Add gensim-eks scenario: end-to-end episode runner with full observability

[scottopell]

What does this PR do?

Adds a new e2e-framework scenario (aws/gensim-eks) that provisions a persistent EKS cluster and runs GenSim episodes end-to-end. Supports two modes:

• record-parquet (default): Collects observer data to parquet files for offline analysis, uploads to S3
• live-anomaly-detection: Runs live edge anomaly detection with all correlators enabled, sends detected anomaly events to Datadog via the Events v2 API

For each episode, the scenario deploys application services, installs a fully-configured Datadog agent, executes fault injection cycles, and collects results (parquet files or live anomaly events depending on mode).

Motivation

GenSim episodes are self-contained incident simulations -- each ships a Helm chart with application services, a play-episode.sh fault injection script, and (in 261 of 267 episodes) a stub Datadog agent Deployment. To evaluate the observer against these episodes at scale, we need infrastructure that:

1. Replaces the stub agent with a custom observer-instrumented build
2. Autodiscovers per-episode integration check configs (redis, postgres, etc.) via pod annotations
3. Enables full observability -- APM traces, logs, DogStatsD, integration checks -- not just internal agent metrics
4. Runs multiple episodes serially with clean teardown between each
5. Supports both offline parquet collection and live anomaly detection modes

Data flow

On your laptop (dda inv aws.eks.gensim.submit --image=... --episodes=A:s1,B:s2 --mode=live-anomaly-detection):

Pulumi reads each episode's play-episode.sh, scenario YAML, and Helm chart from your local GENSIM_REPO_PATH, packages them into per-episode ConfigMaps, renders agent-values.yaml.tmpl with the target image and mode, and creates an orchestrator Job. Your laptop is done.

In the EKS cluster (orchestrator Job, alpine/k8s:1.31.0):

for each episode:

  # EPISODE: helm install episode chart with --set agent.enabled=false
  #          (suppresses stub agent resources; post-renderer fixes imagePullPolicy)

  # AGENT: helm install datadog-agent with:
  #   - agent-values.yaml (APM, logs, DogStatsD, process agent, cluster agent,
  #     admission controller, observer config based on mode)
  #   (autodiscovery annotations on episode pods handle check configs)

  # RUN: play-episode.sh executes warmup -> baseline -> fault injection -> cooldown
  #      (20-80 min per episode)

  # COLLECT (record-parquet mode only):
  #   kubectl cp parquet from agent pod
  #   aws s3 cp to $BUCKET/$IMAGE_TAG/$EPISODE--$SCENARIO/gensim-$DATE-$SHA/
  # (live-anomaly-detection mode: anomaly events sent to Datadog in real-time)

  # REPORT:  DD events + metrics via API, update gensim-run-status ConfigMap
  # TEARDOWN: helm uninstall episode + agent, clean workspace

Observer mode configuration

The --mode flag controls what the agent's observer component does:

Config	record-parquet	live-anomaly-detection
observer.recording.enabled	true	false
observer.analysis.enabled	(default)	true
observer.event_reporter.sending_enabled	false	true
Parquet collection + S3 upload	yes	skipped

Key design decisions

• fullnameOverride: datadog-agent -- episode pods hardcode datadog-agent:8125 and datadog-agent:8126 as DNS names for DogStatsD and APM. This makes our Helm chart create Services with the names they expect.
• Single-pass episode deployment -- helm install with --set agent.enabled=false suppresses stub agent resources. A lightweight sed post-renderer patches imagePullPolicy: Never to Always for the few episodes that set it. Autodiscovery annotations on episode pods handle check configs -- no extraction step needed.
• pullPolicy: Always on the agent image -- prevents stale cached images on EKS nodes when the same tag is rebuilt with new code.
• Admission controller with mutateUnlabelled: true -- auto-injects DD_AGENT_HOST into all episode pods without requiring labels.
• Agent values as a Go template (agent-values.yaml.tmpl) -- embedded via go:embed, rendered at Pulumi plan time with target image repo/tag and mode.
• --refresh on pulumi up -- reconciles Pulumi state with cluster reality before planning changes. Prevents silent no-ops when resources are deleted outside Pulumi (e.g. by the orchestrator's helm uninstall during teardown).
• --non-interactive on pulumi up/destroy -- disables the TUI progress display so error output is captured in logs instead of being overwritten by cursor repositioning.
• Content-addressed build cache -- docker-compose.yaml is included in the services hash (alongside the services/ directory) so adding/removing images triggers a rebuild.
• Image reference validation -- renderAgentValues validates the image contains : before splitting into repo:tag, returning a clear error instead of panicking.

