Consume control capacity (#11665)

* Select control node before start task

Consume capacity on control nodes for controlling tasks and consider
remainging capacity on control nodes before selecting them.

This depends on the requirement that control and hybrid nodes should all
be in the instance group named 'controlplane'. Many tests do not satisfy that
requirement. I'll update the tests in another commit.

* update tests to use controlplane

We don't start any tasks if we don't have a controlplane instance group

Due to updates to fixtures, update tests to set node type and capacity
explicitly so they get expected result.

* Fixes for accounting of control capacity consumed

Update method is used to account for currently consumed capacity for
instance groups in the in-memory capacity tracking data structure we initialize in
after_lock_init and then update via calculate_capacity_consumed (both in
task_manager.py)

Also update fit_task_to_instance to consider control impact on instances

Trust that these functions do the right thing looking for a
node with capacity, and cut out redundant check for the whole group's
capacity per Alan's reccomendation.

* Refactor now redundant code

Deal with control type tasks before we loop over the preferred instance
groups, which cuts out the need for some redundant logic.

Also, fix a bug where I was missing assigning the execution node in one case!

* set job explanation on tasks that need capacity

move the job explanation for jobs that need capacity to a function
so we can re-use it in the three places we need it.

* project updates always run on the controlplane

Instance group ordering makes no sense on project updates because they
always need to run on the control plane.

Also, since hybrid nodes should always run the control processes for the
jobs running on them as execution nodes, account for this when looking for a
execution node.

* fix misleading message

the variables and wording were both misleading, fix to be more accurate
description in the two different cases where this log may be emitted.

* use settings correctly

use settings.DEFAULT_CONTROL_PLANE_QUEUE_NAME instead of a hardcoded
name
cache the controlplane_ig object during the after lock init to avoid
an uneccesary query
eliminate mistakenly duplicated AWX_CONTROL_PLANE_TASK_IMPACT and use
only AWX_CONTROL_NODE_TASK_IMPACT

* add test for control capacity consumption

add test to verify that when there are 2 jobs and only capacity for one
that one will move into waiting and the other stays in pending

* add test for hybrid node capacity consumption

assert that the hybrid node is used for both control and execution and
capacity is deducted correctly

* add test for task.capacity_type = control

Test that control type tasks have the right capacity consumed and
get assigned to the right instance group

Also fix lint in the tests

* jobs_running not accurate for control nodes

We can either NOT use "idle instances" for control nodes, or we need
to update the jobs_running property on the Instance model to count
jobs where the node is the controller_node.

I didn't do that because it may be an expensive query, and it would be
hard to make it match with jobs_running on the InstanceGroup which
filters on tasks assigned to the instance group.

This change chooses to stop considering "idle" control nodes an option,
since we can't acurrately identify them.

The way things are without any change, is we are continuing to over consume capacity on control nodes
because this method sees all control nodes as "idle" at the beginning
of the task manager run, and then only counts jobs started in that run
in the in-memory tracking. So jobs which last over a number of task
manager runs build up consuming capacity, which is accurately reported
via Instance.consumed_capacity

* Reduce default task impact for control nodes

This is something we can experiment with as far as what users
want at install time, but start with just 1 for now.

* update capacity docs

Describe usage of the new setting and the concept of control impact.

Co-authored-by: Alan Rominger <arominge@redhat.com>
Co-authored-by: Rebeccah <rhunter@redhat.com>

awx/main/scheduler/task_manager.py

@@ -70,6 +70,7 @@ class TaskManager:
         """
         instances = Instance.objects.filter(hostname__isnull=False, enabled=True).exclude(node_type='hop')
         self.real_instances = {i.hostname: i for i in instances}
+        self.controlplane_ig = None
 
         instances_partial = [
             SimpleNamespace(
@@ -86,6 +87,8 @@ class TaskManager:
         instances_by_hostname = {i.hostname: i for i in instances_partial}
 
