FAQ
I wrote a proxy checker recently, it sends hundreds of thousands of
requests to a url to check whether proxies are available. Here is a part
of my checker code, the check proxy function:

func checkProxy(proxy string, timeout int) bool {
proxyUrl, err := url.Parse("http://" + proxy)
httpClient := &http.Client{
Transport: &http.Transport{

Proxy: http.ProxyURL(proxyUrl),

Dial: func(netw, addr string) (net.Conn, error) {

deadline := time.Now().Add(time.Duration(timeout) * time.Second)

c, err := net.DialTimeout(netw, addr, time.Second*time.Duration(timeout))

if err != nil {

return nil, err

}

c.SetDeadline(deadline)

return c, nil

},

ResponseHeaderTimeout: time.Duration(timeout) * time.Second,

DisableKeepAlives: true,

},

}

req, err := http.NewRequest("GET", "http://www.urltocheck.com", nil)
req.Close = true
resp, err := httpClient.Do(req)

if err != nil {

return false

}

defer resp.Body.Close()
body, _ := ioutil.ReadAll(resp.Body)
if strings.Contains(string(body), "xxx") {

return true

} else {

return false

}

}
After import "net/http/pprof", and after an hours or two, i could see
these two writeLoop and readLoop goroutines keep growing, the longer i run
the checker, the more they are (could be more than ten thousands):

8054 @ 0x41a716 0x4080d4 0x407d22 0x488021 0x41a8e0
# 0x4080d4 selectgo+0x384 /usr/local/go/src/pkg/runtime/chan.c:996
# 0x407d22 runtime.selectgo+0x12 /usr/local/go/src/pkg/runtime/chan.c:840
# 0x488021 net/http.(*persistConn).writeLoop+0x271 /usr/local/go/src/pkg/net/http/transport.go:791

8030 @ 0x41a716 0x4072d2 0x407718 0x4879cf 0x41a8e0
# 0x4879cf net/http.(*persistConn).readLoop+0x68f /usr/local/go/src/pkg/net/http/transport.go:778


And my memory leaks so bad, I find a bug been reports two years ago
https://code.google.com/p/go/issues/detail?id=4531 (net/http: Transport
leaks goroutines when request.ContentLength is explicitly short) which had
been marked fixed, but I think somewhere someplace transport still leaking.
Anyone would help, thanks a lot!



--
You received this message because you are subscribed to the Google Groups "golang-nuts" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
For more options, visit https://groups.google.com/d/optout.

Search Discussions

  • Billy Shea at Mar 26, 2014 at 1:46 am
    My go version is go1.2.1 linux/amd64, on a centos 6.3 64 bit machine.
    On Tuesday, March 25, 2014 9:01:46 PM UTC+8, Billy Shea wrote:

    I wrote a proxy checker recently, it sends hundreds of thousands of
    requests to a url to check whether proxies are available. Here is a part
    of my checker code, the check proxy function:

    func checkProxy(proxy string, timeout int) bool {
    proxyUrl, err := url.Parse("http://" + proxy)
    httpClient := &http.Client{
    Transport: &http.Transport{

    Proxy: http.ProxyURL(proxyUrl),

    Dial: func(netw, addr string) (net.Conn, error) {

    deadline := time.Now().Add(time.Duration(timeout) * time.Second)

    c, err := net.DialTimeout(netw, addr, time.Second*time.Duration(timeout))

    if err != nil {

    return nil, err

    }

    c.SetDeadline(deadline)

    return c, nil

    },

    ResponseHeaderTimeout: time.Duration(timeout) * time.Second,

    DisableKeepAlives: true,

    },

    }

    req, err := http.NewRequest("GET", "http://www.urltocheck.com", nil)
    req.Close = true
    resp, err := httpClient.Do(req)

    if err != nil {

    return false

    }

    defer resp.Body.Close()
    body, _ := ioutil.ReadAll(resp.Body)
    if strings.Contains(string(body), "xxx") {

    return true

    } else {

    return false

    }

    }
    After import "net/http/pprof", and after an hours or two, i could see
    these two writeLoop and readLoop goroutines keep growing, the longer i run
    the checker, the more they are (could be more than ten thousands):

    8054 @ 0x41a716 0x4080d4 0x407d22 0x488021 0x41a8e0
    # 0x4080d4 selectgo+0x384 /usr/local/go/src/pkg/runtime/chan.c:996
    # 0x407d22 runtime.selectgo+0x12 /usr/local/go/src/pkg/runtime/chan.c:840
    # 0x488021 net/http.(*persistConn).writeLoop+0x271 /usr/local/go/src/pkg/net/http/transport.go:791

    8030 @ 0x41a716 0x4072d2 0x407718 0x4879cf 0x41a8e0
    # 0x4879cf net/http.(*persistConn).readLoop+0x68f /usr/local/go/src/pkg/net/http/transport.go:778


    And my memory leaks so bad, I find a bug been reports two years ago
    https://code.google.com/p/go/issues/detail?id=4531 (net/http: Transport
    leaks goroutines when request.ContentLength is explicitly short) which had
    been marked fixed, but I think somewhere someplace transport still
    leaking. Anyone would help, thanks a lot!

    --
    You received this message because you are subscribed to the Google Groups "golang-nuts" group.
    To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
    For more options, visit https://groups.google.com/d/optout.
  • James Bardin at Mar 26, 2014 at 3:00 am
    You do realize that the number on the left is the goroutine ID, and not the number of currently running goroutines? (since your stack trace does only show 2)

    --
    You received this message because you are subscribed to the Google Groups "golang-nuts" group.
    To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
    For more options, visit https://groups.google.com/d/optout.
  • Billy Shea at Mar 26, 2014 at 3:05 am
    It's not the goroutine ID, it's the number of goroutines. I'm sure about
    that, i create about 10000 goroutines, but the stack trace shows there are
    total 36000+ goroutines. The track url is debug/pprof/goroutine?debug=1, if
    I switch to debug/pprof/goroutine?debug=2, I can see the full goroutine
    stack dump, it's too long too list...
    On Wednesday, March 26, 2014 11:00:01 AM UTC+8, James Bardin wrote:

    You do realize that the number on the left is the goroutine ID, and not
    the number of currently running goroutines? (since your stack trace does
    only show 2)
    --
    You received this message because you are subscribed to the Google Groups "golang-nuts" group.
    To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
    For more options, visit https://groups.google.com/d/optout.
  • James Bardin at Mar 26, 2014 at 3:13 am
    Oops, that may be the case (hard to confirm on mobile :) )

    You might want to try tip just to be sure, there's been some work done around http client and keepalive connections.

    --
    You received this message because you are subscribed to the Google Groups "golang-nuts" group.
    To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
    For more options, visit https://groups.google.com/d/optout.
  • Billy Shea at Mar 26, 2014 at 3:18 am
    Yes, I've set DisableKeepAlives to true to disable keep alive connections,
    and set req.Close = true, but it won't work, the goroutines keep growing.
    I'll try the tip.

    在 2014年3月26日星期三UTC+8上午11时13分09秒,James Bardin写道:
    Oops, that may be the case (hard to confirm on mobile :) )

    You might want to try tip just to be sure, there's been some work done
    around http client and keepalive connections.
    --
    You received this message because you are subscribed to the Google Groups "golang-nuts" group.
    To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
    For more options, visit https://groups.google.com/d/optout.
  • James Bardin at Mar 26, 2014 at 1:25 pm
    The only way I can imagine this leaking is that you occasionally get a
    short response, with a missing or partial body, and the remote server isn't
    closing the connection (I've seen this more often that you'd would think).

    Try setting an overall timeout on the entire process. If you're trying tip,
    in go1.3 the http.Client has a Timeout parameter which can handle this.
    With the current release you can
    try http://godoc.org/github.com/mreiferson/go-httpclient, which implements
    basically the same thing, but in Transport.RequestTimeout.


    On Tuesday, March 25, 2014 11:18:45 PM UTC-4, Billy Shea wrote:

    Yes, I've set DisableKeepAlives to true to disable keep alive connections,
    and set req.Close = true, but it won't work, the goroutines keep growing.
    I'll try the tip.

    在 2014年3月26日星期三UTC+8上午11时13分09秒,James Bardin写道:
    Oops, that may be the case (hard to confirm on mobile :) )

    You might want to try tip just to be sure, there's been some work done
    around http client and keepalive connections.
    --
    You received this message because you are subscribed to the Google Groups "golang-nuts" group.
    To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
    For more options, visit https://groups.google.com/d/optout.
  • Billy Shea at Mar 26, 2014 at 3:25 pm
    Thanks a lot, I've tried the tip, the leak becomes slower, but it's still
    leaking. I'll try the package you recommend, thanks for your help.

