[clusteragent/autoscaling] Use PodWatcher to update current replicas in status

[jennchenn]

What does this PR do?

Use PodWatcher to update current replica count in DatadogPodAutoscaler status instead of Horizontal /scale sub-resource.

Motivation

If horizontal scaling has no changes/is not activated, the current replica count may be inaccurate. This change moves the logic to update the number of current replicas to the controller loop to ensure it is as up-to-date as possible.

Additional Notes

Possible Drawbacks / Trade-offs

During testing, sometimes some lag between when PodWatcher information is updated and when a scaling action happens exists. This can lead to a maximum of 5 min of inconsistent number of replicas being reported (info is fixed on next run of the controller loop).

There is currently duplicated logic between the controller and vertical controller to get the number of pods - would it make sense to combine these two to avoid calling podWatcher.GetPodsForOwner twice? The now duplicated logic:

datadog-agent/pkg/clusteragent/autoscaling/workload/controller_vertical.go

Lines 70 to 84 in 2f3053a

	targetGVK, err := autoscalerInternal.TargetGVK()
	if err != nil {
	autoscalerInternal.SetError(err)
	return autoscaling.NoRequeue, err
	}
	// Get the pod owner from the workload
	target := NamespacedPodOwner{
	Namespace: autoscalerInternal.Namespace(),
	Name: autoscalerInternal.Spec().TargetRef.Name,
	Kind: targetGVK.Kind,
	}
	// Get the pods for the pod owner
	pods := u.podWatcher.GetPodsForOwner(target)

Describe how to test/QA your changes

1. Run an autoscaling workload
2. Verify that the Current Replicas count in the status is updated after scaling actions (even if horizontal scaling is disabled)

[pr-commenter]

Test changes on VM

Use this command from test-infra-definitions to manually test this PR changes on a VM:

inv create-vm --pipeline-id=43652514 --os-family=ubuntu

Note: This applies to commit 734b934

[agent-platform-auto-pr]

[Fast Unit Tests Report]

On pipeline 43652514 (CI Visibility). The following jobs did not run any unit tests:

Jobs:

• tests_flavor_dogstatsd_deb-x64
• tests_flavor_heroku_deb-x64
• tests_flavor_iot_deb-x64

If you modified Go files and expected unit tests to run in these jobs, please double check the job logs. If you think tests should have been executed reach out to #agent-devx-help

[pr-commenter]

Regression Detector

Regression Detector Results

Run ID: ba351497-4dbb-466e-9361-1ff7da428ecf Metrics dashboard Target profiles

Baseline: 6b60c2c
Comparison: 734b934

Performance changes are noted in the perf column of each table:

• ✅ = significantly better comparison variant performance
• ❌ = significantly worse comparison variant performance
• ➖ = no significant change in performance

No significant changes in experiment optimization goals

Confidence level: 90.00%
Effect size tolerance: |Δ mean %| ≥ 5.00%

There were no significant changes in experiment optimization goals at this confidence level and effect size tolerance.

Fine details of change detection per experiment

perf	experiment	goal	Δ mean %	Δ mean % CI	trials	links
➖	tcp_syslog_to_blackhole	ingress throughput	+7.30	[-5.85, +20.46]	1	Logs
➖	file_tree	memory utilization	+1.23	[+1.15, +1.31]	1	Logs
➖	basic_py_check	% cpu utilization	+0.74	[-1.98, +3.46]	1	Logs
➖	pycheck_lots_of_tags	% cpu utilization	+0.70	[-1.87, +3.27]	1	Logs
➖	tcp_dd_logs_filter_exclude	ingress throughput	-0.00	[-0.01, +0.01]	1	Logs
➖	uds_dogstatsd_to_api	ingress throughput	-0.00	[-0.00, +0.00]	1	Logs
➖	idle	memory utilization	-0.13	[-0.17, -0.10]	1	Logs
➖	otel_to_otel_logs	ingress throughput	-0.48	[-1.28, +0.33]	1	Logs
➖	uds_dogstatsd_to_api_cpu	% cpu utilization	-0.73	[-1.51, +0.05]	1	Logs

Bounds Checks

perf	experiment	bounds_check_name	replicates_passed
✅	idle	memory_usage	10/10

Explanation

A regression test is an A/B test of target performance in a repeatable rig, where "performance" is measured as "comparison variant minus baseline variant" for an optimization goal (e.g., ingress throughput). Due to intrinsic variability in measuring that goal, we can only estimate its mean value for each experiment; we report uncertainty in that value as a 90.00% confidence interval denoted "Δ mean % CI".

For each experiment, we decide whether a change in performance is a "regression" -- a change worth investigating further -- if all of the following criteria are true:

1. Its estimated |Δ mean %| ≥ 5.00%, indicating the change is big enough to merit a closer look.

2. Its 90.00% confidence interval "Δ mean % CI" does not contain zero, indicating that if our statistical model is accurate, there is at least a 90.00% chance there is a difference in performance between baseline and comparison variants.

3. Its configuration does not mark it "erratic".

[vboulineau]

Yes we should dedupe the code and give the necessary info to vertical controller.

[vboulineau]

In that case you'd miss status update

[jennchenn]

Added a check to update status and return the error at the end; priority for error during scaling > error to get target > error when updating status

[vboulineau]

Actually if get a targetErr, it's not worth going further, everything is going to fail.

[jennchenn]

Makes sense, I can wait for the changes from #28723

datadog-agent/pkg/clusteragent/autoscaling/workload/controller.go

Lines 413 to 427 in f35a065

	func (c *Controller) updateAutoscalerStatusAndUnlock(ctx context.Context, key, ns, name string, err error, podAutoscalerInternal model.PodAutoscalerInternal, podAutoscaler *datadoghq.DatadogPodAutoscaler) error {
	// Update status based on latest state
	statusErr := c.updatePodAutoscalerStatus(ctx, podAutoscalerInternal, podAutoscaler)
	if statusErr != nil {
	log.Errorf("Failed to update status for PodAutoscaler: %s/%s, err: %v", ns, name, statusErr)
	// We want to return the status error if none to count in the requeue retries.
	if err == nil {
	err = statusErr
	}
	}
	c.store.UnlockSet(key, podAutoscalerInternal, c.ID)
	return err
	}

to be merged so I can make a status update and return early once targetErr is encountered

[vboulineau]

Functional error, will be reflected in status, we should not increase number of errors logs in DCA for functional error (usually mistake in name)

[jennchenn]

/merge

[dd-devflow]

🚂 MergeQueue: pull request added to the queue

The median merge time in main is 23m.

Use /merge -c to cancel this operation!