none
HPC Remote Service RRS feed

  • Question

  • When trying to access an application, event viewer was showing the following message:

    "TCP/IP failed to establish an outgoing connection because the selected local endpoint was recently used to connect to the same remote endpoint. This error typically occurs when outgoing connections are opened and closed at a high rate, causing all available local ports to be used and forcing TCP/IP to reuse a local port for an outgoing connection. To minimize the risk of data corruption, the TCP/IP standard requires a minimum time period to elapse between successive connections from a given local endpoint to a given remote endpoint."

    Using netstat -qno, I could see that the service HPC remote had lots of tcp sockets open so I restarted that service and everything started working as expected.

    What can be done so that error message doesn't appear again and the application continues working?

    Tuesday, May 21, 2019 4:34 PM

All replies

  • Hi erikllor,

    What's the HPC Pack version (HPC Cluster Manager -> Help -> About)?

    Could you copy and paste the 'netstat -qno' output to indicate the port number and PID with lots of tcp sockets open? Could you find the process name of the PID and indicate the name of the HPC service that holds the connections?

    Regards,

    Yutong Sun

    Wednesday, May 22, 2019 2:41 AM
  • Server Version:  5.2.6291.0

    HPC Pack 2016

    This is how it is currently:

    TCP    [::]:64960             [::]:0                 BOUND           12964

    The number of bound connections for that pid is:

    netstat -qno | find "BOUND" | find "12964" /c
    214

    This process belongs to HpcReporting.

    Wednesday, May 22, 2019 4:22 PM
  • Checking again, the bound connections are increasing and are at these values:

    C:\Users\xiss>netstat -qno | find "BOUND" | find "12964" /c
    386

    HpcReporting.exe

    Thursday, May 23, 2019 1:49 PM
  • Hi erikllor,

    Right, this is a known issue for 5.2, the HPCDiagnostics service has an issue in WCF frontend which would cause HpcReporting service and Web Portal (if enabled) to leak ports in Bound state. We've fixed it in the upcoming HPC Pack 2016 Update 3 (5.3) this July. For a temporary workaround, you may just restart the HpcReporting service to release the ports.

    Regards,

    Yutong Sun

    Wednesday, May 29, 2019 3:27 AM