Describe how you validated your changes

Record-parquet mode -- 3-episode end-to-end run:

Episode	Parquet Files	Duration	Signals
213_PagerDuty (cassandra)	381	49m	APM + logs + DogStatsD + observer
food-delivery (redis)	235	33m	APM + logs + DogStatsD + redis check + postgres check + observer
353_postmark (dns)	602	84m	APM + logs + DogStatsD + observer

Live-anomaly-detection mode -- 213_PagerDuty episode:

• 20 Datadog events successfully delivered via the event reporter (source:agent-q-branch-observer)
• TimeCluster correlator detected temporal co-occurrence of BOCPD anomalies across trace agent metrics (2-8 anomalies per cluster)
• Surprise correlator detected statistically unusual metric co-occurrences (lift scores 2.0-24.0)
• Monitor received data (DD_ENV propagation working), went to Alert during disruption, recovered to OK
• Clean teardown, no resource leaks

Future checklist: hacks to retire

• awk filter stripping agent resources -- Retired. Episodes now gate agent resources behind agent.enabled: false (gensim-episodes PR [aggregator] Allow aggregator to handle rate metrics from checks #31).
• extract_check_configs with yq -- Retired. Episodes migrated to autodiscovery annotations (gensim-episodes PR [aggregator] Support all existing check metric types #33).
• helm template + kubectl apply instead of helm install -- Retired. With agent.enabled=false, episodes install cleanly via helm install.
• imagePullPolicy: Never -> Always patching (awk filter) -- Retired as an awk filter. Replaced by a 3-line sed post-renderer on helm install.
• Build VM + docker buildx bake + ECR push -- EC2 instance to build episode service images from source and push to ECR. ~150 lines of Pulumi + bash. Retires when: gensim-episodes publishes pre-built images to a shared registry.

Evaluated and kept as-is:

• fullnameOverride: datadog-agent -- load-bearing DNS name that 227 episodes depend on.
• mutateUnlabelled: true -- correct for single-tenant namespace.
• kubelet.tlsVerify: false -- standard EKS workaround, permanent.

Files

File	What
test/e2e-framework/scenarios/aws/gensim-eks/run.go	Scenario: EKS cluster, orchestrator Job, build VM, orchestrator bash script, mode plumbing
test/e2e-framework/scenarios/aws/gensim-eks/agent-values.yaml.tmpl	Agent Helm values template (mode-conditional observer config)
tasks/e2e_framework/aws/gensim_eks.py	Invoke tasks: submit (with --mode), status, destroy, logs
tasks/e2e_framework/deploy.py	Pulumi deploy: --refresh, --non-interactive, pty=False
tasks/e2e_framework/destroy.py	Pulumi destroy: --non-interactive, pty=False

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: dd4d8f599b

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[agent-platform-auto-pr]

Files inventory check summary

File checks results against ancestor 0b457eee:

Results for datadog-agent_7.78.0~devel.git.723.16412a7.pipeline.103341240-1_amd64.deb:

No change detected

[agent-platform-auto-pr]

Static quality checks

✅ Please find below the results from static quality gates
Comparison made with ancestor 0b457ee
📊 Static Quality Gates Dashboard
🔗 SQG Job

30 successful checks with minimal change (< 2 KiB)

	Quality gate	Current Size
✅	agent_deb_amd64	748.919 MiB
✅	agent_deb_amd64_fips	707.285 MiB
✅	agent_heroku_amd64	312.824 MiB
✅	agent_rpm_amd64	748.903 MiB
✅	agent_rpm_amd64_fips	707.268 MiB
✅	agent_rpm_arm64	727.411 MiB
✅	agent_rpm_arm64_fips	688.727 MiB
✅	agent_suse_amd64	748.903 MiB
✅	agent_suse_amd64_fips	707.268 MiB
✅	agent_suse_arm64	727.411 MiB
✅	agent_suse_arm64_fips	688.727 MiB
✅	docker_agent_amd64	809.250 MiB
✅	docker_agent_arm64	812.559 MiB
✅	docker_agent_jmx_amd64	1000.165 MiB
✅	docker_agent_jmx_arm64	992.253 MiB
✅	docker_cluster_agent_amd64	203.688 MiB
✅	docker_cluster_agent_arm64	218.167 MiB
✅	docker_cws_instrumentation_amd64	7.142 MiB
✅	docker_cws_instrumentation_arm64	6.689 MiB
✅	docker_dogstatsd_amd64	39.207 MiB
✅	docker_dogstatsd_arm64	37.445 MiB
✅	dogstatsd_deb_amd64	29.847 MiB
✅	dogstatsd_deb_arm64	28.003 MiB
✅	dogstatsd_rpm_amd64	29.847 MiB
✅	dogstatsd_suse_amd64	29.847 MiB
✅	iot_agent_deb_amd64	43.065 MiB
✅	iot_agent_deb_arm64	40.124 MiB
✅	iot_agent_deb_armhf	40.860 MiB
✅	iot_agent_rpm_amd64	43.066 MiB
✅	iot_agent_suse_amd64	43.066 MiB