    在 2014年3月26日星期三UTC+8下午9时25分13秒,James Bardin写道:

    The only way I can imagine this leaking is that you occasionally get a
    short response, with a missing or partial body, and the remote server isn't
    closing the connection (I've seen this more often that you'd would think).

    Try setting an overall timeout on the entire process. If you're trying
    tip, in go1.3 the http.Client has a Timeout parameter which can handle
    this. With the current release you can try
    http://godoc.org/github.com/mreiferson/go-httpclient<http://www.google.com/url?q=http%3A%2F%2Fgodoc.org%2Fgithub.com%2Fmreiferson%2Fgo-httpclient&sa=D&sntz=1&usg=AFQjCNFY635FDau2-cLIGGhKdy1u53SOGw>,
    which implements basically the same thing, but in Transport.RequestTimeout.


    On Tuesday, March 25, 2014 11:18:45 PM UTC-4, Billy Shea wrote:

    Yes, I've set DisableKeepAlives to true to disable keep alive
    connections, and set req.Close = true, but it won't work, the goroutines
    keep growing. I'll try the tip.

    在 2014年3月26日星期三UTC+8上午11时13分09秒,James Bardin写道:
    Oops, that may be the case (hard to confirm on mobile :) )

    You might want to try tip just to be sure, there's been some work done
    around http client and keepalive connections.
    --
    You received this message because you are subscribed to the Google Groups "golang-nuts" group.
    To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
    For more options, visit https://groups.google.com/d/optout.
  • Julian at Apr 5, 2014 at 9:11 am
    Hi,

    I'm having the exact same problem. I have a program that creates many many
    http requests using proxies (via Transport.Proxy).
    After I stop a batch of 1000 requests I do a stack trace I have a lot of
    goroutines like these, which never finish:

    goroutine 3317 [chan receive]:
    net/http.(*persistConn).readLoop(0xc216c06680)
      /usr/local/go/src/pkg/net/http/transport.go:778 +0x68f
    created by net/http.(*Transport).dialConn
      /usr/local/go/src/pkg/net/http/transport.go:528 +0x607

    goroutine 2908 [select]:
    net/http.(*persistConn).writeLoop(0xc216a68880)
      /usr/local/go/src/pkg/net/http/transport.go:791 +0x271
    created by net/http.(*Transport).dialConn
      /usr/local/go/src/pkg/net/http/transport.go:529 +0x61e

    goroutine 3318 [select]:
    net/http.(*persistConn).writeLoop(0xc216c06680)
      /usr/local/go/src/pkg/net/http/transport.go:791 +0x271
    created by net/http.(*Transport).dialConn
      /usr/local/go/src/pkg/net/http/transport.go:529 +0x61e

    goroutine 3513 [select]:
    net/http.(*persistConn).writeLoop(0xc21627d180)
      /usr/local/go/src/pkg/net/http/transport.go:791 +0x271
    created by net/http.(*Transport).dialConn
      /usr/local/go/src/pkg/net/http/transport.go:529 +0x61e

    goroutine 2205 [chan receive]:
    net/http.(*persistConn).readLoop(0xc213a08b80)
      /usr/local/go/src/pkg/net/http/transport.go:778 +0x68f
    created by net/http.(*Transport).dialConn
      /usr/local/go/src/pkg/net/http/transport.go:528 +0x607

    goroutine 2206 [select]:
    net/http.(*persistConn).writeLoop(0xc213a08b80)
      /usr/local/go/src/pkg/net/http/transport.go:791 +0x271
    created by net/http.(*Transport).dialConn
      /usr/local/go/src/pkg/net/http/transport.go:529 +0x61e

    goroutine 3391 [chan receive]:
    net/http.(*persistConn).readLoop(0xc216a3cc00)
      /usr/local/go/src/pkg/net/http/transport.go:778 +0x68f
    created by net/http.(*Transport).dialConn
      /usr/local/go/src/pkg/net/http/transport.go:528 +0x607

    goroutine 3392 [select]:
    net/http.(*persistConn).writeLoop(0xc216a3cc00)
      /usr/local/go/src/pkg/net/http/transport.go:791 +0x271
    created by net/http.(*Transport).dialConn
      /usr/local/go/src/pkg/net/http/transport.go:529 +0x61e


    I'm using go 1.2.1 on ubuntu/amd64 and almost the same code as the OP:
    disable keep alive, custom dial func with dial timeout and conn deadline.
    ioutil.ReadAll would also hang sometimes but I found a workaround using
    io.Copy and a timer that runs resp.Body.Close() after a specified timeout.
    Now I just have leaky goroutines from the http/transport.go read/write
    loops.

    --
    You received this message because you are subscribed to the Google Groups "golang-nuts" group.
    To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
    For more options, visit https://groups.google.com/d/optout.
  • Billy Shea at Apr 5, 2014 at 1:08 pm
    I've tried this client http://godoc.org/github.com/mreiferson/go-httpclient,
    the leaking continues..., here's my new transport, i've set as many
    timeouts as i could :(

    transport := &httpclient.Transport{
    Proxy: http.ProxyURL(proxyUrl),

    Dial: func(netw, addr string) (net.Conn, error) {

    deadline := time.Now().Add(time.Duration(timeout) * time.Second)

    c, err := net.DialTimeout(netw, addr, time.Second*time.Duration(timeout))

    if err != nil {

    return nil, err

    }

    c.SetDeadline(deadline)

    return c, nil

    },

    ResponseHeaderTimeout: time.Duration(timeout) * time.Second,

    DisableKeepAlives: true,

    ConnectTimeout: time.Duration(timeout) * time.Second,

    RequestTimeout: time.Duration(timeout) * time.Second * 2,

    ReadWriteTimeout: time.Duration(timeout) * time.Second,

    }
    But my proxy checker keeps leaking goroutines as always, this has trapped
    me over a month. Does http transport has some send request rate limit, i
    can see thousands net.runtime_pollWait goroutines like this:

    4421 @ 0x416549 0x429630 0x428e9d 0x44ebd6 0x44eca2 0x451826 0x460cc9 0x53e48e 0x53e389 0x499646 0x4167e0
    # 0x429630 netpollblock+0x130 /usr/local/go/src/pkg/runtime/netpoll.goc:349
    # 0x428e9d net.runtime_pollWait+0x5d /usr/local/go/src/pkg/runtime/netpoll.goc:146
    # 0x44ebd6 net.(*pollDesc).Wait+0x46 /usr/local/go/src/pkg/net/fd_poll_runtime.go:84
    # 0x44eca2 net.(*pollDesc).WaitWrite+0x42 /usr/local/go/src/pkg/net/fd_poll_runtime.go:93
    # 0x451826 net.(*netFD).Write+0x466 /usr/local/go/src/pkg/net/fd_unix.go:325
    # 0x460cc9 net.(*conn).Write+0xe9 /usr/local/go/src/pkg/net/net.go:130
    # 0x53e48e bufio.(*Writer).flush+0xde /usr/local/go/src/pkg/bufio/bufio.go:501
    # 0x53e389 bufio.(*Writer).Flush+0x39 /usr/local/go/src/pkg/bufio/bufio.go:490
    # 0x499646 net/http.(*persistConn).writeLoop+0x206 /usr/local/go/src/pkg/net/http/transport.go:890

    4445 @ 0x416549 0x429630 0x428e9d 0x44ebd6 0x44ec42 0x4504d2 0x460ba9 0x49a7ba 0x4a38df 0x53c573 0x53c75d 0x498a16 0x4167e0
    # 0x429630 netpollblock+0x130 /usr/local/go/src/pkg/runtime/netpoll.goc:349
    # 0x428e9d net.runtime_pollWait+0x5d /usr/local/go/src/pkg/runtime/netpoll.goc:146
    # 0x44ebd6 net.(*pollDesc).Wait+0x46 /usr/local/go/src/pkg/net/fd_poll_runtime.go:84
    # 0x44ec42 net.(*pollDesc).WaitRead+0x42 /usr/local/go/src/pkg/net/fd_poll_runtime.go:89
    # 0x4504d2 net.(*netFD).Read+0x332 /usr/local/go/src/pkg/net/fd_unix.go:232
    # 0x460ba9 net.(*conn).Read+0xe9 /usr/local/go/src/pkg/net/net.go:122
    # 0x49a7ba net/http.noteEOFReader.Read+0x7a /usr/local/go/src/pkg/net/http/transport.go:1148
    # 0x4a38df net/http.(*noteEOFReader).Read+0xdf /usr/local/go/src/pkg/net/http/chunked.go:1
    # 0x53c573 bufio.(*Reader).fill+0x143 /usr/local/go/src/pkg/bufio/bufio.go:92
    # 0x53c75d bufio.(*Reader).Peek+0x11d /usr/local/go/src/pkg/bufio/bufio.go:120
    # 0x498a16 net/http.(*persistConn).readLoop+0xd6 /usr/local/go/src/pkg/net/http/transport.go:770
    Do I create too many goroutines that exceed the net package could handle
    which breaks the transport? I currently has 8500 goroutines that keep
    sending http request via proxies with a producer/consumer concurrency
    pattern:
    go addJobs(jobs, results)

    for i := 0; i < workers; i++ {

    go doJobs(jobs, timeout)

    }
    Anyone else has the same problem? Need your help!

