wed Oct 14 15:14:49 2015  zwoop:jinx
Wed Oct 14 15:14:52 2015  PSUdaemon:doh
Wed Oct 14 15:14:53 2015  sudheerv:yeah, we already have a way to pass some params in open_write
Wed Oct 14 15:14:56 2015  PSUdaemon:i looked at this yesterday :)
Wed Oct 14 15:15:03 2015  sudheerv:PSUdaemon: i know why ;)
Wed Oct 14 15:15:03 2015  PSUdaemon:cause knutsel asked
Wed Oct 14 15:15:07 2015  zwoop: :)
Wed Oct 14 15:15:29 2015  sudheerv:who's knutsel? is that a reference to me :)
Wed Oct 14 15:15:55 2015  PSUdaemon:no
Wed Oct 14 15:15:57 2015  PSUdaemon:jvd
Wed Oct 14 15:15:59 2015  zwoop:Jan van Doork
Wed Oct 14 15:16:00 2015  sudheerv:ah, ok
Wed Oct 14 15:16:18 2015  sudheerv:i know why Jan would have asked you that question (PSUdaemon probably even knows)
Wed Oct 14 15:16:25 2015  sudheerv:it was all because of me :-/
Wed Oct 14 15:16:26 2015  PSUdaemon:yeah, i know :)
Wed Oct 14 15:16:37 2015  zwoop:lol
Wed Oct 14 15:16:39 2015  PSUdaemon:well, that was actually really helpful
Wed Oct 14 15:16:48 2015  PSUdaemon:i think it explained some specific errors we were seeing
Wed Oct 14 15:16:55 2015  sudheerv:yeah, indeed - i didn't imagine that the empty doc is for 0 body
Wed Oct 14 15:16:59 2015  sudheerv:yeah, it does
Wed Oct 14 15:17:15 2015  zwoop:sudheerv you could have asked me ;)
Wed Oct 14 15:17:21 2015  sudheerv:my day always goes well, if I hop on to #trafficserver early on :)
Wed Oct 14 15:17:29 2015  sudheerv:zwoop: yeah, i should have!
Wed Oct 14 15:17:53 2015  zwoop:that's the thing that is really great with our community, except, it's very non-Apache
Wed Oct 14 15:17:55 2015  Becoming:Joined the channel
Wed Oct 14 15:18:01 2015  zwoop:but, our iRC is sprawling
Wed Oct 14 15:18:16 2015  sudheerv:absolutely :)
Wed Oct 14 15:18:51 2015  zwoop:Humbedooh you should watch our IRC logs in your project status page, and count number of messages and number of active participants ;)
Wed Oct 14 15:18:55 2015  sudheerv:so, i'm thinking for certain file types, i'd disable allow empty doc (assuming, i can make it overridable)
Wed Oct 14 15:19:16 2015  Humbedooh:zwoop: maybe I will, maybe I'll....forget you said that :)
Wed Oct 14 15:19:25 2015  zwoop:Humbedooh as usual
Wed Oct 14 15:19:36 2015  sudheerv:lol
Wed Oct 14 15:22:17 2015  blattj:Joined the channel
Wed Oct 14 15:23:40 2015  _klk_:Joined the channel
Wed Oct 14 15:25:16 2015  sudheerv:PSUdaemon: zwoop: it seems straightforward enough to override empty doc setting?
Wed Oct 14 15:25:24 2015  sudheerv:there's an *options* param being passed to open_write
Wed Oct 14 15:25:34 2015  sudheerv:#define CACHE_WRITE_OPT_OVERWRITE 0x0001
Wed Oct 14 15:25:34 2015  sudheerv:#define CACHE_WRITE_OPT_CLOSE_COMPLETE 0x0002
Wed Oct 14 15:25:34 2015  sudheerv:#define CACHE_WRITE_OPT_SYNC (CACHE_WRITE_OPT_CLOSE_COMPLETE | 0x0004)
Wed Oct 14 15:25:34 2015  sudheerv:#define CACHE_WRITE_OPT_OVERWRITE_SYNC (CACHE_WRITE_OPT_SYNC | CACHE_WRITE_OPT_OVERWRITE)
Wed Oct 14 15:25:37 2015  sudheerv:which currently allows these
Wed Oct 14 15:26:04 2015  sudheerv:we could pass the empty doc (and possibly even a few other settings) as options, no?
Wed Oct 14 15:26:15 2015  sudheerv:options is a bit mask
Wed Oct 14 15:29:24 2015  zwoop:I would have to like, but in general, we need to avoid adding more of these one-offs communicating stuff from HttpSM into CacheSM
Wed Oct 14 15:29:48 2015  sudheerv:hmm..but, that (options) interface seems defined for that purpose?