         for rampart_group in InstanceGroup.objects.prefetch_related('instances'):
+            if rampart_group.name == settings.DEFAULT_CONTROL_PLANE_QUEUE_NAME:
+                self.controlplane_ig = rampart_group
             self.graph[rampart_group.name] = dict(
                 graph=DependencyGraph(),
                 execution_capacity=0,
@@ -283,39 +286,13 @@ class TaskManager:
                 task.send_notification_templates('running')
                 logger.debug('Transitioning %s to running status.', task.log_format)
                 schedule_task_manager()
-            elif rampart_group.is_container_group:
-                task.instance_group = rampart_group
-                if task.capacity_type == 'execution':
-                    # find one real, non-containerized instance with capacity to
-                    # act as the controller for k8s API interaction
-                    try:
-                        task.controller_node = Instance.choose_online_control_plane_node()
-                        task.log_lifecycle("controller_node_chosen")
-                    except IndexError:
-                        logger.warning("No control plane nodes available to run containerized job {}".format(task.log_format))
-                        return
-                else:
-                    # project updates and system jobs don't *actually* run in pods, so
-                    # just pick *any* non-containerized host and use it as the execution node
-                    task.execution_node = Instance.choose_online_control_plane_node()
-                    task.log_lifecycle("execution_node_chosen")
-                    logger.debug('Submitting containerized {} to queue {}.'.format(task.log_format, task.execution_node))
+            # at this point we already have control/execution nodes selected for the following cases
             else:
                 task.instance_group = rampart_group
-                task.execution_node = instance.hostname
-                task.log_lifecycle("execution_node_chosen")
-                if instance.node_type == 'execution':
-                    try:
-                        task.controller_node = Instance.choose_online_control_plane_node()
-                        task.log_lifecycle("controller_node_chosen")
-                    except IndexError:
-                        logger.warning("No control plane nodes available to manage {}".format(task.log_format))
-                        return
-                else:
-                    # control plane nodes will manage jobs locally for performance and resilience
-                    task.controller_node = task.execution_node
-                    task.log_lifecycle("controller_node_chosen")
-                logger.debug('Submitting job {} to queue {} controlled by {}.'.format(task.log_format, task.execution_node, task.controller_node))
+                execution_node_msg = f' and execution node {task.execution_node}' if task.execution_node else ''
+                logger.debug(
+                    f'Submitting job {task.log_format} controlled by {task.controller_node} to instance group {rampart_group.name}{execution_node_msg}.'
+                )
             with disable_activity_stream():
                 task.celery_task_id = str(uuid.uuid4())
                 task.save()
@@ -323,6 +300,13 @@ class TaskManager:
 
             if rampart_group is not None:
                 self.consume_capacity(task, rampart_group.name, instance=instance)
+            if task.controller_node:
+                self.consume_capacity(
+                    task,
+                    settings.DEFAULT_CONTROL_PLANE_QUEUE_NAME,
+                    instance=self.real_instances[task.controller_node],
+                    impact=settings.AWX_CONTROL_NODE_TASK_IMPACT,
+                )
 
         def post_commit():
             if task.status != 'failed' and type(task) is not WorkflowJob:
@@ -497,9 +481,10 @@ class TaskManager:
                         task.job_explanation = job_explanation
                         tasks_to_update_job_explanation.append(task)
                 continue
-            preferred_instance_groups = task.preferred_instance_groups
 
             found_acceptable_queue = False
+            preferred_instance_groups = task.preferred_instance_groups
+
             if isinstance(task, WorkflowJob):
                 if task.unified_job_template_id in running_workflow_templates:
                     if not task.allow_simultaneous:
@@ -510,9 +495,36 @@ class TaskManager:
                 self.start_task(task, None, task.get_jobs_fail_chain(), None)
                 continue
 
