Change default of [kubernetes] enable_tcp_keepalive to True (#15338)

We've seen instances of connection resets happening, particularly in
Azure, that are remedied by enabling tcp_keepalive. Enabling it by
default should be safe and sane regardless of where we are running.

UPDATING.md

@@ -74,7 +74,7 @@ https://developers.google.com/style/inclusive-documentation
 
 Moved the pod launcher from `airflow.kubernetes.pod_launcher` to `airflow.providers.cncf.kubernetes.utils.pod_launcher`
 
-This will alow users to update the pod_launcher for the KubernetesPodOperator without requiring an airflow upgrade
+This will allow users to update the pod_launcher for the KubernetesPodOperator without requiring an airflow upgrade
 
 ### Default `[webserver] worker_refresh_interval` is changed to `6000` seconds
 
@@ -91,6 +91,10 @@ serve as a DagBag cache burst time.
 
 The `default_queue` configuration option has been moved from `[celery]` section to `[operators]` section to allow for re-use between different executors.
 
+### Default `[kubernetes] enable_tcp_keepalive` is changed to `True`
+
+This allows Airflow to work more reliably with some environments (like Azure) by default.
+
 ## Airflow 2.0.1
 
 ### Permission to view Airflow Configurations has been removed from `User` and `Viewer` role

airflow/config_templates/config.yml

@@ -2116,7 +2116,7 @@
       version_added: ~
       type: boolean
       example: ~
-      default: "False"
+      default: "True"
     - name: tcp_keep_idle
       description: |
         When the `enable_tcp_keepalive` option is enabled, TCP probes a connection that has

airflow/config_templates/default_airflow.cfg

@@ -1045,7 +1045,7 @@ delete_option_kwargs =
 
 # Enables TCP keepalive mechanism. This prevents Kubernetes API requests to hang indefinitely
 # when idle connection is time-outed on services like cloud load balancers or firewalls.
-enable_tcp_keepalive = False
+enable_tcp_keepalive = True
 
 # When the `enable_tcp_keepalive` option is enabled, TCP probes a connection that has
 # been idle for `tcp_keep_idle` seconds.

airflow/kubernetes/kube_client.py

@@ -85,9 +85,9 @@ def _enable_tcp_keepalive() -> None:
 
     from urllib3.connection import HTTPConnection, HTTPSConnection
 
-    tcp_keep_idle = conf.getint('kubernetes', 'tcp_keep_idle', fallback=120)
-    tcp_keep_intvl = conf.getint('kubernetes', 'tcp_keep_intvl', fallback=30)
-    tcp_keep_cnt = conf.getint('kubernetes', 'tcp_keep_cnt', fallback=6)
+    tcp_keep_idle = conf.getint('kubernetes', 'tcp_keep_idle')
+    tcp_keep_intvl = conf.getint('kubernetes', 'tcp_keep_intvl')
+    tcp_keep_cnt = conf.getint('kubernetes', 'tcp_keep_cnt')
 
     socket_options = [
         (socket.SOL_SOCKET, socket.SO_KEEPALIVE, 1),
@@ -125,10 +125,10 @@ def get_kube_client(
         if config_file is None:
             config_file = conf.get('kubernetes', 'config_file', fallback=None)
 
-    if conf.getboolean('kubernetes', 'enable_tcp_keepalive', fallback=False):
+    if conf.getboolean('kubernetes', 'enable_tcp_keepalive'):
         _enable_tcp_keepalive()
 
-    if not conf.getboolean('kubernetes', 'verify_ssl', fallback=True):
+    if not conf.getboolean('kubernetes', 'verify_ssl'):
         _disable_verify_ssl()
 
     client_conf = _get_kube_config(in_cluster, cluster_context, config_file)