    -Billy


    在 2014年4月5日星期六UTC+8下午5时11分21秒,Julian写道:
    Hi,

    I'm having the exact same problem. I have a program that creates many many
    http requests using proxies (via Transport.Proxy).
    After I stop a batch of 1000 requests I do a stack trace I have a lot of
    goroutines like these, which never finish:

    goroutine 3317 [chan receive]:
    net/http.(*persistConn).readLoop(0xc216c06680)
    /usr/local/go/src/pkg/net/http/transport.go:778 +0x68f
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:528 +0x607

    goroutine 2908 [select]:
    net/http.(*persistConn).writeLoop(0xc216a68880)
    /usr/local/go/src/pkg/net/http/transport.go:791 +0x271
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:529 +0x61e

    goroutine 3318 [select]:
    net/http.(*persistConn).writeLoop(0xc216c06680)
    /usr/local/go/src/pkg/net/http/transport.go:791 +0x271
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:529 +0x61e

    goroutine 3513 [select]:
    net/http.(*persistConn).writeLoop(0xc21627d180)
    /usr/local/go/src/pkg/net/http/transport.go:791 +0x271
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:529 +0x61e

    goroutine 2205 [chan receive]:
    net/http.(*persistConn).readLoop(0xc213a08b80)
    /usr/local/go/src/pkg/net/http/transport.go:778 +0x68f
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:528 +0x607

    goroutine 2206 [select]:
    net/http.(*persistConn).writeLoop(0xc213a08b80)
    /usr/local/go/src/pkg/net/http/transport.go:791 +0x271
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:529 +0x61e

    goroutine 3391 [chan receive]:
    net/http.(*persistConn).readLoop(0xc216a3cc00)
    /usr/local/go/src/pkg/net/http/transport.go:778 +0x68f
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:528 +0x607

    goroutine 3392 [select]:
    net/http.(*persistConn).writeLoop(0xc216a3cc00)
    /usr/local/go/src/pkg/net/http/transport.go:791 +0x271
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:529 +0x61e


    I'm using go 1.2.1 on ubuntu/amd64 and almost the same code as the OP:
    disable keep alive, custom dial func with dial timeout and conn deadline.
    ioutil.ReadAll would also hang sometimes but I found a workaround using
    io.Copy and a timer that runs resp.Body.Close() after a specified timeout.
    Now I just have leaky goroutines from the http/transport.go read/write
    loops.
    --
    You received this message because you are subscribed to the Google Groups "golang-nuts" group.
    To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
    For more options, visit https://groups.google.com/d/optout.
  • Robert Melton at Apr 5, 2014 at 1:30 pm
    In the testing around this issue -- are you controlling both sides?
    IE: A known well behaving server you can interrogate for information?
    Would be curious if you wrote a server in Go for it, and the client
    and confirmed the behavior. Then you could also give people a fully
    working test case to play with and reproduce the errors.
    On Sat, Apr 5, 2014 at 9:07 AM, Billy Shea wrote:
    I've tried this client http://godoc.org/github.com/mreiferson/go-httpclient,
    the leaking continues..., here's my new transport, i've set as many timeouts
    as i could :(
    transport := &httpclient.Transport{

    Proxy: http.ProxyURL(proxyUrl),

    Dial: func(netw, addr string) (net.Conn, error) {

    deadline := time.Now().Add(time.Duration(timeout) * time.Second)

    c, err := net.DialTimeout(netw, addr, time.Second*time.Duration(timeout))

    if err != nil {

    return nil, err

    }

    c.SetDeadline(deadline)

    return c, nil

    },

    ResponseHeaderTimeout: time.Duration(timeout) * time.Second,

    DisableKeepAlives: true,

    ConnectTimeout: time.Duration(timeout) * time.Second,

    RequestTimeout: time.Duration(timeout) * time.Second * 2,

    ReadWriteTimeout: time.Duration(timeout) * time.Second,

    }

    But my proxy checker keeps leaking goroutines as always, this has trapped me
    over a month. Does http transport has some send request rate limit, i can
    see thousands net.runtime_pollWait goroutines like this:
    4421 @ 0x416549 0x429630 0x428e9d 0x44ebd6 0x44eca2 0x451826 0x460cc9
    0x53e48e 0x53e389 0x499646 0x4167e0

    # 0x429630 netpollblock+0x130 /usr/local/go/src/pkg/runtime/netpoll.goc:349

    # 0x428e9d net.runtime_pollWait+0x5d /usr/local/go/src/pkg/runtime/netpoll.goc:146

    # 0x44ebd6 net.(*pollDesc).Wait+0x46 /usr/local/go/src/pkg/net/fd_poll_runtime.go:84

    # 0x44eca2 net.(*pollDesc).WaitWrite+0x42 /usr/local/go/src/pkg/net/fd_poll_runtime.go:93

    # 0x451826 net.(*netFD).Write+0x466 /usr/local/go/src/pkg/net/fd_unix.go:325
    # 0x460cc9 net.(*conn).Write+0xe9 /usr/local/go/src/pkg/net/net.go:130

    # 0x53e48e bufio.(*Writer).flush+0xde /usr/local/go/src/pkg/bufio/bufio.go:501

    # 0x53e389 bufio.(*Writer).Flush+0x39 /usr/local/go/src/pkg/bufio/bufio.go:490

    # 0x499646 net/http.(*persistConn).writeLoop+0x206 /usr/local/go/src/pkg/net/http/transport.go:890

    4445 @ 0x416549 0x429630 0x428e9d 0x44ebd6 0x44ec42 0x4504d2 0x460ba9
    0x49a7ba 0x4a38df 0x53c573 0x53c75d 0x498a16 0x4167e0

    # 0x429630 netpollblock+0x130 /usr/local/go/src/pkg/runtime/netpoll.goc:349

    # 0x428e9d net.runtime_pollWait+0x5d /usr/local/go/src/pkg/runtime/netpoll.goc:146

    # 0x44ebd6 net.(*pollDesc).Wait+0x46 /usr/local/go/src/pkg/net/fd_poll_runtime.go:84

    # 0x44ec42 net.(*pollDesc).WaitRead+0x42 /usr/local/go/src/pkg/net/fd_poll_runtime.go:89

    # 0x4504d2 net.(*netFD).Read+0x332 /usr/local/go/src/pkg/net/fd_unix.go:232
    # 0x460ba9 net.(*conn).Read+0xe9 /usr/local/go/src/pkg/net/net.go:122

    # 0x49a7ba net/http.noteEOFReader.Read+0x7a /usr/local/go/src/pkg/net/http/transport.go:1148

    # 0x4a38df net/http.(*noteEOFReader).Read+0xdf /usr/local/go/src/pkg/net/http/chunked.go:1

    # 0x53c573 bufio.(*Reader).fill+0x143 /usr/local/go/src/pkg/bufio/bufio.go:92

    # 0x53c75d bufio.(*Reader).Peek+0x11d /usr/local/go/src/pkg/bufio/bufio.go:120

    # 0x498a16 net/http.(*persistConn).readLoop+0xd6 /usr/local/go/src/pkg/net/http/transport.go:770

    Do I create too many goroutines that exceed the net package could handle
    which breaks the transport? I currently has 8500 goroutines that keep
    sending http request via proxies with a producer/consumer concurrency
    pattern:
    go addJobs(jobs, results)

    for i := 0; i < workers; i++ {

    go doJobs(jobs, timeout)

    }

    Anyone else has the same problem? Need your help!