+            # Determine if there is control capacity for the task
+            if task.capacity_type == 'control':
+                control_impact = task.task_impact + settings.AWX_CONTROL_NODE_TASK_IMPACT
+            else:
+                control_impact = settings.AWX_CONTROL_NODE_TASK_IMPACT
+            control_instance = InstanceGroup.fit_task_to_most_remaining_capacity_instance(
+                task, self.graph[settings.DEFAULT_CONTROL_PLANE_QUEUE_NAME]['instances'], impact=control_impact, capacity_type='control'
+            )
+            if not control_instance:
+                self.task_needs_capacity(task, tasks_to_update_job_explanation)
+                logger.debug(f"Skipping task {task.log_format} in pending, not enough capacity left on controlplane to control new tasks")
+                continue
+
+            task.controller_node = control_instance.hostname
+
+            # All task.capacity_type == 'control' jobs should run on control plane, no need to loop over instance groups
+            if task.capacity_type == 'control':
+                task.execution_node = control_instance.hostname
+                control_instance.remaining_capacity = max(0, control_instance.remaining_capacity - control_impact)
+                control_instance.jobs_running += 1
+                self.graph[settings.DEFAULT_CONTROL_PLANE_QUEUE_NAME]['graph'].add_job(task)
+                execution_instance = self.real_instances[control_instance.hostname]
+                self.start_task(task, self.controlplane_ig, task.get_jobs_fail_chain(), execution_instance)
+                found_acceptable_queue = True
+                continue
+
             for rampart_group in preferred_instance_groups:
-                if task.capacity_type == 'execution' and rampart_group.is_container_group:
-                    self.graph[rampart_group.name]['graph'].add_job(task)
+                if rampart_group.is_container_group:
+                    control_instance.jobs_running += 1
+                    self.graph[settings.DEFAULT_CONTROL_PLANE_QUEUE_NAME]['graph'].add_job(task)
                     self.start_task(task, rampart_group, task.get_jobs_fail_chain(), None)
                     found_acceptable_queue = True
                     break
@@ -521,28 +533,32 @@ class TaskManager:
                 if settings.IS_K8S and task.capacity_type == 'execution':
                     logger.debug("Skipping group {}, task cannot run on control plane".format(rampart_group.name))
                     continue
-
-                remaining_capacity = self.get_remaining_capacity(rampart_group.name, capacity_type=task.capacity_type)
-                if task.task_impact > 0 and remaining_capacity <= 0:
-                    logger.debug("Skipping group {}, remaining_capacity {} <= 0".format(rampart_group.name, remaining_capacity))
-                    continue
-
+                # at this point we know the instance group is NOT a container group
+                # because if it was, it would have started the task and broke out of the loop.
                 execution_instance = InstanceGroup.fit_task_to_most_remaining_capacity_instance(
-                    task, self.graph[rampart_group.name]['instances']
+                    task, self.graph[rampart_group.name]['instances'], add_hybrid_control_cost=True
                 ) or InstanceGroup.find_largest_idle_instance(self.graph[rampart_group.name]['instances'], capacity_type=task.capacity_type)
 
-                if execution_instance or rampart_group.is_container_group:
-                    if not rampart_group.is_container_group:
-                        execution_instance.remaining_capacity = max(0, execution_instance.remaining_capacity - task.task_impact)
-                        execution_instance.jobs_running += 1
-                        logger.debug(
-                            "Starting {} in group {} instance {} (remaining_capacity={})".format(
-                                task.log_format, rampart_group.name, execution_instance.hostname, remaining_capacity
-                            )
-                        )
+                if execution_instance:
+                    task.execution_node = execution_instance.hostname
+                    # If our execution instance is a hybrid, prefer to do control tasks there as well.
+                    if execution_instance.node_type == 'hybrid':
+                        control_instance = execution_instance
+                        task.controller_node = execution_instance.hostname
 
