Displaying #traffic-server/2015-12-04.log:

Fri Dec 4 00:59:22 2015  biilmann:Joined the channel
Fri Dec 4 01:04:23 2015  biilmann:Joined the channel
Fri Dec 4 01:04:33 2015  zwoop:maskit nice!
Fri Dec 4 01:13:47 2015  maskit::)
Fri Dec 4 01:32:30 2015  masaori:zwoop: maskit's patch looks good to me. should I wait somebody's review before land this?
Fri Dec 4 01:33:37 2015  zwoop:there are no set rules, if you are comfortable with it, and it seems safe and solid, commit it. If you are slightly uncertain, ask someone else to take a second look, but it's your discretion. Although, thinking about it, maybe worth leaving it for a day at least?
Fri Dec 4 01:36:05 2015  masaori:ok, leave it a while
Fri Dec 4 01:36:18 2015  biilmann:Joined the channel
Fri Dec 4 03:57:10 2015  zwoop:bcall I don't know if it's a "serious" problem, but clang-format on the 6.0.x branch is not correct.
Fri Dec 4 03:57:11 2015  zwoop:http://paste.fedoraproject.org/297296/01394144
Fri Dec 4 04:19:10 2015  jpeach:swoc: how can auto_ptr be deprecated but tr1 is not? that's crazy town!
Fri Dec 4 04:19:51 2015  jpeach:igalic: right at the top of that page it says "This utility is deprecated as of v6.0.0, and replaced with traffic_ctl."
Fri Dec 4 04:19:53 2015  zwoop:the gcc people are nuts I guess
Fri Dec 4 04:20:37 2015  jpeach:just turn #pragma Wno-deprecated around it
Fri Dec 4 04:20:49 2015  zwoop:jpeach the damn OSX builds are hanging again, at
Fri Dec 4 04:20:53 2015  zwoop:Making check in net
Fri Dec 4 04:20:53 2015  zwoop: CXX test_certlookup.o
Fri Dec 4 04:20:53 2015  zwoop: CXXLD test_certlookup
Fri Dec 4 04:21:05 2015  zwoop: /Applications/Xcode.app/Contents/Developer/usr/bin/make check-TESTS
Fri Dec 4 04:21:17 2015  zwoop:is that an issue with XCode license ?
Fri Dec 4 04:22:16 2015  jpeach:dunno, check the box?
Fri Dec 4 04:34:54 2015  jpeach:hey zwoop
Fri Dec 4 04:56:45 2015  biilmann:Joined the channel
Fri Dec 4 05:01:45 2015  biilmann:Joined the channel
Fri Dec 4 05:33:56 2015  biilmann:Joined the channel
Fri Dec 4 08:33:34 2015  biilmann:Joined the channel
Fri Dec 4 09:19:34 2015  mturk:Joined the channel
Fri Dec 4 09:19:34 2015  mturk:Joined the channel
Fri Dec 4 09:24:49 2015  Lethalman:Joined the channel
Fri Dec 4 09:50:04 2015  Lethalman:hi I did map /_ats_stats/ http://{stat}
Fri Dec 4 09:50:10 2015  Lethalman:but /_ats_stats/cache for example is empty
Fri Dec 4 09:50:19 2015  Lethalman:every endpoint there is empty
Fri Dec 4 09:50:55 2015  Lethalman:maybe I shouldn't use {stat} directly ?
Fri Dec 4 09:51:14 2015  Lethalman:ah yeah
Fri Dec 4 09:51:18 2015  Lethalman:{cache} ecc. for example, ok
Fri Dec 4 09:52:22 2015  Lethalman:ok seems to work
Fri Dec 4 09:53:03 2015  Lethalman:oh too bad the urls are broken with two levels of dirs
Fri Dec 4 09:53:43 2015  Lethalman::(
Fri Dec 4 09:59:14 2015  biilmann:Joined the channel
Fri Dec 4 10:12:07 2015  BlackCobra:Joined the channel
Fri Dec 4 11:16:19 2015  lrea:Joined the channel
Fri Dec 4 11:29:38 2015  Lethalman:mh is healthchecks.so working for anybody?
