#techempower-fwbm

brianhauer_: hey btw, did Matrix.org register on your radar ? http://matrix.org/blog/faq/

brianhauer_: IIRC it should fit some of the requirements you have of services like that

(sorry for being offtopic guys)

hamiltont: is this OK? https://github.com/lpereira/FrameworkBenchmarks...

LGTM!

I like the balance of “are you cheating”’s answer with the answer to “can I use this?”

:-P

:P

That blog post is likely to be really useful to F/W developers

If I’m reading this correctly, you’re basically shooting to create an upper limit on performance with lwan

I'm actually planning to make a talk out of it. I believe my spoken English got a lot better since the last time I tried to speak at a conference, so it's less likely I'll give a whole auditorium enough reason to take a nap.

Yea, it seems like you took an common exercise and really ran with it, nice

[FrameworkBenchmarks] lpereira opened pull request #1375: Add a FAQ to the Lwan benchmark. (master...update-lwan-readme) http://git.io/xxW3

[FrameworkBenchmarks] hamiltont pushed 2 new commits to master: http://git.io/xxW7

FrameworkBenchmarks/master bf2af09 Leandro Pereira: Add a FAQ to the Lwan benchmark.

FrameworkBenchmarks/master a4ec473 Hamilton Turner: Merge pull request #1375 from lpereira/update-lwan-readme...

hamiltont_away: :)

[FrameworkBenchmarks] methane opened pull request #1378: uwsgi: Stop using -d option (master...stop-uwsgi-daemon) http://git.io/xpIW

TE-msmith: the DB install log is apparently from a fresh VM. I'll run a new VM install to check if it's still working for me.

btw, the benchmark & CI installs are different for Cassandra. i've assumed a machine will be used only in one mode. running both on a machine may cause problems currently. run-ci.py should be changed to use the method from database.sh after R10.

created issue #1379, i plan to address this after R10.

i just tried installing a fresh dev VM with Vagrant and Cassandra installation seems to run ok in this test. please see: http://sprunge.us/iGDF

so, cassandra is still not "upgrading" appropriately for me on our peak hardware

any thoughts?

also, I'm still having trouble getting lapis running after an upgrade in version

` sudo luarocks install --server=http://rocks.moonscript.org lapis` successfully installs "lapis 1.1.0-1 is now built and installed in /usr/local/"

but when I run ` lapis server production` I get "/usr/bin/lua: cannot open /usr/local/lib/luarocks/rocks/lapis/1.0.6-1/bin/lapis: No such file or directory"

that makes me sad

there's a file /usr/local/lib/luarocks/rocks/manifest

that contains the information on all the installed rocks and where they live on the filesystem

mine says

lapis = {

"lapis/1.1.0-1", "lapis/1.0.5-1", "lapis/1.0.4-1", "lapis/1.0.3-1", "lapis/dev-1", "lapis/0.0.10-1", "lapis/0.0.9-1", "lapis/0.0.7-1", "lapis/0.0.2-1"

so my guess is that manifest is wrong for you

hmm

let me test

btw: it would be really great of lua and luarocks could be installed into the IROOT directory so that when I do a fresh run and I completely remove that dir, this problem magically goes away

okay, whiping out the lapis block from that rock file broke the installation (sudo luarocks install --server=http://rocks.moonscript.org lapis)

just wipe the entire file I guess

then it will reinstall

but should be quick

looking at the manpage I see luarocks has a parameter --to for destination, but I have no idea on how it works, so I guess I would have to look into it for next round

I have only ever used luarocks to install into /usr/local

it's still having issues and looking for an older version

ah, by default it can install into home..[[/.luarocks]]

maybe that's what happened here?

it installed inself into the homeuser dir?

for the benchmark user

no user has a luarocks dir in their home

nor does IROOT

I nuked the /usr/local-dir and the manifest locally, then I had to aptitude reinstall luarocks, that got me a fresh manifest

maybe that's a hard way out, atleast

i think I may have just missed some lapis in the manifest

trying again

that didn't work either - removed the manifest entirely

nope, that failed too

how do you force a reinstall?

