net/http having "use of closed network connection" error

Question

I am getting a lot of errors like below mentioned,

read tcp xx.xx.xx.xx:80: use of closed network connection

read tcp xx.xx.xx.xx:80: connection reset by peer

//function for HTTP connection

func GetResponseBytesByURL_raw(restUrl, connectionTimeOutStr, readTimeOutStr string) ([]byte, error) {
    connectionTimeOut, _ /*err*/ := time.ParseDuration(connectionTimeOutStr)
    readTimeOut, _ /*err*/ := time.ParseDuration(readTimeOutStr)
    timeout := connectionTimeOut + readTimeOut // time.Duration((strconv.Atoi(connectionTimeOutStr) + strconv.Atoi(readTimeOutStr)))
    //timeout = 200 * time.Millisecond
    client := http.Client{
        Timeout: timeout,
    }
    resp, err := client.Get(restUrl)
    if nil != err {
        logger.SetLog("Error GetResponseBytesByURL_raw |err: ", logs.LevelError, err)
        return make([]byte, 0), err
    }
    defer resp.Body.Close()
    body, err := ioutil.ReadAll(resp.Body)
    return body, err
}

Update (July 14):

Server : NumCPU=8, RAM=24GB, GO=go1.4.2.linux-amd64

I am getting such error during some high traffic. 20000-30000 request per minutes, and I have a time frame of 500ms to fetch response from third party api.

netstat status from my server (using : netstat -nat | awk '{print $6}' | sort | uniq -c | sort -n) to get frequency

      1 established)
      1 Foreign
      9 LISTEN
     33 FIN_WAIT1
    338 ESTABLISHED
   5530 SYN_SENT
  32202 TIME_WAIT

sysctl -p

**sysctl -p**
fs.file-max = 2097152
vm.swappiness = 10
vm.dirty_ratio = 60
vm.dirty_background_ratio = 2
net.ipv4.tcp_synack_retries = 2
net.ipv4.ip_local_port_range = 2000 65535
net.ipv4.tcp_rfc1337 = 1
net.ipv4.tcp_fin_timeout = 5
net.ipv4.tcp_keepalive_time = 300
net.ipv4.tcp_keepalive_probes = 5
net.ipv4.tcp_keepalive_intvl = 15
net.core.rmem_default = 31457280
net.core.rmem_max = 12582912
net.core.wmem_default = 31457280
net.core.wmem_max = 12582912
net.core.somaxconn = 65536
net.core.netdev_max_backlog = 65536
net.core.optmem_max = 25165824
net.ipv4.tcp_mem = 65536 131072 262144
net.ipv4.udp_mem = 65536 131072 262144
net.ipv4.tcp_rmem = 8192 87380 16777216
net.ipv4.udp_rmem_min = 16384
net.ipv4.tcp_wmem = 8192 65536 16777216
net.ipv4.udp_wmem_min = 16384
net.ipv4.tcp_max_tw_buckets = 1440000
net.ipv4.tcp_tw_recycle = 0
net.ipv4.tcp_tw_reuse = 1
net.ipv6.bindv6only = 1
net.ipv4.conf.all.rp_filter = 1
net.ipv4.conf.all.accept_source_route = 0
net.ipv4.icmp_echo_ignore_broadcasts = 1
error: "net.ipv4.icmp_ignore_bogus_error_messages" is an unknown key
kernel.exec-shield = 1
kernel.randomize_va_space = 1
net.ipv4.conf.all.log_martians = 1
net.ipv4.conf.default.log_martians = 1
net.ipv4.icmp_ignore_bogus_error_responses = 1
net.ipv4.ip_forward = 0
net.ipv4.conf.all.send_redirects = 0
net.ipv4.conf.default.send_redirects = 0
net.ipv4.conf.all.secure_redirects = 0
net.ipv4.conf.default.secure_redirects = 0

Answer 1

When making connections at a high rate over the internet, it's very likely you're going to encounter some connection problems. You can't mitigate them completely, so you may want to add retry logic around the request. The actual error type at this point probably doesn't matter, but matching the error string for use of closed network connection or connection reset by peer is about the best you can do if you want to be specific. Make sure to limit the retries with a backoff, as some systems will drop or reset connections as a way to limit request rates, and you may get more errors the faster you reconnect.

Depending on the number of remote hosts you're communicating with, you will want to increase Transport.MaxIdleConnsPerHost (the default is only 2). The fewer hosts you talk to, the higher you can set this. This will decrease the number of new connections made, and speed up the requests overall.

If you can, try the go1.5 beta. There have been a couple changes around keep-alive connections that may help reduce the number of errors you see.

Answer 2

I recommend implementing an exponential back off or some other rate limiting mechanism on your side of the wire. There's not really anything you can do about those error, and using exponential back off won't necessarily make you get the data any faster either. But it can ensure that you get all the data and the API you're pulling from will surely appreciate the reduced traffic. Here's a link to one I found on GitHub; https://github.com/cenkalti/backoff

There was another popular option as well though I haven't used either. Implementing one yourself isn't terribly difficult either and I could provide some sample of that on request. One thing I do recommend based off my experience is make sure you're using a retry function that has an abort channel. If you get to really long back off times then you'll want some way for the caller to kill it.