Fri Dec 4 11:30:06 2015  Lethalman:I get a 404 for the healthcheck uri
Fri Dec 4 12:01:02 2015  biilmann:Joined the channel
Fri Dec 4 12:05:47 2015  Lethalman:weird, I'm not getting errors from plugins
Fri Dec 4 12:05:48 2015  Lethalman::(
Fri Dec 4 12:13:20 2015  Lethalman:oh ok
Fri Dec 4 12:13:24 2015  Lethalman:relative paths of the config just don't work
Fri Dec 4 12:13:25 2015  Lethalman:pff
Fri Dec 4 12:47:38 2015  JSeymour:Joined the channel
Fri Dec 4 13:01:28 2015  biilmann:Joined the channel
Fri Dec 4 13:51:10 2015  reveller:Joined the channel
Fri Dec 4 14:02:12 2015  biilmann:Joined the channel
Fri Dec 4 14:20:11 2015  Lethalman:anybody to explain me the difference between proxy.config.cache.target_fragment_size and proxy.config.cache.min_average_object_size ?
Fri Dec 4 14:20:36 2015  Lethalman:does target_fragment_size impact the disk layout, or it's just the block size to read?
Fri Dec 4 14:20:48 2015  Lethalman:because by default it's 1mb and sounds scary
Fri Dec 4 14:23:51 2015  zwoop:the fragment size impacts the disk layout, you generally don't want to change it
Fri Dec 4 14:23:58 2015  zwoop:why does it sound scary ?
Fri Dec 4 14:24:14 2015  Lethalman:zwoop, in what way does it impact the disk layout?
Fri Dec 4 14:24:28 2015  Lethalman:zwoop, I'm trying to understand what a fragment is but I fail to understand
Fri Dec 4 14:24:34 2015  zwoop:it changes the fragment sizes :)
Fri Dec 4 14:24:41 2015  Lethalman:zwoop, if I store 10k documents, will it waste 1mb-10k?
Fri Dec 4 14:24:49 2015  zwoop:no
Fri Dec 4 14:24:52 2015  Lethalman:ok
Fri Dec 4 14:25:01 2015  zwoop:it will bundle multiple small objects into a fragment
Fri Dec 4 14:25:10 2015  zwoop:but, a large object gets split up into 1MB fragments, and they get linked
Fri Dec 4 14:25:25 2015  Lethalman:ok
Fri Dec 4 14:25:47 2015  Lethalman:while proxy.config.cache.min_average_object_size ?
Fri Dec 4 14:25:58 2015  Lethalman:if I set it to 20kb, will it waste 20kb-10kb?
Fri Dec 4 14:26:13 2015  Lethalman:or it only impacts the directory entries?
Fri Dec 4 14:29:09 2015  zwoop:only impacts directory entries
Fri Dec 4 14:29:16 2015  zwoop:if you set it too large, you risk running out of dir entries
Fri Dec 4 14:29:18 2015  Lethalman:zwoop, awesome thanks
Fri Dec 4 14:29:22 2015  zwoop:if you set it too small, you waste memory
Fri Dec 4 14:29:42 2015  zwoop:you should check with amc though, to make sure I'm not lying (re: fragment sizes)
Fri Dec 4 14:29:45 2015  Lethalman:zwoop, so target_fragment_size is the kind of "block size" for objects
Fri Dec 4 14:29:57 2015  zwoop:yeah, I think so
Fri Dec 4 14:30:00 2015  Lethalman:thanks
Fri Dec 4 14:30:15 2015  zwoop:except, it doesn't waste the block
Fri Dec 4 14:30:33 2015  zwoop:https://docs.trafficserver.apache.org/en/latest/developer-guide/architecture/index.en.html
Fri Dec 4 14:31:56 2015  Lethalman:zwoop, yeah reading that, sometimes I feel something is not deeply explained though :)
Fri Dec 4 14:32:47 2015  Humbedooh:zwoop: if you want asfbot move, just make a ticket ;)
Fri Dec 4 14:32:51 2015  Humbedooh:moved*
Fri Dec 4 14:35:53 2015  esproul:Joined the channel
Fri Dec 4 14:47:12 2015  jrushford:Joined the channel
Fri Dec 4 15:03:18 2015  biilmann:Joined the channel
Fri Dec 4 15:11:06 2015  zwoop:Humbedooh well, we want him here, but maybe not all the Github PRs?
