#techempower-fwbm

markoa: the benchmarker now runs all tests and installations using a dedicated user account

so to “stop” things we just kill any processes owned by that user

travis does impose VM limits, and the only option I know of is to try and reduce sbt’s memory usage

using stuff like this: http://stackoverflow.com/a/12484846/119592

According to this you have 3GB of memory available though: http://docs.travis-ci.com/user/ci-environment/#...

so….odd that SBT is *still* failing

So…what’s the hashtag we want to use if tweeting the TFB results?

#TFB, or #FrameworkBenchmarks, or #Round10, or #ItsHappening :-P

[FrameworkBenchmarks] hamiltont opened pull request #1482: Fix ASP.NET (master...pdonald-aspnet) http://git.io/veIlP

hamiltont: I'm using java 8 with sbt, so the process should not run out of permgen. I've also increased heap space to a value that's been a safe number in my environment.

i'm been banging my head against the wall with CI test failures in spray-es. I got it to pass once (http://bit.ly/1aqSyYL), but now tests are failing again.

it's the sbt build that's failing and the only explanation I can come up with is that some resource limits get exceeded and the process is terminated.

any ideas on how to solve this issue?

all those kernel output lines look... suspect

TE-msmith: kernel output lines from here - https://github.com/TechEmpower/FrameworkBenchma...

howdy

Still, there’s too many. We should mod that to write to a file then store it inside results

yo, morning

yeah

oh god... look at the work you've done

gh#1482 has SOOOO many changes

Fix ASP.NET : https://github.com/TechEmpower/FrameworkBenchma...

why!?

Hm, I can squash some of the commits, but most of them are real commits

oh ok

nm

it’s everything pdonald did plus the final changes we asked for in the original PR

sorry, this is covering all the MONO stuff

yup

and you've only done the `apt-get download` for that

good

Yea

Just becuase it was a huge PITA to get the proper binary without

I want, if it's possible, to not even use `apt-get download` because it would make our suite "flavorless" in terms of linux distros

I understand that

Their CI setup doesn’t *always* build the binary for our system, but the only way to know is to cycle through curl-ing each folder, which sucks

lol yeah

In terms of distros - why not just have *deb based and *rpm based like most major linux software vendors?

then we could rely (mostly) on the respective pacakge managers, and probably without ever requiring sudo

Would really help stuff like “download+compile python”

That's fine with me

I want to move towards something like that

and supporting deb first makes sense to me

BUT

Debian and Ubuntu (the two biggest deb-based distros) have slightly different package names

I tested this on Raspbian (same package names as Debian) and couldn't get all the `prerequisites.sh` to complete

additionally, java didn't get installed correctly, etc

lame, but I follow

`apt-get install openjdk-7-jdk` works on Ubuntu, but not Debian

which is totally annoying

So let’s support apt-get download for ubuntu for now, and then when we add some features to support different base OS we can add in debain first as it’ll be the easiest transition?

so, if we're going to do it that way, we need to 1) come up with a solution that cleanly can differentiate between *deb systems and *rpm systems, 2) come up with a way to get `.deb` files without relying on `apt`, and 3) make it completely user-doable

yeah

sounds good

I can add a big-ass comment warning that it won’t work on all debian in the source code

but what I want to avoid is something like `if (ubuntu) { apt-get install openjdk-7-jdk } else if (debian) { apt-get install oraclejdk-7 } else if ( gentoo ) { etc }`

warnings are good

but I think we just need to formalize our process for installation a bit

1) totally agree 2) why bother? Just say we’re using ubuntu package names for now and call it a day, since that’s fast and easy 3) you mean without sudo, right?

