Displaying #traffic-server/2015-10-29.log:

Thu Oct 29 00:19:27 2015  psp:Joined the channel
Thu Oct 29 00:20:53 2015  _klk_:Joined the channel
Thu Oct 29 01:23:52 2015  jrushford:Joined the channel
Thu Oct 29 01:41:29 2015  blattj:Joined the channel
Thu Oct 29 01:43:14 2015  blattj1:Joined the channel
Thu Oct 29 01:51:10 2015  masaori_:Left the channel
Thu Oct 29 01:51:19 2015  masaori_:Joined the channel
Thu Oct 29 02:03:31 2015  _klk_:Joined the channel
Thu Oct 29 03:03:22 2015  reveller1:Joined the channel
Thu Oct 29 03:08:17 2015  msekimura:Joined the channel
Thu Oct 29 03:10:57 2015  dynek:Joined the channel
Thu Oct 29 04:09:46 2015  reveller:Joined the channel
Thu Oct 29 05:04:11 2015  _klk_:Joined the channel
Thu Oct 29 07:01:31 2015  akotani:Joined the channel
Thu Oct 29 07:02:02 2015  rokubo:Joined the channel
Thu Oct 29 07:03:32 2015  taksasak:Joined the channel
Thu Oct 29 07:04:04 2015  shinya:Joined the channel
Thu Oct 29 07:17:15 2015  reveller1:Joined the channel
Thu Oct 29 09:05:55 2015  Lethalman:Joined the channel
Thu Oct 29 10:16:18 2015  Amaryllis:dcarlin: regarding our discussion yesterday about x-forwarded-for, do you think "
Thu Oct 29 10:16:35 2015  Amaryllis:proxy.config.http.cache.vary_default_text = X-Forwarded-Proto" would be a neater solution?
Thu Oct 29 10:52:43 2015  masaori:Joined the channel
Thu Oct 29 12:04:52 2015  JSeymour:Joined the channel
Thu Oct 29 12:57:00 2015  skaji:Joined the channel
Thu Oct 29 13:05:42 2015  shinrich1:Joined the channel
Thu Oct 29 13:15:19 2015  mturk:Joined the channel
Thu Oct 29 13:15:19 2015  mturk:Joined the channel
Thu Oct 29 13:25:15 2015  bahumbug:Joined the channel
Thu Oct 29 13:32:14 2015  esproul:Joined the channel
Thu Oct 29 13:38:03 2015  es:Joined the channel
Thu Oct 29 13:39:01 2015  es1:Joined the channel
Thu Oct 29 13:58:07 2015  zwoop:Amaryllis is this some sort of parent setup? Clients wouldn't send X-Forwarded-Proto
Thu Oct 29 14:17:38 2015  Amaryllis:zwoop: there's a load balancer/ssl terminator in front of ATS
Thu Oct 29 14:18:11 2015  zwoop:ah, that inserts the X-Forward-<x> header ?
Thu Oct 29 14:19:18 2015  Amaryllis:yes
Thu Oct 29 14:20:27 2015  zwoop:I've never used proxy.config.http.cache.vary_default_text, but seems like it ought to work.
Thu Oct 29 14:21:05 2015  zwoop:it's not overridable, so it's a global for all requests. It might be neater to inject the Vary: header specifically for those remap rules / origins where this applies (using e.g. header rewrite plugin)
Thu Oct 29 14:23:43 2015  zwoop:PSUdaemon can you take a look at TS-3989
Thu Oct 29 14:23:44 2015  zwoop:TS-3989
Thu Oct 29 14:24:29 2015  Amaryllis:zwoop: that's what i did at first, but it puts the Vary header in the response as well - and all incoming requests go through the load balancer, so doing it per-remap doesn't really gain anything
Thu Oct 29 14:27:17 2015  dcarlin:you're adding x-forwarded-proto with header_rewrite - what hook are you using?