Wed Oct 14 15:29:50 2015  zwoop:we did something like that recently (a Yahoo addition I think), adding APIs to set CacheSM state based on HttpSM state (for a single value)
Wed Oct 14 15:30:13 2015  zwoop:yeah, maybe this is a case where it's reasonable
Wed Oct 14 15:30:53 2015  sudheerv:yeah, i'll do some more code reading to make sure i'm not messing up things
Wed Oct 14 15:47:12 2015  gancho:Joined the channel
Wed Oct 14 16:01:02 2015  bahumbug:Joined the channel
Wed Oct 14 16:03:21 2015  jpeach:Humbedooh: pull request #298 is merged; what is the magic to close it?
Wed Oct 14 16:13:15 2015  Becoming:Joined the channel
Wed Oct 14 16:28:14 2015  zwoop:sudheerv if a client connects to ATS, but sends nothing, which timeout would trip? (I did turn off defer accept, but still not seeing a timeout)
Wed Oct 14 16:28:35 2015  sudheerv:accept timeout?
Wed Oct 14 16:28:49 2015  zwoop:but it has already accepted the connection, no ?
Wed Oct 14 16:29:11 2015  sudheerv:yeah, the setting is actually <proxy.config.http.accept_no_activity_timeout> :)
Wed Oct 14 16:29:16 2015  sudheerv:so, accepted but no activity :)
Wed Oct 14 16:29:29 2015  sudheerv:i loosely call it accept timeout, but, you are right it's post accept
Wed Oct 14 16:29:29 2015  _klk_:Joined the channel
Wed Oct 14 16:30:00 2015  dcarlin:pretty sure we've all encountered this issue no? :)
Wed Oct 14 16:30:02 2015  zwoop:loki (10:25) 284/0 $ telnet localhost 80
Wed Oct 14 16:30:02 2015  zwoop:Trying 127.0.0.1...
Wed Oct 14 16:30:02 2015  zwoop:Connected to localhost.
Wed Oct 14 16:30:02 2015  zwoop:Escape character is '^]'.
Wed Oct 14 16:30:08 2015  dcarlin:that the accept timeout is really a TTFB timeout
Wed Oct 14 16:30:10 2015  zwoop:it just sits there, never times out ?
Wed Oct 14 16:30:27 2015  sudheerv:zwoop: that timer is default to a slighlty long value
Wed Oct 14 16:30:29 2015  sudheerv:i think 2 min?
Wed Oct 14 16:30:33 2015  sudheerv:let me check..
Wed Oct 14 16:30:34 2015  zwoop:which one ?
Wed Oct 14 16:30:44 2015  sudheerv:proxy.config.http.accept_no_activity_timeout
Wed Oct 14 16:30:54 2015  sudheerv:dcarlin: yeah, you are right - this is a problem with all browsers
Wed Oct 14 16:30:58 2015  zwoop:I set that to 30s, but no timeout
Wed Oct 14 16:31:01 2015  sudheerv:hmm..
Wed Oct 14 16:31:04 2015  sudheerv:weird
Wed Oct 14 16:31:41 2015  dcarlin:maybe its waiting for proxy.config.net.default_inactivity_timeout
Wed Oct 14 16:31:46 2015  dcarlin:which defaults to 1 day
Wed Oct 14 16:32:03 2015  zwoop:hmmm
Wed Oct 14 16:32:05 2015  zwoop:let me try
Wed Oct 14 16:32:50 2015  sudheerv:yeah, that should be the case with defer accept, but, zwoop says he turned that off
Wed Oct 14 16:33:11 2015  zwoop:bingo, dcarlin gets a cookie
Wed Oct 14 16:33:21 2015  sudheerv::)
Wed Oct 14 16:33:27 2015  zwoop:that's bloody weak, because that setting is intended to clean out "leaked" connections
Wed Oct 14 16:33:30 2015  sudheerv:but, does that mean the defer accept is not working?
Wed Oct 14 16:33:39 2015  sudheerv:zwoop: fwiw, we have that setting set to 5 min :)
Wed Oct 14 16:33:45 2015  zwoop:I'm not sure, I set it to 0
Wed Oct 14 16:33:45 2015  dcarlin:yeah that shouldn't be.. I would think proxy.config.http.accept_no_activity_timeout should kick in
Wed Oct 14 16:33:54 2015  zwoop:it doesn't :/
Wed Oct 14 16:34:04 2015  sudheerv:zwoop: any way you can check if the connection was accepted?
Wed Oct 14 16:34:08 2015  sudheerv:gdb?