    -Billy


    在 2014年4月5日星期六UTC+8下午5时11分21秒,Julian写道:
    Hi,

    I'm having the exact same problem. I have a program that creates many many
    http requests using proxies (via Transport.Proxy).
    After I stop a batch of 1000 requests I do a stack trace I have a lot of
    goroutines like these, which never finish:

    goroutine 3317 [chan receive]:
    net/http.(*persistConn).readLoop(0xc216c06680)
    /usr/local/go/src/pkg/net/http/transport.go:778 +0x68f
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:528 +0x607

    goroutine 2908 [select]:
    net/http.(*persistConn).writeLoop(0xc216a68880)
    /usr/local/go/src/pkg/net/http/transport.go:791 +0x271
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:529 +0x61e

    goroutine 3318 [select]:
    net/http.(*persistConn).writeLoop(0xc216c06680)
    /usr/local/go/src/pkg/net/http/transport.go:791 +0x271
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:529 +0x61e

    goroutine 3513 [select]:
    net/http.(*persistConn).writeLoop(0xc21627d180)
    /usr/local/go/src/pkg/net/http/transport.go:791 +0x271
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:529 +0x61e

    goroutine 2205 [chan receive]:
    net/http.(*persistConn).readLoop(0xc213a08b80)
    /usr/local/go/src/pkg/net/http/transport.go:778 +0x68f
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:528 +0x607

    goroutine 2206 [select]:
    net/http.(*persistConn).writeLoop(0xc213a08b80)
    /usr/local/go/src/pkg/net/http/transport.go:791 +0x271
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:529 +0x61e

    goroutine 3391 [chan receive]:
    net/http.(*persistConn).readLoop(0xc216a3cc00)
    /usr/local/go/src/pkg/net/http/transport.go:778 +0x68f
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:528 +0x607

    goroutine 3392 [select]:
    net/http.(*persistConn).writeLoop(0xc216a3cc00)
    /usr/local/go/src/pkg/net/http/transport.go:791 +0x271
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:529 +0x61e


    I'm using go 1.2.1 on ubuntu/amd64 and almost the same code as the OP:
    disable keep alive, custom dial func with dial timeout and conn deadline.
    ioutil.ReadAll would also hang sometimes but I found a workaround using
    io.Copy and a timer that runs resp.Body.Close() after a specified timeout.
    Now I just have leaky goroutines from the http/transport.go read/write
    loops.
    --
    You received this message because you are subscribed to the Google Groups
    "golang-nuts" group.
    To unsubscribe from this group and stop receiving emails from it, send an
    email to [email protected].
    For more options, visit https://groups.google.com/d/optout.


    --
    Robert Melton | http://robertmelton.com

    --
    You received this message because you are subscribed to the Google Groups "golang-nuts" group.
    To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
    For more options, visit https://groups.google.com/d/optout.
  • Ucgggg at Apr 5, 2014 at 3:45 pm
    There is a bug in "net\http\client.go".
    I created an issues: http://code.google.com/p/go/issues/detail?id=7620.
    Just modify "net\http\client.go":
    ------------------------------------------
      if err != nil {
       if resp != nil {
    +++++ if resp.Body != nil {
    +++++ resp.Body.Close()
    +++++ }
        log.Printf("RoundTripper returned a response & error; ignoring response")
       }
       return nil, err
      }
    ------------------------------------------
    Why nobody deal with the issus since I reported it weeks ago?


    在 2014年3月25日星期二UTC+8下午9时01分46秒,Billy Shea写道:
    I wrote a proxy checker recently, it sends hundreds of thousands of
    requests to a url to check whether proxies are available. Here is a part
    of my checker code, the check proxy function:

    func checkProxy(proxy string, timeout int) bool {
    proxyUrl, err := url.Parse("http://" + proxy)
    httpClient := &http.Client{
    Transport: &http.Transport{

    Proxy: http.ProxyURL(proxyUrl),

    Dial: func(netw, addr string) (net.Conn, error) {

    deadline := time.Now().Add(time.Duration(timeout) * time.Second)

    c, err := net.DialTimeout(netw, addr, time.Second*time.Duration(timeout))

    if err != nil {

    return nil, err

    }

    c.SetDeadline(deadline)

    return c, nil

    },

    ResponseHeaderTimeout: time.Duration(timeout) * time.Second,

    DisableKeepAlives: true,

    },

    }

    req, err := http.NewRequest("GET", "http://www.urltocheck.com", nil)
    req.Close = true
    resp, err := httpClient.Do(req)

    if err != nil {

    return false

    }

    defer resp.Body.Close()
    body, _ := ioutil.ReadAll(resp.Body)
    if strings.Contains(string(body), "xxx") {

    return true

    } else {

    return false

    }

    }
    After import "net/http/pprof", and after an hours or two, i could see
    these two writeLoop and readLoop goroutines keep growing, the longer i run
    the checker, the more they are (could be more than ten thousands):

    8054 @ 0x41a716 0x4080d4 0x407d22 0x488021 0x41a8e0
    # 0x4080d4 selectgo+0x384 /usr/local/go/src/pkg/runtime/chan.c:996
    # 0x407d22 runtime.selectgo+0x12 /usr/local/go/src/pkg/runtime/chan.c:840
    # 0x488021 net/http.(*persistConn).writeLoop+0x271 /usr/local/go/src/pkg/net/http/transport.go:791

    8030 @ 0x41a716 0x4072d2 0x407718 0x4879cf 0x41a8e0
    # 0x4879cf net/http.(*persistConn).readLoop+0x68f /usr/local/go/src/pkg/net/http/transport.go:778


    And my memory leaks so bad, I find a bug been reports two years ago
    https://code.google.com/p/go/issues/detail?id=4531 (net/http: Transport
    leaks goroutines when request.ContentLength is explicitly short) which had
    been marked fixed, but I think somewhere someplace transport still
    leaking. Anyone would help, thanks a lot!

    --
    You received this message because you are subscribed to the Google Groups "golang-nuts" group.
    To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
    For more options, visit https://groups.google.com/d/optout.
  • Carlos Castillo at Apr 6, 2014 at 6:43 am
    RTFM: http://golang.org/doc/contribute.html
    On Saturday, April 5, 2014 4:41:39 AM UTC-7, [email protected] wrote:

    There is a bug in "net\http\client.go".
    I created an issues: http://code.google.com/p/go/issues/detail?id=7620.
    Just modify "net\http\client.go":
    ------------------------------------------
    if err != nil {
    if resp != nil {
    +++++ if resp.Body != nil {
    +++++ resp.Body.Close()
    +++++ }
    log.Printf("RoundTripper returned a response & error; ignoring response")
    }
    return nil, err
    }
    ------------------------------------------
    Why nobody deal with the issus since I reported it weeks ago?


    在 2014年3月25日星期二UTC+8下午9时01分46秒,Billy Shea写道:
    I wrote a proxy checker recently, it sends hundreds of thousands of
    requests to a url to check whether proxies are available. Here is a part
    of my checker code, the check proxy function:

    func checkProxy(proxy string, timeout int) bool {
    proxyUrl, err := url.Parse("http://" + proxy)
    httpClient := &http.Client{
    Transport: &http.Transport{

    Proxy: http.ProxyURL(proxyUrl),

    Dial: func(netw, addr string) (net.Conn, error) {

    deadline := time.Now().Add(time.Duration(timeout) * time.Second)

    c, err := net.DialTimeout(netw, addr, time.Second*time.Duration(timeout))

    if err != nil {

    return nil, err

    }

    c.SetDeadline(deadline)

    return c, nil

    },

    ResponseHeaderTimeout: time.Duration(timeout) * time.Second,

    DisableKeepAlives: true,

    },

    }

    req, err := http.NewRequest("GET", "http://www.urltocheck.com", nil)
    req.Close = true
    resp, err := httpClient.Do(req)

    if err != nil {

    return false

    }

    defer resp.Body.Close()
    body, _ := ioutil.ReadAll(resp.Body)
    if strings.Contains(string(body), "xxx") {

    return true

    } else {

    return false

    }

    }
    After import "net/http/pprof", and after an hours or two, i could see
    these two writeLoop and readLoop goroutines keep growing, the longer i run
    the checker, the more they are (could be more than ten thousands):

    8054 @ 0x41a716 0x4080d4 0x407d22 0x488021 0x41a8e0
    # 0x4080d4 selectgo+0x384 /usr/local/go/src/pkg/runtime/chan.c:996
    # 0x407d22 runtime.selectgo+0x12 /usr/local/go/src/pkg/runtime/chan.c:840
    # 0x488021 net/http.(*persistConn).writeLoop+0x271 /usr/local/go/src/pkg/net/http/transport.go:791

    8030 @ 0x41a716 0x4072d2 0x407718 0x4879cf 0x41a8e0
    # 0x4879cf net/http.(*persistConn).readLoop+0x68f /usr/local/go/src/pkg/net/http/transport.go:778


    And my memory leaks so bad, I find a bug been reports two years ago
    https://code.google.com/p/go/issues/detail?id=4531 (net/http: Transport
    leaks goroutines when request.ContentLength is explicitly short) which had
    been marked fixed, but I think somewhere someplace transport still
    leaking. Anyone would help, thanks a lot!