Thu Oct 29 14:30:17 2015  Amaryllis:oh, we changed that since yesterday. it's not set in ATS anymore, it comes with the request
Thu Oct 29 14:34:16 2015  zwoop:Vary has to come from origin
Thu Oct 29 14:34:27 2015  zwoop:(or using that config to pretend the origin set it)
Thu Oct 29 14:59:45 2015  reveller:Joined the channel
Thu Oct 29 15:25:11 2015  PSUdaemon:zwoop: RE TS-3989: http://paste.fedoraproject.org/284866/13228414
Thu Oct 29 15:35:31 2015  amc:Why is there there a check for first_thread on line 183? It's just been set to 1, that's always true.
Thu Oct 29 15:37:33 2015  shinrich1:Joined the channel
Thu Oct 29 15:37:55 2015  amc:Otherwise I think that's reasonable, PSUdaemon
Thu Oct 29 15:38:45 2015  PSUdaemon:saying i shoud just factor that out?
Thu Oct 29 15:39:05 2015  PSUdaemon:amc: while you are reviewing, did you look at my other patch? :)
Thu Oct 29 15:39:38 2015  amc:No, fling it at me.
Thu Oct 29 15:40:03 2015  PSUdaemon:amc: http://ur1.ca/o7lxx
Thu Oct 29 15:40:11 2015  amc:I would just remove first_thread from the "if (first_thread && !i)" because it's pointless.
Thu Oct 29 15:41:14 2015  davet_:Joined the channel
Thu Oct 29 15:41:35 2015  amc:For the other patch, it assumes that if you don't have POSIX_MADVISE you have the non-POSIX variant?
Thu Oct 29 15:41:49 2015  PSUdaemon:yeah
Thu Oct 29 15:42:03 2015  PSUdaemon:that's more or less existing code
Thu Oct 29 15:42:07 2015  PSUdaemon:that works
Thu Oct 29 15:42:22 2015  PSUdaemon:i refactored it a bit because i don't care about the vars anymore
Thu Oct 29 15:43:31 2015  amc:Why call it at all, if you don't care about the return value?
Thu Oct 29 15:44:39 2015  PSUdaemon:i do care about the return value
Thu Oct 29 15:44:49 2015  PSUdaemon:i just don't need the extra variable
Thu Oct 29 15:45:10 2015  amc:But the call site just drops teh return value on the floor.
Thu Oct 29 15:45:24 2015  amc:Line 58
Thu Oct 29 15:45:25 2015  PSUdaemon:the call to ats_madvise does
Thu Oct 29 15:45:34 2015  PSUdaemon:but we are wrapping an API there
Thu Oct 29 15:45:47 2015  PSUdaemon:so i wanted to mimic that API incase it's used elsewhere
Thu Oct 29 15:46:03 2015  amc:So, why not delete line 58?
Thu Oct 29 15:46:13 2015  PSUdaemon:because i care about the side effect
Thu Oct 29 15:46:22 2015  amc:What's the side effect?
Thu Oct 29 15:46:31 2015  amc:The data at newp is modified?
Thu Oct 29 15:46:36 2015  PSUdaemon:no
Thu Oct 29 15:46:51 2015  PSUdaemon:http://man7.org/linux/man-pages/man2/madvise.2.html
Thu Oct 29 15:47:01 2015  blattj:Joined the channel
Thu Oct 29 15:47:07 2015  amc:OK
Thu Oct 29 15:47:36 2015  PSUdaemon:in the future it might be more important to deal with the return code
Thu Oct 29 15:47:45 2015  PSUdaemon:if we advise something that we have stronger requirements on
Thu Oct 29 15:47:50 2015  PSUdaemon:in this case, if it works, great
Thu Oct 29 15:47:53 2015  PSUdaemon:if not, oh well
Thu Oct 29 15:48:02 2015  PSUdaemon:i suppose o could add a Warning()
Thu Oct 29 15:48:06 2015  PSUdaemon:if it doesn't work
Thu Oct 29 15:48:45 2015  PSUdaemon:amc: http://paste.fedoraproject.org/284875/33712144
Thu Oct 29 15:48:48 2015  PSUdaemon:updated version of the first one
Thu Oct 29 15:50:07 2015  amc:Heh. I've been mucking about in that code for Y! and I did the same thing of moving the (void)tid to an #else :-)
Thu Oct 29 15:50:33 2015  PSUdaemon:who void tid's outside an else?!
