/tmp/input.xml:1:86: error: element "tag-list" not allowed here; expected the element end-tag, element "artist", "artist-list", "label", "label-list", "release", "release-group", "release-group-list", "release-list", "track" or "track-list" or an element from another namespace
so commit your schema changes, then we can update mbtest and those two tests should pass.
2011-04-14 10418, 2011
ocharles
done! time to run master and tests and we could see a green light!
2011-04-14 10432, 2011
warp
Updated to revision 12968.
2011-04-14 10441, 2011
ocharles
great
2011-04-14 10446, 2011
ocharles
you running them or am I?
2011-04-14 10453, 2011
warp
I can run them.
2011-04-14 10428, 2011
ocharles
gogogo
2011-04-14 10438, 2011
ocharles
nikki: ngs has new data
2011-04-14 10445, 2011
warp
running. ofcourse that takes ages nowadays with those selenium tests. quite a workout for hobbes.
2011-04-14 10453, 2011
ocharles
heh, yea
2011-04-14 10415, 2011
ocharles
we should be able to switch 'test.mb' to point to the ngs servers soon I think
2011-04-14 10421, 2011
ocharles
though I'm not sure what rob's plan is there
2011-04-14 10433, 2011
warp
I think we keep test as is.
2011-04-14 10442, 2011
ocharles
well test runs next, ngs runs master
2011-04-14 10446, 2011
ocharles
but ngs is where the testing should take place, oddly
2011-04-14 10455, 2011
warp
we need a test server to test changes after ngs release anyway, I see no need to move that subdomain name somewhere else.
2011-04-14 10403, 2011
ocharles
yea, true
2011-04-14 10432, 2011
warp
functionality changes can still be tested on test. we just need performance testing on ngs.
2011-04-14 10441, 2011
warp
ideally, ngs should only run master IMO.
2011-04-14 10447, 2011
ocharles
yep, agreed there
2011-04-14 10453, 2011
ocharles
ngs is also failing all the performance tests atm
2011-04-14 10412, 2011
warp
I glimpsed some of the discussion with rob on that.
2011-04-14 10417, 2011
warp
but didn't follow it closely.
2011-04-14 10419, 2011
ocharles
t'aint looking too good right now
2011-04-14 10437, 2011
warp
any idea where the trouble is?
2011-04-14 10408, 2011
ocharles
not really at the moment, I'm about to start doing some profiling
2011-04-14 10414, 2011
warp
ah, ok
2011-04-14 10452, 2011
ocharles
warp: are the tests still running?
2011-04-14 10455, 2011
ocharles
there's a report at 44%
2011-04-14 10402, 2011
warp
yes, still running.
2011-04-14 10412, 2011
ocharles
ok, that might have been me
2011-04-14 10416, 2011
ocharles removes that report
2011-04-14 10449, 2011
warp
yay! 100%
2011-04-14 10410, 2011
ocharles
sigh, devel-nytprof isn't installed on the new servers so I can't profile there
2011-04-14 10437, 2011
ocharles
i wonder if hobbes can access totoro
2011-04-14 10406, 2011
warp does not know.
2011-04-14 10407, 2011
ocharles
shoot, hobbes can't even ping it
2011-04-14 10455, 2011
warp
that doesn't necessarily mean that other kinds of traffic are also not allowed.
2011-04-14 10408, 2011
warp
(depends on how paranoid djce is ;)
2011-04-14 10402, 2011
ocharles
true :) but psql can't connect to it either
2011-04-14 10454, 2011
djce
64 bytes from totoro.localdomain (10.1.1.19): icmp_req=1 ttl=64 time=17.4 ms
2011-04-14 10455, 2011
djce
64 bytes from totoro.localdomain (10.1.1.19): icmp_req=2 ttl=64 time=0.214 ms
2011-04-14 10459, 2011
djce
orly?
2011-04-14 10428, 2011
ocharles
hrm
2011-04-14 10412, 2011
ocharles
--- 10.1.1.9 ping statistics ---
2011-04-14 10420, 2011
ocharles
what the hell djce, you mean I have to ping the RIGHT address too?
2011-04-14 10444, 2011
ocharles puts the dunce hat on
2011-04-14 10439, 2011
warp suggests 'echo search localdomain >> /etc/resolv.conf' for all our servers.
2011-04-14 10404, 2011
djce
There is a ".mb" domain for lazy typers.
2011-04-14 10416, 2011
djce
hobbes.mb, etc.
