Hedis hangs on reading big zset

[k-bx]

Here is a piece of code:

{-# LANGUAGE OverloadedStrings #-}

import Database.Redis
import System.Random
import qualified Data.ByteString.Char8 as BC8

main :: IO ()
main = do
  conn <- connect (defaultConnectInfo { connectHost="localhost" })

  stdGen <- getStdGen
  runRedis conn (fillBigZset stdGen 20000)

  resp <- runRedis conn $ zrangebyscoreLimit "bigzset" 0 1403194724 0 15000
  putStrLn $ "Response: " ++ show resp
  return ()

fillBigZset :: (RedisCtx m f, RandomGen t)
            => t -> Int -> m ()
fillBigZset _ 0 = return ()
fillBigZset gen n = do
    let (v, newGen) = randomR (1,1000000) gen
    _ <- zadd "bigzset" [(v, BC8.pack ("value-value-value-value-value-" ++ show v))]
    fillBigZset newGen (n - 1)

You can comment out 2 lines that fill redis after first time you run it. On my machine, this hangs with redis 2.6. Also tried with latest 2.8 one.

strace output: https://gist.github.com/k-bx/cd3cb178412ea9c50e71

[Davorak]

It looks like this bug is only present when the library is compiled with profiling on. I recompiled with profiling off and this bug went away.

The tip off to look examine the weather the profiling level was on or off was the repeated use of rt_sigreturn in the strace output which can be utilized when library profiling is turned on, or so I remember, unfortunately I can not find the link at the moment.

[k-bx]

Just tried GHC 7.8. Profiling-enabled hedis still goes into the loop. Profiling-disabled finishes it's work, but instead of result prints "RedisBug: ConnectionLost" twice,

strace for non-profiling version run:
.tar.xz (36Kb): https://dl.dropboxusercontent.com/u/865508/RedisBug.strace.stdout.tar.xz
.txt (1.8 Mb): https://dl.dropboxusercontent.com/u/865508/RedisBug.strace.stderr

[k-bx]

Update: played a bit with non-profiling version. Most of the time it does hang, sometimes it works good, and I saw this "ConnectionLost" only when running with strace.

[k-bx]

Seems part of hang is due to attoparsec bug haskell/attoparsec#70

Downgrading to 0.11.3.4 helps.

@Davorak mentioned that this doesn't fix all errors, since code from original report still fails on his machine with 0.11 versions of attoparsec producing output

KonRedisTest: <socket: 3>: hFlush: illegal operation (handle is closed)
KonRedisTest: <socket: 3>: hFlush: illegal operation (handle is closed)

[ozataman]

Also got bitten by this while using hedis and had to downgrade to < attoparsec 0.12. Is there a fix in sight? Specifically, rpop on a fairly large value hangs indefinitely.

[k-bx]

@ozataman fix is quite straight-forward -- fix attoparsec's behaviour and make a pull-request!

[ozataman]

Thanks, but we can’t all be solving all the problems we encounter in OSS. I was more interested in bumping the issue and finding out if anybody had anything in the works.
On Sep 4, 2014, at 3:26 AM, Konstantine Rybnikov notifications@github.com wrote:

@ozataman fix is quite straight-forward -- fix attoparsec's behaviour and make a pull-request!

—
Reply to this email directly or view it on GitHub.

[k-bx]

So, this is now working with latest attoparsec.