Wed Oct 14 16:34:20 2015  dcarlin:either way, proxy.config.http.accept_no_activity_timeout doesn't do what its supposed to anyways since its after accept
Wed Oct 14 16:34:56 2015  zwoop:I tried setting defer accept to -1
Wed Oct 14 16:35:24 2015  zwoop:fwiw, I'm trying to see if I can get ATS to send a 408 response
Wed Oct 14 16:35:44 2015  sudheerv:zwoop: 408 is only implemented for a POST timeout scenario i think
Wed Oct 14 16:39:42 2015  frantz_:Joined the channel
Wed Oct 14 16:40:48 2015  jpeach:hey amc can you look at TS-3969
Wed Oct 14 16:43:46 2015  zwoop:sudheerv ok. Well, I can not get it to timeout on a client that sends nothing. Wonder if we have a bug here.
Wed Oct 14 16:44:25 2015  sudheerv:zwoop: even the default inactivity timeout?
Wed Oct 14 16:44:35 2015  sudheerv:that's a safe gaurd, no?
Wed Oct 14 16:44:43 2015  sudheerv:i think if no other timer is active, that kicks in
Wed Oct 14 16:45:00 2015  sudheerv:yeah, the accept timeout, i swear i tested it (a while back though ) :)
Wed Oct 14 16:45:06 2015  sudheerv:i remember it does timeout in 2 min
Wed Oct 14 16:45:12 2015  sudheerv:or may be it's broken
Wed Oct 14 16:45:22 2015  sudheerv:let me try on my version (5.0 :( )
Wed Oct 14 16:50:23 2015  Becoming:Joined the channel
Wed Oct 14 16:57:04 2015  sudheerv:zwoop: about the accept no activity timeout
Wed Oct 14 16:57:08 2015  sudheerv:it does seem to be broken :(
Wed Oct 14 16:57:16 2015  sudheerv:i wonder if its' the protocol probe thing
Wed Oct 14 16:57:25 2015  sudheerv:we expect a byte atleast before accepting?
Wed Oct 14 16:57:38 2015  sudheerv:i mean the session accept (not the socket level accept)
Wed Oct 14 16:57:49 2015  sudheerv:so, the timeout should be added in protocol probe session accept handling
Wed Oct 14 16:58:04 2015  sudheerv:jpeach: wdyt
Wed Oct 14 16:58:48 2015  sudheerv:the current accept no activity timer is started at http session accept level (which only happens after protocol probe session accept)
Wed Oct 14 16:58:51 2015  jpeach:yeh it could be ... we drive a zero-length read, but the event would not be triggered unless some data arrived
Wed Oct 14 16:58:57 2015  sudheerv:so, basically, no timer is running during that window
Wed Oct 14 16:59:03 2015  sudheerv:yeah, seems like it
Wed Oct 14 16:59:08 2015  jpeach:write a tsqa test for it!
Wed Oct 14 16:59:19 2015  sudheerv:lol..first need to fix it
Wed Oct 14 16:59:24 2015  sudheerv:i can easily reproduce it with telnet
Wed Oct 14 16:59:29 2015  sudheerv:zwoop did that too just now :)
Wed Oct 14 16:59:59 2015  jacksontj:Joined the channel
Wed Oct 14 16:59:59 2015  perry:Joined the channel
Wed Oct 14 17:00:26 2015  blattj:Joined the channel
Wed Oct 14 17:00:44 2015  es:Joined the channel
Wed Oct 14 17:01:21 2015  zwoop:yeah, it times out if I lower the default disconnect time
Wed Oct 14 17:03:34 2015  sudheerv:default inactivity time?
Wed Oct 14 17:03:48 2015  sudheerv:yeah, that's a safety net (thanks to bcall)
Wed Oct 14 17:04:00 2015  psp:Joined the channel
Wed Oct 14 17:05:04 2015  Randeep:Joined the channel
Wed Oct 14 17:07:39 2015  daemonkeeper:Joined the channel
Wed Oct 14 17:07:39 2015  luc:Joined the channel
Wed Oct 14 17:09:44 2015  Humbedooh:Joined the channel
Wed Oct 14 17:10:18 2015  Humbedooh:Joined the channel
Wed Oct 14 17:10:30 2015  jacksontj:Joined the channel
Wed Oct 14 17:10:30 2015  perry:Joined the channel
Wed Oct 14 17:10:58 2015  Randeep:zwoop: yesterday you had advised me to use raw disk for the cashing. Usually people will use RAID also on that? If the disk is SSD then also we will use it as raw?
Wed Oct 14 17:15:37 2015  PSUdaemon:Randeep: no RAID
Wed Oct 14 17:15:47 2015  PSUdaemon:use all the disks individually
Wed Oct 14 17:17:43 2015  zwoop:+1
Wed Oct 14 17:17:58 2015  zwoop:Randeep ATS manages the disks itself, looking for errors and "striping" etc.
