[AMLII-1994] Improve expected_tags_duration for metrics

[andrewqian2001datadog]

What does this PR do?

Fixes configurable host tags that are added to metrics expected_tags_duration is configured.

This PR changes the logic for adding host tags and moves it after aggregation occurs so that a metric's context hash does not change and the aggregator does not mistake the same metric as a new one.

Motivation

Issue

Describe how to test/QA your changes

To verify that configurable host tags work as expected

1. Add the following settings to datadog.yaml

tags: "foo:bar"
expected_tags_duration: 1m
log_level: 'debug'
log_payloads: true 

2. Start the agent, and tail the logs.
3. grep for tags foo:bar
4. For the first ~1m of the agent running you should see foo:bar tags on every metric logged in the payloads.
5. After ~1m passes, you should no longer see these tags.

To verify that the bug was resolved

1. Use the same datadog.yaml settings as before
2. Create a test file
   dev/dist/checks.d/testcheck.py
3. Inside the file, configure the following check

from datadog_checks.base import AgentCheck

class ServerCheck(AgentCheck):
    inc = 0  

    def check(self, instance):
        ServerCheck.inc += 100  
        self.monotonic_count('brian.test.check.count', ServerCheck.inc, flush_first_value=True)

4. Run the agent then the check
5. Verify that a spike does not occur in 1m, here is how the spike used to look

QA steps are also given in the ticket

Additional Notes

I've benchmarked these changes with dogstatsd stress tests and displayed the results in this document.

Notebook

RSS pod memory usage

Old host tag feature:

• Tags enabled: 237M average
• Tags disabled: 228M average

New host tag feature:

• Tags enabled: 229M average
• Tags disabled 224M average

CPU usage

Old host tag feature

• Tags enabled: 1.18 cores average
• Tags disabled: 1.15 cores average

New host tag feature

• Tags enabled: 1.16 cores average
• Tags disabled: 1.11 cores average

Results:
3.7 % reduction in memory usage when tags are enabled while using the new host tag feature
1.8 % reduction in memory usage when tags are disabled while using the new host tag feature

1.7 % reduction in average cores used when tags are enabled while using the new host tag feature
3.5 % reduction in average cores used when tags are disabled while using the new host tag feature

The old version also has a CPU drop when the host tags expire which the new version does not have.

[agent-platform-auto-pr]

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=46767706 --os-family=ubuntu

Note: This applies to commit 35be39e

[cit-pr-commenter]

Regression Detector

Regression Detector Results

Run ID: 85573b84-e3c0-4a6a-a404-00503f3dda40 Metrics dashboard Target profiles

Baseline: 03a353c
Comparison: 1ef8fcd

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
➖	idle_all_features	memory utilization	+1.42	[+1.32, +1.52]	1	Logs bounds checks dashboard
➖	tcp_syslog_to_blackhole	ingress throughput	+1.06	[+0.97, +1.14]	1	Logs
➖	pycheck_lots_of_tags	% cpu utilization	+0.50	[-2.08, +3.08]	1	Logs
➖	file_to_blackhole_500ms_latency	egress throughput	+0.08	[-0.16, +0.33]	1	Logs
➖	file_tree	memory utilization	+0.07	[-0.06, +0.20]	1	Logs
➖	file_to_blackhole_0ms_latency	egress throughput	+0.01	[-0.32, +0.34]	1	Logs
➖	file_to_blackhole_100ms_latency	egress throughput	+0.01	[-0.21, +0.23]	1	Logs
➖	basic_py_check	% cpu utilization	+0.00	[-2.97, +2.97]	1	Logs
➖	tcp_dd_logs_filter_exclude	ingress throughput	-0.00	[-0.01, +0.01]	1	Logs
➖	file_to_blackhole_300ms_latency	egress throughput	-0.00	[-0.19, +0.18]	1	Logs
➖	uds_dogstatsd_to_api	ingress throughput	-0.01	[-0.08, +0.07]	1	Logs
➖	uds_dogstatsd_to_api_cpu	% cpu utilization	-0.10	[-0.83, +0.64]	1	Logs
➖	file_to_blackhole_1000ms_latency	egress throughput	-0.10	[-0.59, +0.38]	1	Logs
➖	idle	memory utilization	-0.65	[-0.70, -0.61]	1	Logs bounds checks dashboard
➖	otel_to_otel_logs	ingress throughput	-0.88	[-1.69, -0.08]	1	Logs

Bounds Checks

perf	experiment	bounds_check_name	replicates_passed
✅	file_to_blackhole_0ms_latency	memory_usage	10/10
✅	file_to_blackhole_1000ms_latency	memory_usage	10/10
✅	file_to_blackhole_100ms_latency	memory_usage	10/10
✅	file_to_blackhole_300ms_latency	memory_usage	10/10
✅	file_to_blackhole_500ms_latency	memory_usage	10/10
✅	idle	memory_usage	10/10
✅	idle_all_features	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".

[estherk15]

Non-blocking suggestion! Approved for docs.

[gh123man]

NIT: Can we change the title to something like: "Improve expected_tags_duration for metrics"

[gh123man]

This looks very promising and the numbers look great 👌

There are a few more files you can delete things from introduced in the old PR:

This function and it's usage:

datadog-agent/comp/core/tagger/taggerimpl/collectors/workloadmeta_extract.go

Line 297 in 1003799

func (c *WorkloadMetaCollector) handleHostTags(ev workloadmeta.Event) []*types.TagInfo {

This const and it's usage:

datadog-agent/comp/core/workloadmeta/def/types.go

Line 50 in 1003799

KindHost Kind = "host"

[gh123man]

Nice work!

[gh123man]

Suggested change

	sketch.Tags = tagset.CombineCompositeTagsAndSlice(sketch.Tags, hostTagProvider.GetHostTags())
	sketch.Tags = tagset.CombineCompositeTagsAndSlice(sketch.Tags, hostTags)

NIT: to avoid calling GetHostTags twice.
(same for line 109 above)

[clamoriniere]

HI @andrewqian2001datadog

I have a question about the approach choose to fix the reported issue: why we don't keep the information into workloadmeta? I fill like in some environment (host less) it can still be interesting to have the logic, since the host-metadata payload is never sent.

also keeping host information inside workloadmeta could solve other issue in the future like, refreshing faster host tags when a new host tag appears after the agent has already sent the host metadata payload.

[clamoriniere]

/lgtm for @DataDog/container-platform files

[andrewqian2001datadog]

/merge

[dd-devflow]

🚂 MergeQueue: pull request added to the queue

The median merge time in main is 22m.

Use /merge -c to cancel this operation!