Thu Oct 29 15:50:39 2015  PSUdaemon:i mean REALLY?!
Thu Oct 29 15:50:58 2015  amc:In public, no less.
Thu Oct 29 15:51:17 2015  PSUdaemon:it's like they don't even have OCD
Thu Oct 29 15:51:40 2015  amc:Not real programmers.
Thu Oct 29 15:51:47 2015  amc:They're not ...
Thu Oct 29 15:52:00 2015  amc:Yeah, both of those seem reasonable.
Thu Oct 29 15:52:07 2015  PSUdaemon:k
Thu Oct 29 15:52:10 2015  PSUdaemon:shipping them!
Thu Oct 29 15:54:25 2015  PSUdaemon:oh, what about the ink_aseert?
Thu Oct 29 15:54:34 2015  PSUdaemon:i think i need to add that back in
Thu Oct 29 15:54:56 2015  amc:line 65?
Thu Oct 29 15:55:12 2015  PSUdaemon:yeah
Thu Oct 29 15:55:52 2015  PSUdaemon:also, i think the ats_madvise call on L59 needs an INK_ALGN too
Thu Oct 29 15:56:02 2015  PSUdaemon:cause the addr and size both need to be page aligned
Thu Oct 29 15:56:08 2015  PSUdaemon:and that is what we alloced anyway
Thu Oct 29 15:57:33 2015  Becoming:Joined the channel
Thu Oct 29 15:59:06 2015  amc:Yes, you should save the size and pass it on.
Thu Oct 29 15:59:32 2015  amc:The people at work here, they *talk* to me. It's freeky.
Thu Oct 29 16:03:17 2015  es:Joined the channel
Thu Oct 29 16:03:53 2015  PSUdaemon:amc: http://ur1.ca/o7ptv
Thu Oct 29 16:05:15 2015  amc:Shouldn't the if atL56 just be an else?
Thu Oct 29 16:05:57 2015  PSUdaemon:no
Thu Oct 29 16:05:58 2015  amc:Yeah, i think it's right to put the ink_assert back in.
Thu Oct 29 16:06:12 2015  amc:Because ats_alloc_hugepage can fail?
Thu Oct 29 16:06:17 2015  PSUdaemon:we can fail at aquiring a huge page
Thu Oct 29 16:06:19 2015  PSUdaemon:yeah
Thu Oct 29 16:06:22 2015  amc:OK
Thu Oct 29 16:06:52 2015  PSUdaemon:there is a case where we don't have a huge page sized contiguous block of mem, but we can still make a regular sized page allocation
Thu Oct 29 16:06:54 2015  msekimura:Joined the channel
Thu Oct 29 16:06:57 2015  amc:+1 then
Thu Oct 29 16:07:29 2015  amc:Ah, a huge page must be contiguous in physical memory even if the VM allows the address span.
Thu Oct 29 16:08:10 2015  _klk_:Joined the channel
Thu Oct 29 16:11:33 2015  PSUdaemon:i don't really follow that, but sure...
Thu Oct 29 16:11:40 2015  PSUdaemon:i'm keeping the +1 anyway
Thu Oct 29 16:14:45 2015  jpeach:a huge vm page maps to the same size of contiguous physical memory by definition, right?
Thu Oct 29 16:15:06 2015  amc:Yes. And it can't be split because otherwise it's not a huge page.
Thu Oct 29 16:15:15 2015  jpeach:if the physical memory was not contiguous you'd need more than one vm page so it would not be huge
Thu Oct 29 16:15:19 2015  jpeach:bingo!