try aptitude reinstall luarocks

aptitude is a desktop application

i'm using apt-get

apt-get --reinstall luarocks

(aptitude isn't a desktop application, it's just more high level than apt-get)

dash-dash... god I hate apt

FINALLY

okay

this needs to be addressed

lapis should be included in preview 3

thank you

np

sorry for the problems, I've never seen luarocks errors like that

I THINK we have an issue for this, but we NEED (in the future) to be able to have all of our installations sandboxed to the IROOT so that any user can delete that directory and be assured that there are no lingering tools that may differ from the actual runs (like in this example)

understandable

additionally, I THINK this would force us to move away from apt-get (yay), and as a side-effect would allow other *nix environments (RedHat, Gentoo, etc) to run the suite

(also yay)

well, sort of

but the big thing is "I want to be able to destroy the IROOT dir, and that should remove ALL the software installed and required by the benchmark suite"

a bunch of tools will have the same problem

like how do you install php's composer

and how do you install npm, and where does npm put its stuff

the correct solution here would be to use a cfg mgmt system from the start :-)

also, confirmed that lapis is running on peak, now

cool

php-composer is currently installed in iroot

npm appears to also be installed to iroot presently

ah, so they dont apt-get themselsves, but install from source?

for those types of softwares, they are usually configurable as to where the libraries they collect are stored (default seems to be some dot-dir under the user's home dir, but I have changed a NUMBER of them to use iroot: sbt, maven, etc)

correct

but even java installs itself as a package right now

it's a more arduous process to install from source, but ultimately a stronger one since we can 1) lock versions and 2) obliterate without side-effects to the system

ghc, hhvm, erlang, lua

yes

we need to move away from that

not now, but eventually

yeah

I want to compile from source; even if that takes 2 hours

(per install)

it really is a drop in the bucket to a run, and it should make our runs more precise AND I won't have to spend 2-3 preview runs fixing these sorts of issues

especially when many contributors don't idle this channel and can provide suggestions and help to get their framework running again

I just want to nuke the IROOT, kick off a preview, and have everything run as if it were a brand new machine

with that said... lunch

[FrameworkBenchmarks] LadyMozzarella opened pull request #1380: Remove Running PID if Present for All Java Play2 Variations (master...play2-java-remove-running-pid) http://git.io/pvxX

TE-msmith: what does the following command output on the DB server in peak env?

dpkg -l '*cassandra*'

is that a lower-case 'L' char ?

if so, dpkg-query: no packages found matching *cassandra*

yes, lowercase l.

hmm. did you try running db install in a fresh VM or just the Peak env?

does a either of these files exist? /var/log/cassandra/system.log, /ssd/log/cassandra/system.log ?

do they contain any errors?

peak

[FrameworkBenchmarks] marko-asplund opened pull request #1381: Set Cassandra log directory permissions during DB installation (master...cass-fixes) http://git.io/pfQM

i found a small bug in cassandra log directory permissions (see PR #1381). due to this the cassandra log is probably missing.

it should be found in /ssd/log/cassandra/system.log

not sure if this can cause cassandra startup to fail or not (doesn't in my env), but at least with this fixed the cass log should reveal something about the root cause.

TE-msmith: would it be possible to run db install with this fix and then check cass system.log ?

sorry

what do I need to od/

okay, pull your PR and test the db install?

yes, exactly. that would be great! thanks!

doing it now

okay, still failed

i'm looking for logs on the db machine?

does /ssd/log/cassandra/system.log exist?

hmmm, /ssd/log/cassandra/

empty

also, /ssd/cassandra is empty

"ls -la /ssd/log/cassandra /ssd/cassandra" ?

srsly empty

both

they have '.' and '..'

>_<

who owns the dirs?