    --
    You received this message because you are subscribed to the Google Groups "golang-nuts" group.
    To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
    For more options, visit https://groups.google.com/d/optout.
  • Billy Shea at Apr 7, 2014 at 12:42 am
    This patch solved my problem
    exactly https://codereview.appspot.com/84850043, the goroutines now stop
    leaking! Cheers, special thanks to ucgggg and snaury.

    在 2014年4月5日星期六UTC+8下午7时41分39秒,[email protected]写道:
    There is a bug in "net\http\client.go".
    I created an issues: http://code.google.com/p/go/issues/detail?id=7620.
    Just modify "net\http\client.go":
    ------------------------------------------
    if err != nil {
    if resp != nil {
    +++++ if resp.Body != nil {
    +++++ resp.Body.Close()
    +++++ }
    log.Printf("RoundTripper returned a response & error; ignoring response")
    }
    return nil, err
    }
    ------------------------------------------
    Why nobody deal with the issus since I reported it weeks ago?


    在 2014年3月25日星期二UTC+8下午9时01分46秒,Billy Shea写道:
    I wrote a proxy checker recently, it sends hundreds of thousands of
    requests to a url to check whether proxies are available. Here is a part
    of my checker code, the check proxy function:

    func checkProxy(proxy string, timeout int) bool {
    proxyUrl, err := url.Parse("http://" + proxy)
    httpClient := &http.Client{
    Transport: &http.Transport{

    Proxy: http.ProxyURL(proxyUrl),

    Dial: func(netw, addr string) (net.Conn, error) {

    deadline := time.Now().Add(time.Duration(timeout) * time.Second)

    c, err := net.DialTimeout(netw, addr, time.Second*time.Duration(timeout))

    if err != nil {

    return nil, err

    }

    c.SetDeadline(deadline)

    return c, nil

    },

    ResponseHeaderTimeout: time.Duration(timeout) * time.Second,

    DisableKeepAlives: true,

    },

    }

    req, err := http.NewRequest("GET", "http://www.urltocheck.com", nil)
    req.Close = true
    resp, err := httpClient.Do(req)

    if err != nil {

    return false

    }

    defer resp.Body.Close()
    body, _ := ioutil.ReadAll(resp.Body)
    if strings.Contains(string(body), "xxx") {

    return true

    } else {

    return false

    }

    }
    After import "net/http/pprof", and after an hours or two, i could see
    these two writeLoop and readLoop goroutines keep growing, the longer i run
    the checker, the more they are (could be more than ten thousands):

    8054 @ 0x41a716 0x4080d4 0x407d22 0x488021 0x41a8e0
    # 0x4080d4 selectgo+0x384 /usr/local/go/src/pkg/runtime/chan.c:996
    # 0x407d22 runtime.selectgo+0x12 /usr/local/go/src/pkg/runtime/chan.c:840
    # 0x488021 net/http.(*persistConn).writeLoop+0x271 /usr/local/go/src/pkg/net/http/transport.go:791

    8030 @ 0x41a716 0x4072d2 0x407718 0x4879cf 0x41a8e0
    # 0x4879cf net/http.(*persistConn).readLoop+0x68f /usr/local/go/src/pkg/net/http/transport.go:778


    And my memory leaks so bad, I find a bug been reports two years ago
    https://code.google.com/p/go/issues/detail?id=4531 (net/http: Transport
    leaks goroutines when request.ContentLength is explicitly short) which had
    been marked fixed, but I think somewhere someplace transport still
    leaking. Anyone would help, thanks a lot!

    --
    You received this message because you are subscribed to the Google Groups "golang-nuts" group.
    To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
    For more options, visit https://groups.google.com/d/optout.
  • Julian at Apr 7, 2014 at 2:56 pm
    I didn't try your patch but after looking at it I disabled compression in
    the http.Transport I'm using and the goroutines stopped leaking, so I
    suppose you're right, that is where the problem lies.
    On Monday, April 7, 2014 3:42:26 AM UTC+3, Billy Shea wrote:

    This patch solved my problem exactly
    https://codereview.appspot.com/84850043, the goroutines now stop leaking!
    Cheers, special thanks to ucgggg and snaury.

    在 2014年4月5日星期六UTC+8下午7时41分39秒,[email protected]写道:
    There is a bug in "net\http\client.go".
    I created an issues: http://code.google.com/p/go/issues/detail?id=7620.
    Just modify "net\http\client.go":
    ------------------------------------------
    if err != nil {
    if resp != nil {
    +++++ if resp.Body != nil {
    +++++ resp.Body.Close()
    +++++ }
    log.Printf("RoundTripper returned a response & error; ignoring response")
    }
    return nil, err
    }
    ------------------------------------------
    Why nobody deal with the issus since I reported it weeks ago?


    在 2014年3月25日星期二UTC+8下午9时01分46秒,Billy Shea写道:
    I wrote a proxy checker recently, it sends hundreds of thousands of
    requests to a url to check whether proxies are available. Here is a part
    of my checker code, the check proxy function:

    func checkProxy(proxy string, timeout int) bool {
    proxyUrl, err := url.Parse("http://" + proxy)
    httpClient := &http.Client{
    Transport: &http.Transport{

    Proxy: http.ProxyURL(proxyUrl),

    Dial: func(netw, addr string) (net.Conn, error) {

    deadline := time.Now().Add(time.Duration(timeout) * time.Second)

    c, err := net.DialTimeout(netw, addr,
    time.Second*time.Duration(timeout))
    if err != nil {

    return nil, err

    }

    c.SetDeadline(deadline)

    return c, nil

    },

    ResponseHeaderTimeout: time.Duration(timeout) * time.Second,

    DisableKeepAlives: true,

    },

    }

    req, err := http.NewRequest("GET", "http://www.urltocheck.com", nil)
    req.Close = true
    resp, err := httpClient.Do(req)

    if err != nil {

    return false

    }

    defer resp.Body.Close()
    body, _ := ioutil.ReadAll(resp.Body)
    if strings.Contains(string(body), "xxx") {

    return true

    } else {

    return false

    }

    }
    After import "net/http/pprof", and after an hours or two, i could see
    these two writeLoop and readLoop goroutines keep growing, the longer i run
    the checker, the more they are (could be more than ten thousands):

    8054 @ 0x41a716 0x4080d4 0x407d22 0x488021 0x41a8e0
    # 0x4080d4 selectgo+0x384 /usr/local/go/src/pkg/runtime/chan.c:996
    # 0x407d22 runtime.selectgo+0x12 /usr/local/go/src/pkg/runtime/chan.c:840
    # 0x488021 net/http.(*persistConn).writeLoop+0x271 /usr/local/go/src/pkg/net/http/transport.go:791

    8030 @ 0x41a716 0x4072d2 0x407718 0x4879cf 0x41a8e0
    # 0x4879cf net/http.(*persistConn).readLoop+0x68f /usr/local/go/src/pkg/net/http/transport.go:778


    And my memory leaks so bad, I find a bug been reports two years ago
    https://code.google.com/p/go/issues/detail?id=4531 (net/http: Transport
    leaks goroutines when request.ContentLength is explicitly short) which had
    been marked fixed, but I think somewhere someplace transport still
    leaking. Anyone would help, thanks a lot!

    --
    You received this message because you are subscribed to the Google Groups "golang-nuts" group.
    To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
    For more options, visit https://groups.google.com/d/optout.
  • Tom Maiaroto at Sep 23, 2014 at 6:04 pm
    Yes. This.
    This is plaguing me BAD right now.

    goroutine 869 [select, 2 minutes]:
    net/http.(*persistConn).readLoop(0xc208956fd0)
             /usr/local/go/src/pkg/net/http/transport.go:868 +0x829
    created by net/http.(*Transport).dialConn
             /usr/local/go/src/pkg/net/http/transport.go:600 +0x93f

    goroutine 778 [select, 2 minutes]:
    net/http.(*persistConn).writeLoop(0xc208956bb0)
             /usr/local/go/src/pkg/net/http/transport.go:885 +0x38f
    created by net/http.(*Transport).dialConn
             /usr/local/go/src/pkg/net/http/transport.go:601 +0x957

    goroutine 515 [select, 2 minutes]:
    net/http.(*persistConn).readLoop(0xc2086878c0)
             /usr/local/go/src/pkg/net/http/transport.go:868 +0x829
    created by net/http.(*Transport).dialConn
             /usr/local/go/src/pkg/net/http/transport.go:600 +0x93f

    goroutine 664 [select, 2 minutes]:
    net/http.(*persistConn).readLoop(0xc208957760)
             /usr/local/go/src/pkg/net/http/transport.go:868 +0x829
    created by net/http.(*Transport).dialConn
             /usr/local/go/src/pkg/net/http/transport.go:600 +0x93f


    They'll sit like that forever. If I refresh the pprof web page, it'll just
    say 24 minutes, etc. The count on the main page never decreases either.
    Eventually, since I keep making more requests on a schedule...It just
    crashes the system. Varies on how long it takes.