Fri Dec 4 15:13:13 2015  ibrezac:Joined the channel
Fri Dec 4 15:16:03 2015  blattj:Joined the channel
Fri Dec 4 15:16:04 2015  ibrezac:Joined the channel
Fri Dec 4 15:19:04 2015  jrushford:Joined the channel
Fri Dec 4 15:19:13 2015  blattj1:Joined the channel
Fri Dec 4 15:22:37 2015  biilmann:Joined the channel
Fri Dec 4 15:24:27 2015  danielxu:Joined the channel
Fri Dec 4 15:27:18 2015  swoc:Joined the channel
Fri Dec 4 15:27:44 2015  swoc:Dang nab it, I just spent time researching when TR1 was made available in gcc (it appears to be in 4.3 so we would be OK to use it).
Fri Dec 4 15:38:58 2015  jrushford:Joined the channel
Fri Dec 4 15:48:11 2015  zwoop:Yeah, I think as of 6.0.0, we can use e.g. unordered_map just fine, right ?
Fri Dec 4 15:48:36 2015  zwoop:swoc but still, more modern compilers barf over that auto_ptr :)
Fri Dec 4 15:49:25 2015  zwoop:I made some changes a while back (for 6.0.0 I think) such that configure will detect if C++11 is available (properly), so we can use that in non-core components (e.g. plugins, and cppapi)
Fri Dec 4 15:50:22 2015  swoc:Yes, unordered_map but I was looking at unique_ptr to replace auto_ptr.
Fri Dec 4 15:56:30 2015  swoc:zwoop - but what happens in such cases when eleventy isn't detected? Compile file? Don't build that component?
Fri Dec 4 15:56:37 2015  zwoop:ah
Fri Dec 4 15:56:47 2015  zwoop:yeah, don't build
Fri Dec 4 15:57:00 2015  zwoop:we do that for e.g. cppapi I think (I don't think we have any plugins, today, that requires C++11)
Fri Dec 4 15:57:12 2015  swoc:That's fine with me.
Fri Dec 4 15:58:08 2015  zwoop:so is unique_ptr available on RHEL6 ?
Fri Dec 4 15:58:13 2015  zwoop:if it is, we should be ok to use it
Fri Dec 4 15:58:33 2015  swoc:It should be. But you have to access it slightly differently because it's in TR1 and not std:
Fri Dec 4 15:59:40 2015  swoc:Look at atscppapi/src/include/atsccppapi/shared_ptr.h for an example.
Fri Dec 4 16:05:40 2015  niq:Joined the channel
Fri Dec 4 16:15:32 2015  jrushford:Joined the channel
Fri Dec 4 16:19:25 2015  blattj:Joined the channel
Fri Dec 4 16:50:15 2015  JSeymour:Joined the channel
Fri Dec 4 16:54:02 2015  zwoop:ah
Fri Dec 4 16:55:47 2015  blattj:Joined the channel
Fri Dec 4 17:27:48 2015  danielxu:Was there ever a point in time where `ink_assert()` crashed ATS? Does it still?
Fri Dec 4 17:28:17 2015  danielxu:Crash as in purposely crash
Fri Dec 4 17:28:33 2015  Lethalman:danielxu, when you compile with debug
Fri Dec 4 17:28:53 2015  sudheerv:Lethalman knows all about it :)
Fri Dec 4 17:29:03 2015  Lethalman:ahah xD
Fri Dec 4 17:29:09 2015  sudheerv:Lethalman: didn't you have a strange crash related to that :)?
Fri Dec 4 17:29:17 2015  Lethalman:yes it's in an issue somewhere
Fri Dec 4 17:29:23 2015  sudheerv:something for which we went in circles
Fri Dec 4 17:29:27 2015  sudheerv:yeah
Fri Dec 4 17:29:28 2015  Lethalman:it's a normal abort()
Fri Dec 4 17:30:50 2015  swoc:ink_release_assert() fails in non-debug mode but be careful about adding that.
Fri Dec 4 17:32:49 2015  danielxu:Ok I see, that explains a lot
Fri Dec 4 17:32:57 2015  danielxu:Thanks!
Fri Dec 4 17:34:10 2015  swoc:Is that already clang formatted?