Thu Oct 29 16:33:13 2015  blattj:Joined the channel
Thu Oct 29 16:34:25 2015  blattj:Joined the channel
Thu Oct 29 16:35:13 2015  blattj:Joined the channel
Thu Oct 29 16:47:46 2015  gancho:Joined the channel
Thu Oct 29 16:48:41 2015  bcall:PSUdaemon: TS-3883 yay!
Thu Oct 29 16:49:35 2015  zwoop:sweet
Thu Oct 29 16:52:54 2015  jrushford:Joined the channel
Thu Oct 29 16:53:51 2015  zwoop:PSUdaemon TS-3883 is going into 5.3.x too right ?
Thu Oct 29 16:54:32 2015  PSUdaemon:i dunno
Thu Oct 29 16:54:37 2015  PSUdaemon:convince me...
Thu Oct 29 16:55:26 2015  zwoop:do it!
Thu Oct 29 16:55:33 2015  PSUdaemon:fwiw, i backported them both to our internal branch
Thu Oct 29 16:55:44 2015  PSUdaemon:not sure if i am ready to foist it upon the masses
Thu Oct 29 16:56:09 2015  PSUdaemon:because it fundamentally changes how freelists allocate memory
Thu Oct 29 16:56:14 2015  PSUdaemon:all page aligned now
Thu Oct 29 16:56:18 2015  PSUdaemon:which i think is a good thing
Thu Oct 29 16:56:25 2015  PSUdaemon:but i may be a moron
Thu Oct 29 16:56:30 2015  PSUdaemon:and this is all a disaster waiting to hapen
Thu Oct 29 16:56:32 2015  PSUdaemon:happen*
Thu Oct 29 16:57:00 2015  PSUdaemon:fwiw, it's more like huge pages now
Thu Oct 29 16:57:09 2015  PSUdaemon:i wonder if that code can be condensed now too
Thu Oct 29 16:57:18 2015  PSUdaemon:since they are essentially the same path with a different size
Thu Oct 29 16:57:26 2015  zwoop:PSUdaemon trying to cherry-pick TS-3883 to 6.0.x has a lot of merge conflicts :/
Thu Oct 29 16:59:19 2015  PSUdaemon:yikes
Thu Oct 29 16:59:24 2015  PSUdaemon:that doesn't merge very well, does it
Thu Oct 29 17:00:41 2015  zwoop:yeah
Thu Oct 29 17:01:14 2015  PSUdaemon:i think there is another commit i did right in there
Thu Oct 29 17:01:45 2015  PSUdaemon:21a692ba242edb32916817535b82621a1aee8e0d
Thu Oct 29 17:03:30 2015  shinrich2:Joined the channel
Thu Oct 29 17:03:31 2015  zwoop:ah, so pick both ?
Thu Oct 29 17:03:39 2015  PSUdaemon:hold one
Thu Oct 29 17:03:58 2015  PSUdaemon:i think this was a big change too: e56d1f9eb6f55b0257d1fb758b773dea468159ee
Thu Oct 29 17:04:20 2015  zwoop:heh
Thu Oct 29 17:04:27 2015  zwoop:ok, so what's the order? :)
Thu Oct 29 17:05:15 2015  PSUdaemon:well, i can make you a 6.0.x patch if you want
Thu Oct 29 17:05:34 2015  PSUdaemon:would that be easier?
