DOCSP-24317 FAQ Diagnostic for tcp_retries2 (#3503)

* DOCSP-24317 tcp_retries2 FAQ

* tcp_retries adjustment in Linux

* Documents windows tcp_retries adjustment

* Minor edits

* Minor edits

* Fixes per Ashley

Co-authored-by: Ashley Brown <[email protected]>

* io code-block setup

* Fixes build issue

* Fixes rendering issue

* Fixes per Shane Harvey

---------

Co-authored-by: Ashley Brown <[email protected]>
Co-authored-by: Kenneth P. J. Dyer <[email protected]>

Author: 3 people

source/faq/diagnostics.txt

@@ -76,6 +76,46 @@ Adjusting the TCP keepalive value:
 You will need to restart :binary:`~bin.mongod` and :binary:`~bin.mongos`
 processes for new system-wide keepalive settings to take effect.
 
+.. _faq-tcp_retries2:
+
+Do TCP Retransmission Timeouts affect MongoDB Deployments?
+----------------------------------------------------------
+
+If you experience long stalls (stalls greater than two minutes) followed
+by network timeouts or socket errors between clients and server
+or between members of a sharded cluster or replica set, 
+check the ``tcp_retries2`` value for the affected systems.
+
+Most Linux operating systems set this value to ``15`` by default, while
+Windows sets it to ``5``.  For MongoDB, you experience better results
+with a lower ``tcp_retries2`` value, on the order of ``5`` (12 seconds)
+or lower.
+
+If your MongoDB deployment experiences TCP retransmission timeout-related 
+issues, change the ``tcp_retries2`` value (``TcpMaxDataRetransmission`` 
+on Windows) for *all* affected systems. This includes all machines running 
+:binary:`~bin.mongod` or :binary:`~bin.mongos` processes and 
+all machines hosting client processes that connect to MongoDB.
+
+Adjust the TCP Retransmission Timeout
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. tabs::
+
+   .. tab:: Linux
+      :tabid: linux
+
+      .. include:: /includes/fact-tcp-retries-linux.rst
+
+   .. tab:: Windows 
+      :tabid: windows
+
+      .. include:: /includes/fact-tcp-retries-windows.rst
+
+
+
+
+
 Why does MongoDB log so many "Connection Accepted" events?
 ----------------------------------------------------------
 

source/includes/fact-tcp-retries-linux.rst

@@ -0,0 +1,47 @@
+
+On most Linux operating systems, control the TCP retransmission
+by adjusting the ``net.ipv4.tcp_retries2`` sysctl setting.
+
+.. note:: 
+
+   Although the setting name includes ``ipv4``, the ``tcp_retries2`` setting
+   applies to both IPv4 and IPv6.
+
+- To view the current setting, use the ``sysctl`` command:
+
+  .. io-code-block::
+
+     .. input::
+        :language: bash
+
+        sysctl net.ipv4.tcp_retries2
+
+     .. output::
+        :language: bash
+
+        net.ipv4.tcp_retries = 15
+
+- To change the ``tcp_retries2`` setting at runtime, use the ``sysctl`` command:
+
+  .. code-block:: bash
+
+     sysctl -w net.ipv4.tcp_retries2=8
+
+- To make the change permanent, edit the configuration file:
+
+  #. Open ``/etc/sysctl.conf`` in your preferred text editor:
+
+     .. code-block:: bash
+
+        vi /etc/sysctl.conf
+
+  #. Configure the ``net.ipv4.tcp_retries2`` setting:
+
+     .. code-block:: conf
+
+        net.ipv4.tcp_retries2 = 8
+
+  #. Restart the system. 
+  
+  Your system now uses the new ``tcp_retries2`` setting.
+

source/includes/fact-tcp-retries-windows.rst

@@ -0,0 +1,20 @@
+On Windows, control TCP Retransmission by adjusting the 
+``TcpMaxDataRetransmissions`` parameter.
+
+- To view the ``TcpMaxDataRetransmissions`` setting on Windows, issue the 
+  following command:
+
+  .. code-block:: powershell
+
+     reg query HKLM\SYSTEM\CurrentControlSet\Services\Tcpip\Parameters /v TcpMaxDataRetransmissions
+
+  By default, the parameter is not set. The system default,
+  used if the value is absent, is ``5`` retries.
+
+- To change the ``TcpMaxDataRetransmissions`` value, use the following command 
+  in an Administrator :guilabel:`Command Prompt`, where ``<value>`` is an integer: 
+
+  .. code-block:: powershell
+
+     reg add HKLM\SYSTEM\CurrentControlSet\Services\Tcpip\Parameters\ /t REG_DWORD /v TcpMaxDataRetransmission /d <value>
+