    I've been trying to close things all over. I tried a client wrapper that
    lets me set timeouts, etc. I just can't get it to work and am a little
    frustrated by it to be honest.

    I'm on version 1.3.1 ... Is there something else I need still?
    Is this referenced patch (5 months old now) not in the stable build?

    On Monday, April 7, 2014 9:55:52 AM UTC-5, Julian wrote:

    I didn't try your patch but after looking at it I disabled compression in
    the http.Transport I'm using and the goroutines stopped leaking, so I
    suppose you're right, that is where the problem lies.
    On Monday, April 7, 2014 3:42:26 AM UTC+3, Billy Shea wrote:

    This patch solved my problem exactly
    https://codereview.appspot.com/84850043, the goroutines now stop
    leaking! Cheers, special thanks to ucgggg and snaury.

    在 2014年4月5日星期六UTC+8下午7时41分39秒,[email protected]写道:
    There is a bug in "net\http\client.go".
    I created an issues: http://code.google.com/p/go/issues/detail?id=7620.
    Just modify "net\http\client.go":
    ------------------------------------------
    if err != nil {
    if resp != nil {
    +++++ if resp.Body != nil {
    +++++ resp.Body.Close()
    +++++ }
    log.Printf("RoundTripper returned a response & error; ignoring response")
    }
    return nil, err
    }
    ------------------------------------------
    Why nobody deal with the issus since I reported it weeks ago?


    在 2014年3月25日星期二UTC+8下午9时01分46秒,Billy Shea写道:
    I wrote a proxy checker recently, it sends hundreds of thousands of
    requests to a url to check whether proxies are available. Here is a part
    of my checker code, the check proxy function:

    func checkProxy(proxy string, timeout int) bool {
    proxyUrl, err := url.Parse("http://" + proxy)
    httpClient := &http.Client{
    Transport: &http.Transport{

    Proxy: http.ProxyURL(proxyUrl),

    Dial: func(netw, addr string) (net.Conn, error) {

    deadline := time.Now().Add(time.Duration(timeout) * time.Second)

    c, err := net.DialTimeout(netw, addr,
    time.Second*time.Duration(timeout))
    if err != nil {

    return nil, err

    }

    c.SetDeadline(deadline)

    return c, nil

    },

    ResponseHeaderTimeout: time.Duration(timeout) * time.Second,

    DisableKeepAlives: true,

    },

    }

    req, err := http.NewRequest("GET", "http://www.urltocheck.com", nil)
    req.Close = true
    resp, err := httpClient.Do(req)

    if err != nil {

    return false

    }

    defer resp.Body.Close()
    body, _ := ioutil.ReadAll(resp.Body)
    if strings.Contains(string(body), "xxx") {

    return true

    } else {

    return false

    }

    }
    After import "net/http/pprof", and after an hours or two, i could see
    these two writeLoop and readLoop goroutines keep growing, the longer i run
    the checker, the more they are (could be more than ten thousands):

    8054 @ 0x41a716 0x4080d4 0x407d22 0x488021 0x41a8e0
    # 0x4080d4 selectgo+0x384 /usr/local/go/src/pkg/runtime/chan.c:996
    # 0x407d22 runtime.selectgo+0x12 /usr/local/go/src/pkg/runtime/chan.c:840
    # 0x488021 net/http.(*persistConn).writeLoop+0x271 /usr/local/go/src/pkg/net/http/transport.go:791

    8030 @ 0x41a716 0x4072d2 0x407718 0x4879cf 0x41a8e0
    # 0x4879cf net/http.(*persistConn).readLoop+0x68f /usr/local/go/src/pkg/net/http/transport.go:778


    And my memory leaks so bad, I find a bug been reports two years ago
    https://code.google.com/p/go/issues/detail?id=4531 (net/http:
    Transport leaks goroutines when request.ContentLength is explicitly short)
    which had been marked fixed, but I think somewhere someplace transport
    still leaking. Anyone would help, thanks a lot!

    --
    You received this message because you are subscribed to the Google Groups "golang-nuts" group.
    To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
    For more options, visit https://groups.google.com/d/optout.
  • James Bardin at Sep 23, 2014 at 6:20 pm
    The gzip related bug was fixed in go1.3, so you're likely leaking
    connections in some other way.
    On Tue, Sep 23, 2014 at 2:04 PM, Tom Maiaroto wrote:

    Yes. This.
    This is plaguing me BAD right now.

    goroutine 869 [select, 2 minutes]:
    net/http.(*persistConn).readLoop(0xc208956fd0)
    /usr/local/go/src/pkg/net/http/transport.go:868 +0x829
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:600 +0x93f

    goroutine 778 [select, 2 minutes]:
    net/http.(*persistConn).writeLoop(0xc208956bb0)
    /usr/local/go/src/pkg/net/http/transport.go:885 +0x38f
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:601 +0x957

    goroutine 515 [select, 2 minutes]:
    net/http.(*persistConn).readLoop(0xc2086878c0)
    /usr/local/go/src/pkg/net/http/transport.go:868 +0x829
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:600 +0x93f

    goroutine 664 [select, 2 minutes]:
    net/http.(*persistConn).readLoop(0xc208957760)
    /usr/local/go/src/pkg/net/http/transport.go:868 +0x829
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:600 +0x93f


    They'll sit like that forever. If I refresh the pprof web page, it'll just
    say 24 minutes, etc. The count on the main page never decreases either.
    Eventually, since I keep making more requests on a schedule...It just
    crashes the system. Varies on how long it takes.

    I've been trying to close things all over. I tried a client wrapper that
    lets me set timeouts, etc. I just can't get it to work and am a little
    frustrated by it to be honest.

    I'm on version 1.3.1 ... Is there something else I need still?
    Is this referenced patch (5 months old now) not in the stable build?

    On Monday, April 7, 2014 9:55:52 AM UTC-5, Julian wrote:

    I didn't try your patch but after looking at it I disabled compression in
    the http.Transport I'm using and the goroutines stopped leaking, so I
    suppose you're right, that is where the problem lies.
    On Monday, April 7, 2014 3:42:26 AM UTC+3, Billy Shea wrote:

    This patch solved my problem exactly https://codereview.
    appspot.com/84850043, the goroutines now stop leaking! Cheers, special
    thanks to ucgggg and snaury.

    在 2014年4月5日星期六UTC+8下午7时41分39秒,[email protected]写道:
    There is a bug in "net\http\client.go".
    I created an issues: http://code.google.com/p/go/issues/detail?id=7620.
    Just modify "net\http\client.go":
    ------------------------------------------
    if err != nil {
    if resp != nil {
    +++++ if resp.Body != nil {
    +++++ resp.Body.Close()
    +++++ }
    log.Printf("RoundTripper returned a response & error; ignoring response")
    }
    return nil, err
    }
    ------------------------------------------
    Why nobody deal with the issus since I reported it weeks ago?