Thu Oct 29 17:09:00 2015  PSUdaemon:http://ur1.ca/o7pzc
Thu Oct 29 17:10:59 2015  bcall:if you want you can add that patch to the bug and mark it backport to 6.0.1
Thu Oct 29 17:13:27 2015  ASFBot:Joined the channel
Thu Oct 29 17:13:57 2015  jpeach:PSUdaemon: all freelist chunks should be page-alligned anyway
Thu Oct 29 17:17:27 2015  PSUdaemon:before they were aligned, but not page aligned
Thu Oct 29 17:17:43 2015  PSUdaemon:they'd be like 8 or 16 byte aligned
Thu Oct 29 17:18:09 2015  PSUdaemon:my changes make them 4096 byte aligned (usually) and add extra objects to fill out the page
Thu Oct 29 17:18:39 2015  zwoop:PSUdaemon it'd be easier if there was an order to cherry-pick
Thu Oct 29 17:18:43 2015  PSUdaemon:or, are you saying +1 to backporting because it should have been that way from the starT?
Thu Oct 29 17:19:11 2015  PSUdaemon:zwoop: do you want to cherry pick jpeach's runtime freelist commit too?
Thu Oct 29 17:19:28 2015  jpeach:PSUdaemon: not necessary
Thu Oct 29 17:19:40 2015  zwoop:if that's required to get this to apply, then probably, but if it can be avoided, that's probably better
Thu Oct 29 17:19:45 2015  jpeach:PSUdaemon: freelist chunks should be a multiple of pagesize imho
Thu Oct 29 17:20:09 2015  jpeach:large allocations will almost certainly be page aligned because malloc will get them from mmap
Thu Oct 29 17:22:23 2015  psp:Joined the channel
Thu Oct 29 17:22:52 2015  PSUdaemon:jpeach: i think we should just move explicitly to mmap at some point
Thu Oct 29 17:23:24 2015  PSUdaemon:zwoop: well, i provided you a patch that applies, but if you want to cherry-pick, and you want it to be clean, i'll have to do some digging
Thu Oct 29 17:23:52 2015  zwoop:PSUdaemon it just gets weird managing a release branch, which is not merging cleanly with the main repo
Thu Oct 29 17:24:21 2015  zwoop:such that, a git merge in the future is likely to fail big time
Thu Oct 29 17:32:00 2015  blattj:Joined the channel
Thu Oct 29 17:42:54 2015  PSUdaemon:zwoop: git cherry-pick 21a692ba242edb32916817535b82621a1aee8e0d 0bae26a9 e56d1f9eb6f55b0257d1fb758b773dea468159ee 471b969
Thu Oct 29 17:43:32 2015  PSUdaemon:i am a magician
Thu Oct 29 17:46:56 2015  blattj:Joined the channel
Thu Oct 29 17:50:01 2015  PSUdaemon:zwoop: 13 files changed, 204 insertions(+), 152 deletions(-)
Thu Oct 29 17:50:31 2015  zwoop:Magic!
Thu Oct 29 17:51:01 2015  zwoop:You should post that on the bug too, in case we want this backported to 6.0.x
Thu Oct 29 17:51:10 2015  zwoop:which I bet bcall wants :)
Thu Oct 29 17:51:51 2015  blattj:Joined the channel
Thu Oct 29 18:02:51 2015  blattj:Joined the channel
Thu Oct 29 18:12:13 2015  zwoop:PSUdaemon and thanks :)
Thu Oct 29 18:20:45 2015  _klk_:Joined the channel
Thu Oct 29 18:47:18 2015  blattj:Left the channel
Thu Oct 29 18:52:14 2015  blattj:Joined the channel
Thu Oct 29 20:26:44 2015  blattj:Left the channel
Thu Oct 29 20:37:16 2015  blattj:Joined the channel
Thu Oct 29 21:53:15 2015  zwoop:bcall doh
Thu Oct 29 22:08:05 2015  blattj:Joined the channel
Thu Oct 29 23:09:33 2015  blattj:Joined the channel
Thu Oct 29 23:31:23 2015  bahumbug:Joined the channel
Thu Oct 29 23:40:29 2015  jpeach:zwoop, amc: is the RTD build timing out because the doxygen is taking too long to generate?
Thu Oct 29 23:41:20 2015  jrushford:Joined the channel
Thu Oct 29 23:46:34 2015  jrushford:Joined the channel

Comments