Wed Oct 14 17:18:26 2015  es:Joined the channel
Wed Oct 14 17:18:58 2015  zwoop:sudheerv yeah, default inactivity time. But, we shouldnt have to rely on that for the idle connection to timeout, no ?
Wed Oct 14 17:26:18 2015  sudheerv:zwoop: agree
Wed Oct 14 17:26:25 2015  sudheerv:we shouldn't be relying on that timer at all - it's just too long
Wed Oct 14 17:26:33 2015  sudheerv:we should fix the protocol probe timer bug
Wed Oct 14 17:28:27 2015  _klk_:Joined the channel
Wed Oct 14 17:31:26 2015  Randeep:zwoop,PSUdaemon : thanks
Wed Oct 14 17:51:46 2015  Randeep:currently I have configured ats as reverse proxy
Wed Oct 14 17:51:47 2015  Randeep:live.domain.com is pointed to ats server and in remap rules I have wrote live.domain.com to original host's ip.
Wed Oct 14 17:51:47 2015  Randeep:Now my networking team wants to forward all http traffic to ats server so that they can check general web pages caching.
Wed Oct 14 17:51:47 2015  Randeep:For that I need to do any modifications? Because caching is enabled and if it is not matched in the remap rules, it will act as a forward proxy right?
Wed Oct 14 17:51:51 2015  Randeep:Do I have to change anything for this?
Wed Oct 14 17:53:17 2015  es1:Joined the channel
Wed Oct 14 17:55:49 2015  _klk_:Joined the channel
Wed Oct 14 18:01:24 2015  blattj:Joined the channel
Wed Oct 14 18:03:44 2015  es:Joined the channel
Wed Oct 14 18:13:37 2015  es:Joined the channel
Wed Oct 14 18:21:58 2015  _klk_:Joined the channel
Wed Oct 14 18:25:25 2015  jpeach:Randeep: sounds like you might want to set proxy.config.url_remap.remap_required=0 so that it will proxy for domains that are not listed in remap.config
Wed Oct 14 18:25:45 2015  jpeach:remember to turn that off before you go to production :)
Wed Oct 14 18:32:45 2015  Randeep:But in that case, what will happen to the rule I have written in remap.config
Wed Oct 14 18:33:33 2015  Randeep:then instead of pointing the live.domain.com to ats, I have to point it to the real host?
Wed Oct 14 18:56:02 2015  _klk_1:Joined the channel
Wed Oct 14 19:22:33 2015  ASFBot:Joined the channel
Wed Oct 14 19:37:38 2015  es:Joined the channel
Wed Oct 14 20:03:29 2015  briang:zwoop, PSUdaemon , hey regarding our call @ 2, I might need to be 15 minutes late because of a meeting
Wed Oct 14 20:03:31 2015  briang:can we start at 2:15?
Wed Oct 14 20:03:40 2015  briang:also I never received hangout information
Wed Oct 14 20:12:32 2015  _klk_:Joined the channel
Wed Oct 14 20:16:17 2015  zwoop:briang we'll just start without you :) And PSUdaemon is setting up the Hangout
Wed Oct 14 20:16:31 2015  briang:thanks
Wed Oct 14 20:16:50 2015  zwoop:PSUdaemon send out the Hangout info :)
Wed Oct 14 20:21:11 2015  es:Joined the channel
Wed Oct 14 20:31:35 2015  PSUdaemon:when we start i will
Wed Oct 14 20:35:03 2015  zwoop:aight
Wed Oct 14 20:35:13 2015  zwoop:email it to summits@ though (not here)
Wed Oct 14 20:46:11 2015  PSUdaemon:ok
Wed Oct 14 20:53:57 2015  es:Joined the channel
Wed Oct 14 21:03:54 2015  zwoop:PSUdaemon ?
Wed Oct 14 21:09:41 2015  PSUdaemon:zwoop: ?
Wed Oct 14 21:09:53 2015  sudheerv:??
Wed Oct 14 21:10:00 2015  zwoop:???
Wed Oct 14 21:10:11 2015  sudheerv:lol
Wed Oct 14 21:26:35 2015  _klk_1:Joined the channel
Wed Oct 14 21:45:59 2015  reveller1:Joined the channel
Wed Oct 14 22:43:31 2015  _klk_:Joined the channel
Wed Oct 14 23:03:12 2015  jpeach:zwoop: did you figure out all the summit scheduling
Wed Oct 14 23:04:28 2015  jpeach:PSUdaemon, bcall: granted you edit access to the proposals docs, sorry it took so long
Wed Oct 14 23:04:42 2015  bcall:jpeach: thx
Wed Oct 14 23:07:19 2015  _klk_:Joined the channel
Wed Oct 14 23:18:14 2015  jpeach:PSUdaemon: does jrushford do IRC?

Comments