Fri Dec 4 17:42:03 2015  danielxu:Shoot, no. I forgot
Fri Dec 4 17:42:26 2015  swoc:Just do a force push to fix it.
Fri Dec 4 17:42:27 2015  bcall:danielxu: I am sure there are some cases where ink_assert() would fail in prod, most of the time we gracefully fail instead of asserting for production
Fri Dec 4 17:44:29 2015  jpeach:ink_assert is basically for additional debugging strictness, ink_release_assert is for programmer errors IMHO
Fri Dec 4 17:54:13 2015  swoc:danielxu - Sorry, I'm waiting for another intern to arrive so I'm stuck in the vestibule.
Fri Dec 4 18:10:37 2015  ibrezac:Joined the channel
Fri Dec 4 18:32:26 2015  reveller:Joined the channel
Fri Dec 4 18:33:39 2015  es:Joined the channel
Fri Dec 4 18:43:27 2015  niq:Joined the channel
Fri Dec 4 19:20:12 2015  reveller1:Joined the channel
Fri Dec 4 19:25:19 2015  jpeach:zwoop, bcall: maybe we have a bunch of cfixed jiras that are still open, eg TS-3648
Fri Dec 4 19:29:48 2015  _klk_:Joined the channel
Fri Dec 4 19:34:22 2015  biilmann:Joined the channel
Fri Dec 4 19:40:34 2015  reveller2:Joined the channel
Fri Dec 4 20:05:00 2015  _klk_:Joined the channel
Fri Dec 4 20:05:37 2015  sudheerv:zwoop: ping
Fri Dec 4 20:05:46 2015  zwoop:png
Fri Dec 4 20:06:04 2015  sudheerv:is CC: max-age=0 identical to must-revalidate?
Fri Dec 4 20:06:20 2015  zwoop:Hmmmm
Fri Dec 4 20:06:31 2015  sudheerv:or do you need both to cache and force revalidation?
Fri Dec 4 20:06:41 2015  sudheerv:there's also the no-cache, which seems simialr
Fri Dec 4 20:07:25 2015  sudheerv:or may be, the easiest is to do <max-age=n, must-revalidate>, where n > 0?
Fri Dec 4 20:11:22 2015  zwoop:Hmmm
Fri Dec 4 20:12:04 2015  zwoop:this is you acting as a client to origin ?
Fri Dec 4 20:12:27 2015  sudheerv:no, proxy acting on a server response
Fri Dec 4 20:12:34 2015  sudheerv:so, we have this failover mechanism
Fri Dec 4 20:12:37 2015  zwoop:who's sending the max-age etc ?
Fri Dec 4 20:12:46 2015  sudheerv:our upstream layer (could be origin)
Fri Dec 4 20:13:06 2015  sudheerv:so, what i'd like is when I get a 404 from origin for example, do a failover
Fri Dec 4 20:13:23 2015  sudheerv:by failover, i mean do a 3xx based redirect follow
Fri Dec 4 20:13:36 2015  sudheerv:but, if i had 404 in the cache even with 1 sec
Fri Dec 4 20:13:42 2015  sudheerv:there's a race condition and sometimes
Fri Dec 4 20:13:47 2015  sudheerv:the 404 is served from teh cache direct
Fri Dec 4 20:13:51 2015  sudheerv:instead of the retry i expect
Fri Dec 4 20:13:59 2015  zwoop:let me rephrase
Fri Dec 4 20:14:02 2015  sudheerv:so, i want the 404 to be saved in cache, but as a must-revalidate
Fri Dec 4 20:14:05 2015  zwoop:are you talking about request headers, or response headers
Fri Dec 4 20:14:09 2015  sudheerv:response headers
Fri Dec 4 20:14:23 2015  zwoop:you wouldn't use must-revalidate in a response, would you ?
Fri Dec 4 20:14:31 2015  sudheerv:really?
Fri Dec 4 20:14:33 2015  zwoop:oh maybe you do
Fri Dec 4 20:14:39 2015  zwoop:looking
Fri Dec 4 20:14:42 2015  sudheerv:yeah, it's a response header, no?
Fri Dec 4 20:14:55 2015  zwoop:The "must-revalidate" response directive indicates that once it has
Fri Dec 4 20:14:56 2015  zwoop: become stale, a cache MUST NOT use the response to satisfy subsequent
Fri Dec 4 20:14:56 2015  zwoop: requests without successful validation on the origin server.