    在 2014年3月25日星期二UTC+8下午9时01分46秒,Billy Shea写道:
    I wrote a proxy checker recently, it sends hundreds of thousands of
    requests to a url to check whether proxies are available. Here is a part
    of my checker code, the check proxy function:

    func checkProxy(proxy string, timeout int) bool {
    proxyUrl, err := url.Parse("http://" + proxy)
    httpClient := &http.Client{
    Transport: &http.Transport{

    Proxy: http.ProxyURL(proxyUrl),

    Dial: func(netw, addr string) (net.Conn, error) {

    deadline := time.Now().Add(time.Duration(timeout) * time.Second)

    c, err := net.DialTimeout(netw, addr, time.Second*time.Duration(
    timeout))
    if err != nil {

    return nil, err

    }

    c.SetDeadline(deadline)

    return c, nil

    },

    ResponseHeaderTimeout: time.Duration(timeout) * time.Second,

    DisableKeepAlives: true,

    },

    }

    req, err := http.NewRequest("GET", "http://www.urltocheck.com", nil)
    req.Close = true
    resp, err := httpClient.Do(req)

    if err != nil {

    return false

    }

    defer resp.Body.Close()
    body, _ := ioutil.ReadAll(resp.Body)
    if strings.Contains(string(body), "xxx") {

    return true

    } else {

    return false

    }

    }
    After import "net/http/pprof", and after an hours or two, i could see
    these two writeLoop and readLoop goroutines keep growing, the longer i run
    the checker, the more they are (could be more than ten thousands):

    8054 @ 0x41a716 0x4080d4 0x407d22 0x488021 0x41a8e0
    # 0x4080d4 selectgo+0x384 /usr/local/go/src/pkg/runtime/chan.c:996
    # 0x407d22 runtime.selectgo+0x12 /usr/local/go/src/pkg/runtime/chan.c:840
    # 0x488021 net/http.(*persistConn).writeLoop+0x271 /usr/local/go/src/pkg/net/http/transport.go:791

    8030 @ 0x41a716 0x4072d2 0x407718 0x4879cf 0x41a8e0
    # 0x4879cf net/http.(*persistConn).readLoop+0x68f /usr/local/go/src/pkg/net/http/transport.go:778


    And my memory leaks so bad, I find a bug been reports two years ago
    https://code.google.com/p/go/issues/detail?id=4531 (net/http:
    Transport leaks goroutines when request.ContentLength is explicitly short)
    which had been marked fixed, but I think somewhere someplace
    transport still leaking. Anyone would help, thanks a lot!

    --
    You received this message because you are subscribed to a topic in the
    Google Groups "golang-nuts" group.
    To unsubscribe from this topic, visit
    https://groups.google.com/d/topic/golang-nuts/FnJZ9iZ0i_g/unsubscribe.
    To unsubscribe from this group and all its topics, send an email to
    [email protected].
    For more options, visit https://groups.google.com/d/optout.
    --
    You received this message because you are subscribed to the Google Groups "golang-nuts" group.
    To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
    For more options, visit https://groups.google.com/d/optout.
  • Tom Maiaroto at Sep 23, 2014 at 6:26 pm
    Hmm, drat.
    This is hard one to track down. I've tried various http client wrappers and
    the native client on my own. Nothing seems to work.
    Are there any things obvious you can suggest I look into?
    Do you know of a solid package to use when looping and making many requests
    (hundreds to thousands) within a goroutine?
    Thanks.
    On Tuesday, September 23, 2014 1:20:41 PM UTC-5, James Bardin wrote:

    The gzip related bug was fixed in go1.3, so you're likely leaking
    connections in some other way.

    On Tue, Sep 23, 2014 at 2:04 PM, Tom Maiaroto <[email protected]
    <javascript:>> wrote:
    Yes. This.
    This is plaguing me BAD right now.

    goroutine 869 [select, 2 minutes]:
    net/http.(*persistConn).readLoop(0xc208956fd0)
    /usr/local/go/src/pkg/net/http/transport.go:868 +0x829
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:600 +0x93f

    goroutine 778 [select, 2 minutes]:
    net/http.(*persistConn).writeLoop(0xc208956bb0)
    /usr/local/go/src/pkg/net/http/transport.go:885 +0x38f
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:601 +0x957

    goroutine 515 [select, 2 minutes]:
    net/http.(*persistConn).readLoop(0xc2086878c0)
    /usr/local/go/src/pkg/net/http/transport.go:868 +0x829
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:600 +0x93f

    goroutine 664 [select, 2 minutes]:
    net/http.(*persistConn).readLoop(0xc208957760)
    /usr/local/go/src/pkg/net/http/transport.go:868 +0x829
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:600 +0x93f


    They'll sit like that forever. If I refresh the pprof web page, it'll
    just say 24 minutes, etc. The count on the main page never decreases
    either. Eventually, since I keep making more requests on a schedule...It
    just crashes the system. Varies on how long it takes.

    I've been trying to close things all over. I tried a client wrapper that
    lets me set timeouts, etc. I just can't get it to work and am a little
    frustrated by it to be honest.

    I'm on version 1.3.1 ... Is there something else I need still?
    Is this referenced patch (5 months old now) not in the stable build?

    On Monday, April 7, 2014 9:55:52 AM UTC-5, Julian wrote:

    I didn't try your patch but after looking at it I disabled compression
    in the http.Transport I'm using and the goroutines stopped leaking, so I
    suppose you're right, that is where the problem lies.
    On Monday, April 7, 2014 3:42:26 AM UTC+3, Billy Shea wrote:

    This patch solved my problem exactly https://codereview.
    appspot.com/84850043, the goroutines now stop leaking! Cheers, special
    thanks to ucgggg and snaury.

    在 2014年4月5日星期六UTC+8下午7时41分39秒,[email protected]写道:
    There is a bug in "net\http\client.go".
    I created an issues: http://code.google.com/p/go/issues/detail?id=7620.
    Just modify "net\http\client.go":
    ------------------------------------------
    if err != nil {
    if resp != nil {
    +++++ if resp.Body != nil {
    +++++ resp.Body.Close()
    +++++ }
    log.Printf("RoundTripper returned a response & error; ignoring response")
    }
    return nil, err
    }
    ------------------------------------------
    Why nobody deal with the issus since I reported it weeks ago?


    在 2014年3月25日星期二UTC+8下午9时01分46秒,Billy Shea写道:
    I wrote a proxy checker recently, it sends hundreds of thousands of
    requests to a url to check whether proxies are available. Here is a part
    of my checker code, the check proxy function:

    func checkProxy(proxy string, timeout int) bool {
    proxyUrl, err := url.Parse("http://" + proxy)
    httpClient := &http.Client{
    Transport: &http.Transport{

    Proxy: http.ProxyURL(proxyUrl),

    Dial: func(netw, addr string) (net.Conn, error) {

    deadline := time.Now().Add(time.Duration(timeout) * time.Second)

    c, err := net.DialTimeout(netw, addr, time.Second*time.Duration(
    timeout))
    if err != nil {

    return nil, err

    }

    c.SetDeadline(deadline)

    return c, nil

    },

    ResponseHeaderTimeout: time.Duration(timeout) * time.Second,

    DisableKeepAlives: true,

    },

    }

    req, err := http.NewRequest("GET", "http://www.urltocheck.com",
    nil)
    req.Close = true
    resp, err := httpClient.Do(req)

    if err != nil {

    return false

    }

    defer resp.Body.Close()
    body, _ := ioutil.ReadAll(resp.Body)
    if strings.Contains(string(body), "xxx") {

    return true

    } else {

    return false

    }

    }
    After import "net/http/pprof", and after an hours or two, i could
    see these two writeLoop and readLoop goroutines keep growing, the longer i
    run the checker, the more they are (could be more than ten thousands):

    8054 @ 0x41a716 0x4080d4 0x407d22 0x488021 0x41a8e0
    # 0x4080d4 selectgo+0x384 /usr/local/go/src/pkg/runtime/chan.c:996
    # 0x407d22 runtime.selectgo+0x12 /usr/local/go/src/pkg/runtime/chan.c:840
    # 0x488021 net/http.(*persistConn).writeLoop+0x271 /usr/local/go/src/pkg/net/http/transport.go:791

    8030 @ 0x41a716 0x4072d2 0x407718 0x4879cf 0x41a8e0
    # 0x4879cf net/http.(*persistConn).readLoop+0x68f /usr/local/go/src/pkg/net/http/transport.go:778


    And my memory leaks so bad, I find a bug been reports two years ago
    https://code.google.com/p/go/issues/detail?id=4531 (net/http:
    Transport leaks goroutines when request.ContentLength is explicitly short)
    which had been marked fixed, but I think somewhere someplace
    transport still leaking. Anyone would help, thanks a lot!

    --
    You received this message because you are subscribed to a topic in the
    Google Groups "golang-nuts" group.
    To unsubscribe from this topic, visit
    https://groups.google.com/d/topic/golang-nuts/FnJZ9iZ0i_g/unsubscribe.
    To unsubscribe from this group and all its topics, send an email to
    [email protected] <javascript:>.
    For more options, visit https://groups.google.com/d/optout.
    --
    You received this message because you are subscribed to the Google Groups "golang-nuts" group.
    To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
    For more options, visit https://groups.google.com/d/optout.
  • Tom Maiaroto at Sep 23, 2014 at 7:21 pm
    This is what I've been using:
    https://gist.github.com/seantalts/11266762

    Could be it be a particular URL? I'm calling the Facebook API.
    Is there anything a server response could do to create these issues?