I LOVE what you did with the installs directory

Totally agreed - a huge ass if statement is …. undesirable

1) yay 2) for now using `apt-get download [package-name]` works, let's do it 3) yes

Cool

Thinking towards the future, perhaps something like mono.debian.sh vs mono.ubuntu12.sh, etc. Although TBH I’d rather use something that actualyl does this well, like puppet or chef (haven’t used either so I”m not aware if they meet our use case, but I’d look at them before writing custom code to do this)

I'll talk to Brittany when she gets in

Sweet

Don’t let me take up your morning, we can chat later

yeah, we should do some investigation to see what's already built

I love rebuilding wheels, but if there's already a good solution then it's cheaper than my time

no no, today is basically a wash for me

I have a half-day to help a friend set up his seder

so i'm catching up on emails, helping brian with the round 10 results, and doing r&d (read: tfb)

hamiltont: what is the difference between `export FOO=someVal` and simply calling `FOO=someVal` in bash?

if you say FOO=someVal, then you can use $FOO in your current script, but no subcommands will see that FOO variable

export means to export for subcommands / subshells

gotcha

so likely the `start.sh` scripts SHOULD use `export FOO=someVal`

yup

okay

if in doubt export

okay

i'm going to go over LadyMozzarella's "remove bash_profile" branch and audit the files a bit (since there are so many)

i'll check back here, but gunna nose-down for a bit

Coolio

\o/ http://bit.ly/1GoHjLX

you winning, mark?

nice

oh hamiltont: I think I'm going to try and tag the final commit number today

maybe :-) i'll just squash some commits, let's see what happens. fingers crossed :-)

depends on brian's response

Cool

I mean, I named the results in TFB-Round-10 as "final", so I think we're in at this point

but we have the commit hash, so I was going to play with tagging it as "round-10" on the FrameworkBenchmarks repo

thoughts?

Yea, I think that’s a good idea

yay

If you want to double dip, take a look at gh#1365 where I advocate tagging 4c67f02 as round-9

Retroactively tag rounds 8 and 9: https://github.com/TechEmpower/FrameworkBenchma...

markoa: yay! Sadly that’s the only way I know how to address the sbt failures in travis currently, just play with the memory heavy options until you get lucky...

ya

I'm kind of sad we didn't bring up the memory constraints on the call

>_<

I think we get 1GB now?

Actually, 3GB is what is claimed

really?

But I think it’s heavily dependant on the load...

... that should be PLENTY of RAM for a dependency manager

yeah

Yea, I know

Kind of suspicious of that 3gb number...

But - http://docs.travis-ci.com/user/ci-environment/#...

just wondering if we could get more verbose output in case of errors. i mean usually when the jvm runs out of memory i get a message indicating what happened.

or is the jvm just killed by some other process?

I think the first is an option, if there are options to tell sbt to print those kind of OOM errors we should enable them

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

FrameworkBenchmarks/master 5af5f37 marko asplund: implement test types 1, 2, 3, 5 and 6 with Scala, Spray + Elasticsearch...

FrameworkBenchmarks/master 902e19b Hamilton Turner: Merge pull request #1477 from marko-asplund/spray-es...

agreed

I think twiddling nobs on these package managers is something we should look into

steve's been spending a LOT of time trying to get php JUST right

bugger. now the test is failing again even though all I did was squash two of the latest commits :-(

"bugger" - you bri'ish?

no, not really ;-)

lol k

holy hell yesod’s build process is intense

My computer sounds like a train is whoosing by

the stdout output from install.sh is captured in the log. usually when i ran into memory exhaustion issues i get an error message either in stdout or stderr. but in this case i don't see anything like that in the log.

I shudder to think of building a real application

ya

So, does the build process get killed by some other process? that would explain why there's no error message in the log.

haskell does a LOT of work at compile time

What do you guys think about gh #1439 (aside from waiting for it to pass Travis)?

markoa: not sure

gh#1439

Increase throughput for Grizzly and undertow by parallelizing DB access: https://github.com/TechEmpower/FrameworkBenchma...

oh, is this the hibernate thing?

I'm not a hibernate expert, but if you can confirm that configuration value of 'false' in terms of caching, then I'm okay with it

markoa: FYI, you can use “TFB_VB_MEM=1024 vagrant up” to simulate a low-memory condition using vagrant and see if you get log output

markoa: I doublt it’s being killed by an external process on travis, I’d expect your mallocs just start to fail (and most software starts to segfault, but sbt should be fairly resistant to that happening)

LadyMozzarella: ^ I’m with TE-msmith on this, if that value does mean no cache then coolio

ok sweet. thanks

My confusion is all the Go pull requests - there seem to be multiple competing thoughts about what they should put into TFB and I’m totally lost on those

Been avoiding them