Enable TCP keep alive on HTTP socket level

This commit changes the socket options of a connection in a way that
keepalive (socket.SO_KEEPALIVE) is enabled by default. If enabled, the
kernel settings can be overwritten by individual options. Setting the
socket options is still handled by urllib3.

Reference: https://urllib3.readthedocs.io/en/latest/reference/urllib3.connection.html#urllib3.connection.HTTPConnection

The new arguments for `crate.client.connect` and respectively
`crate.client.connection.Connection` are:

* `socket_keepalive`: (optional, defaults to ``True``)
  Enable TCP keepalive on socket level.
* `socket_tcp_keepidle`: (optional)
  Set the `TCP_KEEPIDLE` socket option, which overrides
  `net.ipv4.tcp_keepalive_time` kernel setting if `socket_keepalive`
  is `True`.
* `socket_tcp_keepintvl`: (optional)
  Set the `TCP_KEEPINTVL` socket option, which overrides
  `net.ipv4.tcp_keepalive_intvl` kernel setting if `socket_keepalive`
  is `True`.
* `socket_tcp_keepcnt`: (optional)
  Set the `TCP_KEEPCNT` socket option, which overrides
  `net.ipv4.tcp_keepalive_probes` kernel setting if `socket_keepalive`
  is `True`.

Author: chaudum

CHANGES.txt

@@ -5,6 +5,13 @@ Changes for crate
 Unreleased
 ==========
 
+- Enabled TCP keepalive on socket level and support for setting socket options
+  when creating the connection. The supported options are:
+
+  - ``TCP_KEEPIDLE`` (overriding ``net.ipv4.tcp_keepalive_time``)
+  - ``TCP_KEEPINTVL`` (overriding ``net.ipv4.tcp_keepalive_intvl``)
+  - ``TCP_KEEPCNT`` (overriding ``net.ipv4.tcp_keepalive_probes``)
+
 - Propagate connect parameter ``pool_size`` to urllib3 as ``maxsize`` parameter
   in order to make the connection pool size configurable.
 

docs/connect.rst

@@ -167,20 +167,51 @@ the optional ``error_trace`` argument to ``True``, like so::
 
     >>> connection = client.connect(..., error_trace=True)
 
-.. _authentication:
-
 Backoff Factor
-..............
+--------------
 
 When attempting to make a request, the connection can be configured so that
 retries are made in increasing time intervals. This can be configured like so::
 
     >>> connection = client.connect(..., backoff_factor=0.1)
 
-If ``backoff_factor``is set to 0.1, then the delay between retries will be 0.0,
+If ``backoff_factor`` is set to 0.1, then the delay between retries will be 0.0,
 0.1, 0.2, 0.4 etc. The maximum backoff factor cannot exceed 120 seconds and by
 default its value is 0.
 
+Socket Options
+--------------
+
+Creating connections uses `urllib3 default socket options`_ but additionally
+enables TCP keepalive by setting ``socket.SO_KEEPALIVE`` to ``1``.
+
+Keepalive can be disabled using the ``socket_keepalive`` argument, like so::
+
+    >>> connection = client.connect(..., socket_keepalive=False)
+
+If keepalive is enabled (default), there are three additional, optional socket
+options that can be configured via connection arguments.
+
+:``socket_tcp_keepidle``:
+
+    Set the ``TCP_KEEPIDLE`` socket option, which overrides
+    ``net.ipv4.tcp_keepalive_time`` kernel setting if ``socket_keepalive`` is
+    ``True``.
+
+:``socket_tcp_keepintvl``:
+
+    Set the ``TCP_KEEPINTVL`` socket option, which overrides
+    ``net.ipv4.tcp_keepalive_intvl`` kernel setting if ``socket_keepalive`` is
+    ``True``.
+
+:``socket_tcp_keepcnt``:
+
+    Set the ``TCP_KEEPCNT`` socket option, which overrides
+    ``net.ipv4.tcp_keepalive_probes`` kernel setting if ``socket_keepalive`` is
+    ``True``.
+
+.. _authentication:
+
 Authentication
 ==============
 
@@ -247,3 +278,4 @@ Once you're connected, you can :ref:`query CrateDB <query>`.
 .. _socket timeout: https://docs.python.org/2/library/socket.html#socket.getdefaulttimeout
 .. _SQLAlchemy: http://www.sqlalchemy.org/
 .. _tracebacks: https://docs.python.org/3/library/traceback.html
+.. _urllib3 default socket options: https://urllib3.readthedocs.io/en/latest/reference/urllib3.connection.html#urllib3.connection.HTTPConnection

src/crate/client/__init__.py

@@ -19,7 +19,7 @@
 # with Crate these terms will supersede the license and you may use the
 # software solely pursuant to the terms of the relevant commercial agreement.
 