    On Tuesday, September 23, 2014 1:26:09 PM UTC-5, Tom Maiaroto wrote:

    Hmm, drat.
    This is hard one to track down. I've tried various http client wrappers
    and the native client on my own. Nothing seems to work.
    Are there any things obvious you can suggest I look into?
    Do you know of a solid package to use when looping and making many
    requests (hundreds to thousands) within a goroutine?
    Thanks.
    On Tuesday, September 23, 2014 1:20:41 PM UTC-5, James Bardin wrote:

    The gzip related bug was fixed in go1.3, so you're likely leaking
    connections in some other way.

    On Tue, Sep 23, 2014 at 2:04 PM, Tom Maiaroto <[email protected]>
    wrote:
    Yes. This.
    This is plaguing me BAD right now.

    goroutine 869 [select, 2 minutes]:
    net/http.(*persistConn).readLoop(0xc208956fd0)
    /usr/local/go/src/pkg/net/http/transport.go:868 +0x829
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:600 +0x93f

    goroutine 778 [select, 2 minutes]:
    net/http.(*persistConn).writeLoop(0xc208956bb0)
    /usr/local/go/src/pkg/net/http/transport.go:885 +0x38f
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:601 +0x957

    goroutine 515 [select, 2 minutes]:
    net/http.(*persistConn).readLoop(0xc2086878c0)
    /usr/local/go/src/pkg/net/http/transport.go:868 +0x829
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:600 +0x93f

    goroutine 664 [select, 2 minutes]:
    net/http.(*persistConn).readLoop(0xc208957760)
    /usr/local/go/src/pkg/net/http/transport.go:868 +0x829
    created by net/http.(*Transport).dialConn
    /usr/local/go/src/pkg/net/http/transport.go:600 +0x93f


    They'll sit like that forever. If I refresh the pprof web page, it'll
    just say 24 minutes, etc. The count on the main page never decreases
    either. Eventually, since I keep making more requests on a schedule...It
    just crashes the system. Varies on how long it takes.

    I've been trying to close things all over. I tried a client wrapper that
    lets me set timeouts, etc. I just can't get it to work and am a little
    frustrated by it to be honest.

    I'm on version 1.3.1 ... Is there something else I need still?
    Is this referenced patch (5 months old now) not in the stable build?

    On Monday, April 7, 2014 9:55:52 AM UTC-5, Julian wrote:

    I didn't try your patch but after looking at it I disabled compression
    in the http.Transport I'm using and the goroutines stopped leaking, so I
    suppose you're right, that is where the problem lies.
    On Monday, April 7, 2014 3:42:26 AM UTC+3, Billy Shea wrote:

    This patch solved my problem exactly https://codereview.
    appspot.com/84850043, the goroutines now stop leaking! Cheers,
    special thanks to ucgggg and snaury.

    在 2014年4月5日星期六UTC+8下午7时41分39秒,[email protected]写道:
    There is a bug in "net\http\client.go".
    I created an issues: http://code.google.com/p/go/issues/detail?id=7620.
    Just modify "net\http\client.go":
    ------------------------------------------
    if err != nil {
    if resp != nil {
    +++++ if resp.Body != nil {
    +++++ resp.Body.Close()
    +++++ }
    log.Printf("RoundTripper returned a response & error; ignoring response")
    }
    return nil, err
    }
    ------------------------------------------
    Why nobody deal with the issus since I reported it weeks ago?


    在 2014年3月25日星期二UTC+8下午9时01分46秒,Billy Shea写道:
    I wrote a proxy checker recently, it sends hundreds of thousands of
    requests to a url to check whether proxies are available. Here is a part
    of my checker code, the check proxy function:

    func checkProxy(proxy string, timeout int) bool {
    proxyUrl, err := url.Parse("http://" + proxy)
    httpClient := &http.Client{
    Transport: &http.Transport{

    Proxy: http.ProxyURL(proxyUrl),

    Dial: func(netw, addr string) (net.Conn, error) {

    deadline := time.Now().Add(time.Duration(timeout) * time.Second)

    c, err := net.DialTimeout(netw, addr, time.Second*time.Duration(
    timeout))
    if err != nil {

    return nil, err

    }

    c.SetDeadline(deadline)

    return c, nil

    },

    ResponseHeaderTimeout: time.Duration(timeout) * time.Second,

    DisableKeepAlives: true,

    },

    }

    req, err := http.NewRequest("GET", "http://www.urltocheck.com",
    nil)
    req.Close = true
    resp, err := httpClient.Do(req)

    if err != nil {

    return false

    }

    defer resp.Body.Close()
    body, _ := ioutil.ReadAll(resp.Body)
    if strings.Contains(string(body), "xxx") {

    return true

    } else {

    return false

    }

    }
    After import "net/http/pprof", and after an hours or two, i could
    see these two writeLoop and readLoop goroutines keep growing, the longer i
    run the checker, the more they are (could be more than ten thousands):

    8054 @ 0x41a716 0x4080d4 0x407d22 0x488021 0x41a8e0
    # 0x4080d4 selectgo+0x384 /usr/local/go/src/pkg/runtime/chan.c:996
    # 0x407d22 runtime.selectgo+0x12 /usr/local/go/src/pkg/runtime/chan.c:840
    # 0x488021 net/http.(*persistConn).writeLoop+0x271 /usr/local/go/src/pkg/net/http/transport.go:791

    8030 @ 0x41a716 0x4072d2 0x407718 0x4879cf 0x41a8e0
    # 0x4879cf net/http.(*persistConn).readLoop+0x68f /usr/local/go/src/pkg/net/http/transport.go:778


    And my memory leaks so bad, I find a bug been reports two years ago
    https://code.google.com/p/go/issues/detail?id=4531 (net/http:
    Transport leaks goroutines when request.ContentLength is explicitly short)
    which had been marked fixed, but I think somewhere someplace
    transport still leaking. Anyone would help, thanks a lot!

    --
    You received this message because you are subscribed to a topic in the
    Google Groups "golang-nuts" group.
    To unsubscribe from this topic, visit
    https://groups.google.com/d/topic/golang-nuts/FnJZ9iZ0i_g/unsubscribe.
    To unsubscribe from this group and all its topics, send an email to
    [email protected].
    For more options, visit https://groups.google.com/d/optout.
    --
    You received this message because you are subscribed to the Google Groups "golang-nuts" group.
    To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
    For more options, visit https://groups.google.com/d/optout.
  • James Bardin at Sep 23, 2014 at 7:45 pm
    http.Client has a Timeout field, so you don't need your TimeoutTransport.
    In your case though, you've basically re-implemented
    Transport.ResponseHeaderTimeout, since you return before Body is read at
    all.

    No idea what your actual code looks like, but your test stops before
    calling resp.Body.Close if there's a read error. If you used that same
    logic with an early return it would be a problem.
    On Tue, Sep 23, 2014 at 3:21 PM, Tom Maiaroto wrote:

    This is what I've been using:
    https://gist.github.com/seantalts/11266762

    Could be it be a particular URL? I'm calling the Facebook API.
    Is there anything a server response could do to create these issues?
    --
    You received this message because you are subscribed to the Google Groups "golang-nuts" group.
    To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
    For more options, visit https://groups.google.com/d/optout.
  • Tom Maiaroto at Sep 23, 2014 at 11:48 pm
    K. Thanks!
    I had to get medieval on this thing and track it down. Otherwise I would
    have been trying over and over something that was fixed =) So thanks for
    pointing that out.

    On Tuesday, September 23, 2014 2:45:36 PM UTC-5, James Bardin wrote:


    http.Client has a Timeout field, so you don't need your TimeoutTransport.
    In your case though, you've basically re-implemented
    Transport.ResponseHeaderTimeout, since you return before Body is read at
    all.

    No idea what your actual code looks like, but your test stops before
    calling resp.Body.Close if there's a read error. If you used that same
    logic with an early return it would be a problem.

    On Tue, Sep 23, 2014 at 3:21 PM, Tom Maiaroto <[email protected]
    <javascript:>> wrote:
    This is what I've been using:
    https://gist.github.com/seantalts/11266762

    Could be it be a particular URL? I'm calling the Facebook API.
    Is there anything a server response could do to create these issues?
    --
    You received this message because you are subscribed to the Google Groups "golang-nuts" group.
    To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
    For more options, visit https://groups.google.com/d/optout.

Related Discussions

Discussion Navigation
viewthread | post
Discussion Overview
groupgolang-nuts @
categoriesgo
postedMar 25, '14 at 6:41p
activeSep 23, '14 at 11:48p
posts21
users7
websitegolang.org

People

Translate

site design / logo © 2023 Grokbase