From d5de1f9d114808ef415e799c687efec2ea0f56f0 Mon Sep 17 00:00:00 2001
From: Alan Rominger <arominge@redhat.com>
Date: Mon, 20 Feb 2023 09:56:38 -0500
Subject: [PATCH] Make use of new keepalive messages from ansible-runner

Make setting API configurable and process keepalive events
  when seen in the event callback

Use env var in pod spec and make it specific to K8S
---
 awx/main/conf.py           | 10 ++++++++++
 awx/main/tasks/callback.py |  2 ++
 awx/main/tasks/receptor.py |  4 ++++
 awx/settings/defaults.py   |  5 +++++
 4 files changed, 21 insertions(+)

diff --git a/awx/main/conf.py b/awx/main/conf.py
index dab0543a1a..99b995b113 100644
--- a/awx/main/conf.py
+++ b/awx/main/conf.py
@@ -282,6 +282,16 @@ register(
     placeholder={'HTTP_PROXY': 'myproxy.local:8080'},
 )
 
+register(
+    'AWX_RUNNER_KEEPALIVE_SECONDS',
+    field_class=fields.IntegerField,
+    label=_('K8S Ansible Runner Keep-Alive Message Interval'),
+    help_text=_('Only applies to K8S deployments and container_group jobs. If not 0, send a message every so-many seconds to keep connection open.'),
+    category=_('Jobs'),
+    category_slug='jobs',
+    placeholder=240,  # intended to be under common 5 minute idle timeout
+)
+
 register(
     'GALAXY_TASK_ENV',
     field_class=fields.KeyValueField,
diff --git a/awx/main/tasks/callback.py b/awx/main/tasks/callback.py
index 92bfc40368..0046d07d82 100644
--- a/awx/main/tasks/callback.py
+++ b/awx/main/tasks/callback.py
@@ -85,6 +85,8 @@ class RunnerCallback:
         # which generate job events from two 'streams':
         # ansible-inventory and the awx.main.commands.inventory_import
         # logger
+        if event_data.get('event') == 'keepalive':
+            return
 
         if event_data.get(self.event_data_key, None):
             if self.event_data_key != 'job_id':
diff --git a/awx/main/tasks/receptor.py b/awx/main/tasks/receptor.py
index 006c805943..9cb4d49efe 100644
--- a/awx/main/tasks/receptor.py
+++ b/awx/main/tasks/receptor.py
@@ -526,6 +526,10 @@ class AWXReceptorJob:
         pod_spec['spec']['containers'][0]['image'] = ee.image
         pod_spec['spec']['containers'][0]['args'] = ['ansible-runner', 'worker', '--private-data-dir=/runner']
 
+        if settings.AWX_RUNNER_KEEPALIVE_SECONDS:
+            pod_spec['spec']['containers'][0].setdefault('env', [])
+            pod_spec['spec']['containers'][0]['env'].append({'name': 'ANSIBLE_RUNNER_KEEPALIVE_SECONDS', 'value': str(settings.AWX_RUNNER_KEEPALIVE_SECONDS)})
+
         # Enforce EE Pull Policy
         pull_options = {"always": "Always", "missing": "IfNotPresent", "never": "Never"}
         if self.task and self.task.instance.execution_environment:
diff --git a/awx/settings/defaults.py b/awx/settings/defaults.py
index 4d18540bcd..74a36b3e2d 100644
--- a/awx/settings/defaults.py
+++ b/awx/settings/defaults.py
@@ -929,6 +929,11 @@ AWX_RUNNER_OMIT_ENV_FILES = True
 # Allow ansible-runner to save ansible output (may cause performance issues)
 AWX_RUNNER_SUPPRESS_OUTPUT_FILE = True
 
+# https://github.com/ansible/ansible-runner/pull/1191/files
+# Interval in seconds between the last message and keep-alive messages that
+# ansible-runner will send
+AWX_RUNNER_KEEPALIVE_SECONDS = 0
+
 # Delete completed work units in receptor
 RECEPTOR_RELEASE_WORK = True