2011-04-14 10431, 2011
warp
ah
2011-04-14 10438, 2011
hawke_ joined the channel
2011-04-14 10430, 2011
ocharles
warp: ah, good news, ish. if I do time curl "http://127.0.0.1:3000/ws/1/release/138b9e89-e4f0-4bed-99fe-6a963728a292?type=xml&inc=tracks+puids+artist+release-events+labels+isrcs+artist-rels+release-rels+url-rels" on astro, it takes 0.00s, ie - blazingly fast
but if I do it from my machine to ngs.mb.org, it takes 0.6s
2011-04-14 10408, 2011
ocharles
so it seems for some reason there's more overhead to actually transport the request from server to client for ngs.mb, rather than the normal mb servers
2011-04-14 10417, 2011
nikki giggles
2011-04-14 10433, 2011
nikki
thunderbird thinks that profiling script commit is an email scam
2011-04-14 10439, 2011
ocharles
haha
2011-04-14 10428, 2011
warp
ocharles: oh, interesting!
2011-04-14 10455, 2011
ocharles
however, even if I compare the time to do a request to localhost and compare against the logs it still loses
2011-04-14 10413, 2011
ocharles
i guess the next step could be to try and do the tests against an internal catalyst server
2011-04-14 10432, 2011
ocharles
but however you swing it, ngs ws/1 is looking slower
2011-04-14 10453, 2011
ocharles
I timed the portion of /ws/1/release/lookup doing a request and it's about 50/50 for time spent fetching data, and time spent rendering the xml
2011-04-14 10404, 2011
ocharles
it looks like rendering my be the bottleneck
2011-04-14 10417, 2011
warp
real 0m0.084s
2011-04-14 10423, 2011
warp
(from hobbes to ngs.m.o)
2011-04-14 10437, 2011
ocharles
oh, i was reading the wrong time too, which doesn't help :)
2011-04-14 10417, 2011
warp
ocharles: well, we are rendering it properly now. I wouldn't be surprised if that is a bit slower compared to concatenating strings.
2011-04-14 10425, 2011
ocharles
yea
2011-04-14 10440, 2011
ocharles
there's gonna be a overhead from just calling a perl subroutine
2011-04-14 10448, 2011
ocharles
maybe rob will be interested in how much slower it is
2011-04-14 10414, 2011
ocharles
is an example of how ngs is slower, in this case by ~0.06s
2011-04-14 10452, 2011
warp
aka a 50% increase for that particular scenario.
2011-04-14 10437, 2011
ocharles
almost, yea
2011-04-14 10415, 2011
warp
on equivalent hardware?
2011-04-14 10421, 2011
ocharles
on better hardware
2011-04-14 10437, 2011
warp
so the actual difference could be much larger
2011-04-14 10458, 2011
warp
i.e. our code could be taking twice as long
2011-04-14 10422, 2011
ocharles
i mean it's that difference when I run the tests comparing the current solution, to the solution we will deploy in the future
2011-04-14 10437, 2011
warp
yeah.
2011-04-14 10442, 2011
warp
but if that hardware is much beefier, we may be able to find and fix some issues.
2011-04-14 10408, 2011
ocharles
my attempt at briefly profiling it didn't really show any single hotspot
2011-04-14 10456, 2011
warp
how would I go about get a cachegrind file for such a /ws/1 request?
you'll need to run that through nytprofcg to get a cache grind file
2011-04-14 10414, 2011
ocharles
(part of Devel::NYTProf)
2011-04-14 10430, 2011
ocharles
I have been preferring to use nytprofhtml and reading the html reports though
2011-04-14 10405, 2011
warp
makes sense, kcachegrind is a bit confusing.
2011-04-14 10418, 2011
warp
(but i'm somewhat used to it)
2011-04-14 10445, 2011
warp installs Devel::NYTProf
2011-04-14 10411, 2011
warp
Profile data incomplete, inflate error -5 ((null)) at end of input file, perhaps the process didn't exit cleanly or the file has been truncated (refer to TROUBLESHOOTING in the documentation)
2011-04-14 10444, 2011
ocharles
hrm, dunno what to do about that
2011-04-14 10449, 2011
ocharles
try the html reports
2011-04-14 10425, 2011
warp
same thing.
2011-04-14 10451, 2011
ocharles
ok, time to reprofile
2011-04-14 10404, 2011
warp
did the file upload correctly? it's only 307 bytes.