-                    if execution_instance:
-                        execution_instance = self.real_instances[execution_instance.hostname]
+                    control_instance.remaining_capacity = max(0, control_instance.remaining_capacity - settings.AWX_CONTROL_NODE_TASK_IMPACT)
+                    task.log_lifecycle("controller_node_chosen")
+                    if control_instance != execution_instance:
+                        control_instance.jobs_running += 1
+                    execution_instance.remaining_capacity = max(0, execution_instance.remaining_capacity - task.task_impact)
+                    execution_instance.jobs_running += 1
+                    task.log_lifecycle("execution_node_chosen")
+                    logger.debug(
+                        "Starting {} in group {} instance {} (remaining_capacity={})".format(
+                            task.log_format, rampart_group.name, execution_instance.hostname, execution_instance.remaining_capacity
+                        )
+                    )
+                    execution_instance = self.real_instances[execution_instance.hostname]
                     self.graph[rampart_group.name]['graph'].add_job(task)
                     self.start_task(task, rampart_group, task.get_jobs_fail_chain(), execution_instance)
                     found_acceptable_queue = True
@@ -554,18 +570,21 @@ class TaskManager:
                         )
                     )
             if not found_acceptable_queue:
-                task.log_lifecycle("needs_capacity")
-                job_explanation = gettext_noop("This job is not ready to start because there is not enough available capacity.")
-                if task.job_explanation != job_explanation:
-                    if task.created < (tz_now() - self.time_delta_job_explanation):
-                        # Many launched jobs are immediately blocked, but most blocks will resolve in a few seconds.
-                        # Therefore we should only update the job_explanation after some time has elapsed to
-                        # prevent excessive task saves.
-                        task.job_explanation = job_explanation
-                        tasks_to_update_job_explanation.append(task)
-                logger.debug("{} couldn't be scheduled on graph, waiting for next cycle".format(task.log_format))
+                self.task_needs_capacity(task, tasks_to_update_job_explanation)
         UnifiedJob.objects.bulk_update(tasks_to_update_job_explanation, ['job_explanation'])
 
+    def task_needs_capacity(self, task, tasks_to_update_job_explanation):
+        task.log_lifecycle("needs_capacity")
+        job_explanation = gettext_noop("This job is not ready to start because there is not enough available capacity.")
+        if task.job_explanation != job_explanation:
+            if task.created < (tz_now() - self.time_delta_job_explanation):
+                # Many launched jobs are immediately blocked, but most blocks will resolve in a few seconds.
+                # Therefore we should only update the job_explanation after some time has elapsed to
+                # prevent excessive task saves.
+                task.job_explanation = job_explanation
+                tasks_to_update_job_explanation.append(task)
+        logger.debug("{} couldn't be scheduled on graph, waiting for next cycle".format(task.log_format))
+
     def timeout_approval_node(self):
         workflow_approvals = WorkflowApproval.objects.filter(status='pending')
         now = tz_now()
@@ -600,16 +619,17 @@ class TaskManager:
     def calculate_capacity_consumed(self, tasks):
         self.graph = InstanceGroup.objects.capacity_values(tasks=tasks, graph=self.graph)
 
-    def consume_capacity(self, task, instance_group, instance=None):
+    def consume_capacity(self, task, instance_group, instance=None, impact=None):
+        impact = impact if impact else task.task_impact
         logger.debug(
             '{} consumed {} capacity units from {} with prior total of {}'.format(
-                task.log_format, task.task_impact, instance_group, self.graph[instance_group]['consumed_capacity']
+                task.log_format, impact, instance_group, self.graph[instance_group]['consumed_capacity']
             )
         )
-        self.graph[instance_group]['consumed_capacity'] += task.task_impact
+        self.graph[instance_group]['consumed_capacity'] += impact
         for capacity_type in ('control', 'execution'):
             if instance is None or instance.node_type in ('hybrid', capacity_type):
-                self.graph[instance_group][f'consumed_{capacity_type}_capacity'] += task.task_impact
+                self.graph[instance_group][f'consumed_{capacity_type}_capacity'] += impact
 
     def get_remaining_capacity(self, instance_group, capacity_type='execution'):
         return self.graph[instance_group][f'{capacity_type}_capacity'] - self.graph[instance_group][f'consumed_{capacity_type}_capacity']