- Haskell-Cafe - Haskell.org

Long pauses / idle time in a Haskell TCP server
by Alex Iliev 08 Oct '12

08 Oct '12

Hello, I'm seeing long pauses in a server based on the 'scalable-server' package from Hackage, version 0.2.2 [1]. It normally performs very well, about 100-150 micro-secs client latency over the loopback interface, but a fair number of requests are much slower, 38-40 milli-secs, making the mean latency and throughput quite bad. The server is a simple version of a memory cache, with operations to "put" a key-value mapping, and to "get" the value for a key, using a data map. I've tried with the strict versions of 'unordered-containers' and the 'containers' package for the data map implementation, with similar results. I've tried it with ghc 7.4.1 and with 7.6.1 with similar results. Runtime info I got shows: - GC report: GC time is much less than the total time of the slow responses, so does not explain them. - CPU profile: Idle time is very high, and would explain the slow responses. By idle time I mean the part of wall time which is not included in the profile report's "total time". For example, when running a test for 5 minutes, the profile report has total time = 0.05 secs. 'top' output is consistent - quite low CPU usage during such a test. Apart from this there is nothing notable in the profile, most time is spent in the network input and output processing (network-enumerator etc.). One observation is that the pauses seem correlated with overwriting entries in the data map. If I'm mostly adding new keys (as well as reading), the throughput is much better, and the idle time in the CPU profile is lower (though the time spent in GC increases a lot). I've removed all concurrency from the program (including my version of scalable-server) to eliminate that factor, but the problem persists. I tried it on two fairly different Linux machines, with similar results. The client I use to benchmark is a simple C program, so I don't suspect it of causing the pauses. Looking at the timestamps on the network traffic confirms that the pauses are at the server. Any idea what might be causing the idle time? The long pauses are consistently 38-40ms, maybe that points to some aspect of CPU scheduling, leaving the program idle for some time? I can put the code on github if it would help. Many thanks! Alex [1] http://hackage.haskell.org/package/scalable-server-0.2.2

2 1

ANNOUNCE: test-framework-golden-1.1
by Roman Cheplyaka 08 Oct '12

08 Oct '12

I am glad to announce the first public release of test-framework-golden — a golden testing library. Hackage: http://hackage.haskell.org/package/test-framework-golden GitHub: https://github.com/feuerbach/test-framework-golden Golden tests are similar to unit tests (as implemented in HUnit), but the idea is to store the expected result in files (called «golden» files). I was introduced to the idea of golden testing by Bohdan Vlasyuk at ZuriHac in 2010. Since then I've discovered that this is a nice approach and it is already used in variety of projects (e.g. ghc, haddock). Surprisingly, not much is written about golden testing, and I've been unable to find any golden testing libraries for Haskell or any other programming language. Every project has its own ad-hoc implementation in Haskell/Python/Shell/etc. The closest match on the market is Simon Michael's shelltestrunner. But to use it you have to expose the tested functionality via command line, which may be inconvenient. So this is my attempt at a general golden testing library. It consists of two modules. Test.Golden has a simple interface that helps you get started very quickly. Test.Golden.Advanced provides a very generic testing function that you can use to implement the testing system you dream about. The library is integrated with test-framework, so you can use golden tests in addition to SmallCheck/QuickCheck/HUnit tests. In future there's a plan to support some golden test management capabilities. Roman

5 14

a parallel mapM?
by Greg Fitzgerald 07 Oct '12

07 Oct '12

I'm new to concurrent programming in Haskell. I'm looking for a drop-in replacement for 'mapM' to parallelize a set of independent IO operations. I hoped 'mapConcurrently' might be it, but I need something that will only spawn as many threads as I have CPUs available [1]. I also tried Control.Parallel.Strategies [2]. While that route works, I had to use unsafePerformIO. Considering that IO is for sequencing effects and my IO operation doesn't cause any side-effects (besides hogging a file handle), is this a proper use of unsafePerformIO? Attempt 1 -------------- import System.Process(readProcess) import Control.Concurrent.Async(mapConcurrently) main :: IO [String] main = mapConcurrently (\n -> readProcess "echo" ["test: " ++ show n] "") [0..1000] $ ghc --version The Glorious Glasgow Haskell Compilation System, version 7.6.1 $ runghc test.hs test.hs: runInteractiveProcess: pipe: Too many open files test.hs: runInteractiveProcess: pipe: Too many open files test.hs: runInteractiveProcess: pipe: Too many open files test.hs: runInteractiveProcess: pipe: Too many open files test.hs: runInteractiveProcess: pipe: Too many open files test.hs: runInteractiveProcess: pipe: Too many open files test.hs: runInteractiveProcess: pipe: Too many open files test.hs: runInteractiveProcess: pipe: Too many open files test.hs: runInteractiveProcess: pipe: Too many open files test.hs: echo: createProcess: resource exhausted (Too many open files) Attempt 2 -------------- import System.Process(readProcess) import Control.Parallel.Strategies(parMap, rpar) import System.IO.Unsafe(unsafePerformIO) main :: IO [String] main = myMapConcurrently (\n -> readProcess "echo" ["test: " ++ show n] "") [0..1000] where myMapConcurrently f = return . parMap rpar (unsafePerformIO . f) $ runghc test.hs > /dev/null && echo Success Success Thanks, Greg