-from .connection import connect
+from .connection import Connection as connect
 from .exceptions import Error
 
 __all__ = [

src/crate/client/connection.py

@@ -41,7 +41,65 @@ def __init__(self,
                  username=None,
                  password=None,
                  schema=None,
-                 pool_size=None):
+                 pool_size=None,
+                 socket_keepalive=True,
+                 socket_tcp_keepidle=None,
+                 socket_tcp_keepintvl=None,
+                 socket_tcp_keepcnt=None,
+                 ):
+        """
+        :param servers:
+            either a string in the form of '<hostname>:<port>'
+            or a list of servers in the form of ['<hostname>:<port>', '...']
+        :param timeout:
+            (optional)
+            define the retry timeout for unreachable servers in seconds
+        :param backoff_factor:
+            (optional)
+            define the retry interval for unreachable servers in seconds
+        :param client:
+            (optional - for testing)
+            client used to communicate with crate.
+        :param verify_ssl_cert:
+            if set to ``True`` verify the servers SSL server certificate.
+            defaults to ``False``
+        :param ca_cert:
+            a path to a CA certificate to use when verifying the SSL server
+            certificate.
+        :param error_trace:
+            if set to ``True`` return a whole stacktrace of any server error if
+            one occurs
+        :param cert_file:
+            a path to the client certificate to present to the server.
+        :param key_file:
+            a path to the client key to use when communicating with the server.
+        :param username:
+            the username in the database.
+        :param password:
+            the password of the user in the database.
+        :param pool_size:
+            (optional)
+            Number of connections to save that can be reused.
+            More than 1 is useful in multithreaded situations.
+        :param socket_keepalive:
+            (optional, defaults to ``True``)
+            Enable TCP keepalive on socket level.
+        :param socket_tcp_keepidle:
+            (optional)
+            Set the ``TCP_KEEPIDLE`` socket option, which overrides
+            ``net.ipv4.tcp_keepalive_time`` kernel setting if ``socket_keepalive``
+            is ``True``.
+        :param socket_tcp_keepintvl:
+            (optional)
+            Set the ``TCP_KEEPINTVL`` socket option, which overrides
+            ``net.ipv4.tcp_keepalive_intvl`` kernel setting if ``socket_keepalive``
+            is ``True``.
+        :param socket_tcp_keepcnt:
+            (optional)
+            Set the ``TCP_KEEPCNT`` socket option, which overrides
+            ``net.ipv4.tcp_keepalive_probes`` kernel setting if ``socket_keepalive``
+            is ``True``.
+        """
         if client:
             self.client = client
         else:
@@ -56,7 +114,12 @@ def __init__(self,
                                  username=username,
                                  password=password,
                                  schema=schema,
-                                 pool_size=pool_size)
+                                 pool_size=pool_size,
+                                 socket_keepalive=socket_keepalive,
+                                 socket_tcp_keepidle=socket_tcp_keepidle,
+                                 socket_tcp_keepintvl=socket_tcp_keepintvl,
+                                 socket_tcp_keepcnt=socket_tcp_keepcnt,
+                                 )
         self.lowest_server_version = self._lowest_server_version()
         self._closed = False
 
@@ -113,67 +176,5 @@ def __exit__(self, *excs):
         self.close()
 
 
-def connect(servers=None,
-            timeout=None,
-            backoff_factor=0,
-            client=None,
-            verify_ssl_cert=False,
-            ca_cert=None,
-            error_trace=False,
-            cert_file=None,
-            key_file=None,
-            username=None,
-            password=None,
-            schema=None,
-            pool_size=None):
-    """ Create a :class:Connection object
-
-    :param servers:
-        either a string in the form of '<hostname>:<port>'
-        or a list of servers in the form of ['<hostname>:<port>', '...']
-    :param timeout:
-        (optional)
-        define the retry timeout for unreachable servers in seconds
-    :param backoff_factor:
-        (optional)
-        define the retry interval for unreachable servers in seconds
-    :param client:
-        (optional - for testing)
-        client used to communicate with crate.
-    :param verify_ssl_cert:
-        if set to ``True`` verify the servers SSL server certificate.
-        defaults to ``False``
-    :param ca_cert:
-        a path to a CA certificate to use when verifying the SSL server
-        certificate.
-    :param error_trace:
-        if set to ``True`` return a whole stacktrace of any server error if
-        one occurs
-    :param cert_file:
-        a path to the client certificate to present to the server.
-    :param key_file:
-        a path to the client key to use when communicating with the server.
-    :param username:
-        the username in the database.
-    :param password:
-        the password of the user in the database.
-    :param pool_size:
-        (optional)
-        Number of connections to save that can be reused.
-        More than 1 is useful in multithreaded situations.
-    >>> connect(['host1:4200', 'host2:4200'])
-    <Connection <Client ['http://host1:4200', 'http://host2:4200']>>
-    """
-    return Connection(servers=servers,
-                      timeout=timeout,
-                      backoff_factor=backoff_factor,
-                      pool_size=pool_size,
-                      client=client,
-                      verify_ssl_cert=verify_ssl_cert,
-                      ca_cert=ca_cert,
-                      error_trace=error_trace,
-                      cert_file=cert_file,
-                      key_file=key_file,
-                      username=username,
-                      password=password,
-                      schema=schema)
+# For backwards compatibility and not to break existing imports
+connect = Connection

src/crate/client/doctests/blob.txt

@@ -7,7 +7,7 @@ bloba API.
 
 Create a connection::
 
-    >>> from crate.client.connection import connect
+    >>> from crate.client import connect
     >>> client = connect([crate_host])
 
 Get a blob container::