Fri Dec 4 20:15:01 2015  zwoop:yeah
Fri Dec 4 20:15:03 2015  zwoop:I'm smoking crack
Fri Dec 4 20:15:09 2015  zwoop:max-age can be either
Fri Dec 4 20:15:14 2015  sudheerv:yeah, that's true
Fri Dec 4 20:15:28 2015  sudheerv:so, i guess my question is
Fri Dec 4 20:15:43 2015  sudheerv:can i just say max-age=0 OR must-revalidate exclusively?
Fri Dec 4 20:15:55 2015  sudheerv:or do i need to always give some max-age for caching the object in the first place
Fri Dec 4 20:16:10 2015  sudheerv:i coudl probably just say max-age=0 alone too, but, not sure if that works
Fri Dec 4 20:16:16 2015  zwoop:Note that cached responses that contain the "must-revalidate" and/or
Fri Dec 4 20:16:16 2015  zwoop: "s-maxage" response directives are not allowed to be served stale
Fri Dec 4 20:16:17 2015  zwoop: (Section 4.2.4) by shared caches. In particular, a response with
Fri Dec 4 20:16:17 2015  zwoop: either "max-age=0, must-revalidate" or "s-maxage=0" cannot be used to
Fri Dec 4 20:16:17 2015  zwoop: satisfy a subsequent request without revalidating it on the origin
Fri Dec 4 20:16:18 2015  zwoop: server.
Fri Dec 4 20:16:44 2015  sudheerv:cool
Fri Dec 4 20:16:54 2015  sudheerv:so, any of them exclusively should work
Fri Dec 4 20:17:02 2015  sudheerv:now the real question is whether any of them will wokr on ATS ;)?
Fri Dec 4 20:17:05 2015  sudheerv:i guess, i will test and find out
Fri Dec 4 20:17:20 2015  sudheerv:i'm worried about the corner case for max-age=0
Fri Dec 4 20:17:32 2015  sudheerv:i mean, if we expect max-age > 0 to cache
Fri Dec 4 20:17:39 2015  sudheerv:(hopefully, not)
Fri Dec 4 20:18:21 2015  zwoop:what do you mean ?
Fri Dec 4 20:18:30 2015  zwoop:Max-age=1234, must-revalidate ?
Fri Dec 4 20:18:46 2015  sudheerv:no, if i just used max-age=0
Fri Dec 4 20:18:57 2015  sudheerv:will that work as I want it (i.e cache the response, but still force revalidation)
Fri Dec 4 20:19:02 2015  sudheerv:OR if I just used must-revalidate
Fri Dec 4 20:19:04 2015  sudheerv:will that work
Fri Dec 4 20:19:13 2015  sudheerv:OR do I have to do something like "max-age=1, must-revalidate"
Fri Dec 4 20:19:21 2015  sudheerv:to make it cache, but, also force revalidate
Fri Dec 4 20:19:41 2015  zwoop:not sure, test it
Fri Dec 4 20:19:42 2015  sudheerv:i'm going with the last one for now..but, will check to see if the other two will work or fail
Fri Dec 4 20:19:44 2015  sudheerv:yeah
Fri Dec 4 20:20:16 2015  sudheerv:thanks!
Fri Dec 4 20:40:20 2015  _klk_:Joined the channel
Fri Dec 4 20:41:30 2015  reveller:Joined the channel
Fri Dec 4 21:06:28 2015  frantzipan:Joined the channel
Fri Dec 4 21:33:02 2015  ibrezac:Joined the channel
Fri Dec 4 21:47:23 2015  niq:Joined the channel
Fri Dec 4 21:48:16 2015  niq:Joined the channel
Fri Dec 4 21:49:27 2015  niq:Joined the channel
Fri Dec 4 22:37:36 2015  biilmann:Joined the channel
Fri Dec 4 22:54:15 2015  biilmann:Joined the channel
Fri Dec 4 22:56:14 2015  _klk_:Joined the channel
Fri Dec 4 23:33:57 2015  swoc:Joined the channel
Fri Dec 4 23:53:01 2015  _klk_:Joined the channel

Comments