7 8

Cabal dependencies
by José Lopes 06 Oct '12

06 Oct '12

Hello, I'm trying to understand Cabal dependencies. Why does the following situation happen? # cabal install xmobar --dry-run Resolving dependencies... In order, the following would be installed: parsec-3.1.3 (reinstall) changes: mtl-2.1.1 -> 2.0.1.0 xmobar-0.15 (new package) Warning: The following packages are likely to be broken by the reinstalls: texmath-0.6.1.1 regex-tdfa-1.1.8 Unixutils-1.50 network-2.4.0.1 HTTP-4000.2.5 network-2.3.0.14 sendfile-0.7.6 happstack-server-7.0.7 happstack-hsp-7.1.0 happstack-7.0.0 hslogger-1.1.5 citeproc-hs-0.3.4 acid-state-0.6.4 HTTP-4000.2.3 ltk-0.12.1.0 json-0.5 highlighting-kate-0.5.1 Use --force-reinstalls if you want to install anyway. Best regards, José -- José António Branquinho de Oliveira Lopes Instituto Superior Técnico Technical University of Lisbon

4 8

Am I the only one having problems with RWH?
by Janek S. 06 Oct '12

06 Oct '12

I began learning Haskell 9 months ago. I still consider myself a beginner, but I'm progressing towards more advanced concepts. I read scientific papers (simpler ones) and books about Haskell and functional programming. Right now I'm reading Pearls of Functional Algorithm Design, Introduction to Functional Programming, Implementation of Functional Programming Languages and Real World Haskell. RWH is causing me a lot of trouble though. This leads me to frustration because book covers rather basic material. I just spent another 1,5 hour reading chapter 10 again and trying to understand how presented parsing functions work. Even if I am barely able to grasp what is going on I feel that I wouldn't know how to write such code by myself. Am I the only one having such problems with RWH? Jan

4 4

Benchmarks for graph algorithms
by damodar kulkarni 06 Oct '12

06 Oct '12

Hi cafe, Where do Haskell and DPH stand in comparison to C/OpenMP w.r.t. graph algorithms? Are benchmarks for graph algorithms available for Haskell like the ones we find at http://www.graphanalysis.org/benchmark/? - Damodar

1 0

referential transparency? (for fixity of local operators)
by Johannes Waldmann 06 Oct '12

06 Oct '12

I was really surprised at the following: *Main> 1 + 2 * 3 7 *Main> ( \ (+) (*) -> 1 + 2 * 3 ) (+) (*) 9 because I was somehow assuming that either a) the Prelude fixities of the operators are kept b) or they are undefined, so the parser rejects. but the Haskell standard says "Any operator lacking a fixity declaration is assumed to be infixl 9". This really should be "infix 9"?

5 5

does try force evaluation of IO value
by s9gf4ult＠gmail.com 05 Oct '12

05 Oct '12

if i have some lazy io action, assuming (maction :: IO a) which may raise exceptions while running, will "try" force evaluation of "a" to determine is exception raised or not ? Sorry for my broken english.

2 5

Windows - Creating FreeGLUT DLL
by Thiago Negri 05 Oct '12

05 Oct '12

I'm trying OpenGL on Haskell and couldn't create the FreeGLUT DLL myself. I did follow this tutorial: http://netsuperbrain.com/blog/posts/freeglut-windows-hopengl-hglut/ But when I put my DLL on the folder of the binary, it complains about "initGlut". Using the DLL pointed by FreeGLUT official site, created by other person, works fine. I don't know what I did wrong. Anyone had this problem? Thanks.

1 0

Re: [Haskell-cafe] ANNOUNCE: test-framework-golden-1.1
by Roman Cheplyaka 05 Oct '12

05 Oct '12

* Janek S. <fremenzone(a)poczta.onet.pl> [2012-10-05 11:50:53+0200] > > Cool, looking forward to reading it! > Well, the post is already finished: > http://ics.p.lodz.pl/~stolarek/blog/2012/10/code-testing-in-haskell/ > I was just going to publish it and then your email came up on the list. > > > I hope you won't forget to cover SmallCheck in your article as well. > > Being also the maintainer of SmallCheck, I want it to steal some fame > > from QuickCheck :) > Sorry to disappoint you, but I did not mention SmallCheck. So far I'm relying on QuickCheck and > didn't feel like I need to look for other testing library. I might be wrong of course. There are some technical advantages to SmallCheck (determinism, no need to shrink etc.), but the main reason I prefer it is because it gives me more confidence. With quickcheck, I know that it generated 100 tests, but I've no idea what those tests are, and whether the RNG missed some important corner cases. With SmallCheck I know that it tried *all* cases up to certain depth, so it's much more reassuring. (Except for cases when the notion of depth isn't that informative, like floating-point numbers.) Roman

2 1