On-wire sizes (compressed)

	Quality gate	Change	Size (prev → curr → max)
✅	agent_deb_amd64	-4.41 KiB (0.00% reduction)	174.191 → 174.186 → 177.700
✅	agent_deb_amd64_fips	-21.94 KiB (0.01% reduction)	165.191 → 165.170 → 172.230
✅	agent_heroku_amd64	neutral	74.896 MiB → 79.970
✅	agent_rpm_amd64	+35.9 KiB (0.02% increase)	177.138 → 177.173 → 180.780
✅	agent_rpm_amd64_fips	-15.66 KiB (0.01% reduction)	167.197 → 167.181 → 173.370
✅	agent_rpm_arm64	+41.76 KiB (0.03% increase)	159.608 → 159.648 → 161.610
✅	agent_rpm_arm64_fips	-2.48 KiB (0.00% reduction)	151.234 → 151.231 → 155.910
✅	agent_suse_amd64	+35.9 KiB (0.02% increase)	177.138 → 177.173 → 180.780
✅	agent_suse_amd64_fips	-15.66 KiB (0.01% reduction)	167.197 → 167.181 → 173.370
✅	agent_suse_arm64	+41.76 KiB (0.03% increase)	159.608 → 159.648 → 161.610
✅	agent_suse_arm64_fips	-2.48 KiB (0.00% reduction)	151.234 → 151.231 → 155.910
✅	docker_agent_amd64	neutral	267.206 MiB → 271.240
✅	docker_agent_arm64	+4.08 KiB (0.00% increase)	254.476 → 254.480 → 259.800
✅	docker_agent_jmx_amd64	neutral	335.848 MiB → 339.870
✅	docker_agent_jmx_arm64	neutral	319.126 MiB → 324.390
✅	docker_cluster_agent_amd64	neutral	71.289 MiB → 72.920
✅	docker_cluster_agent_arm64	neutral	66.926 MiB → 68.220
✅	docker_cws_instrumentation_amd64	neutral	2.999 MiB → 3.330
✅	docker_cws_instrumentation_arm64	neutral	2.729 MiB → 3.090
✅	docker_dogstatsd_amd64	neutral	15.155 MiB → 15.820
✅	docker_dogstatsd_arm64	neutral	14.478 MiB → 14.830
✅	dogstatsd_deb_amd64	neutral	7.886 MiB → 8.790
✅	dogstatsd_deb_arm64	neutral	6.772 MiB → 7.710
✅	dogstatsd_rpm_amd64	neutral	7.897 MiB → 8.800
✅	dogstatsd_suse_amd64	neutral	7.897 MiB → 8.800
✅	iot_agent_deb_amd64	neutral	11.348 MiB → 12.040
✅	iot_agent_deb_arm64	neutral	9.664 MiB → 10.450
✅	iot_agent_deb_armhf	neutral	9.897 MiB → 10.620
✅	iot_agent_rpm_amd64	-3.48 KiB (0.03% reduction)	11.369 → 11.366 → 12.060
✅	iot_agent_suse_amd64	-3.48 KiB (0.03% reduction)	11.369 → 11.366 → 12.060

[cit-pr-commenter-54b7da]

Regression Detector

Regression Detector Results

Metrics dashboard
Target profiles
Run ID: 2403b26f-8c95-408d-a2da-313a1ce50dac

Baseline: 0b457ee
Comparison: 16412a7
Diff

Optimization Goals: ✅ No significant changes detected

Experiments ignored for regressions

Regressions in experiments with settings containing erratic: true are ignored.

perf	experiment	goal	Δ mean %	Δ mean % CI	trials	links
➖	docker_containers_cpu	% cpu utilization	+3.27	[+0.23, +6.31]	1	Logs

Fine details of change detection per experiment

perf	experiment	goal	Δ mean %	Δ mean % CI	trials	links
➖	docker_containers_cpu	% cpu utilization	+3.27	[+0.23, +6.31]	1	Logs
➖	otlp_ingest_logs	memory utilization	+0.73	[+0.62, +0.84]	1	Logs
➖	ddot_metrics_sum_cumulative	memory utilization	+0.36	[+0.22, +0.50]	1	Logs
➖	quality_gate_metrics_logs	memory utilization	+0.26	[+0.02, +0.50]	1	Logs bounds checks dashboard
➖	docker_containers_memory	memory utilization	+0.24	[+0.17, +0.31]	1	Logs
➖	file_to_blackhole_0ms_latency	egress throughput	+0.07	[-0.47, +0.60]	1	Logs
➖	quality_gate_idle	memory utilization	+0.06	[+0.02, +0.11]	1	Logs bounds checks dashboard
➖	quality_gate_idle_all_features	memory utilization	+0.05	[+0.01, +0.09]	1	Logs bounds checks dashboard
➖	tcp_dd_logs_filter_exclude	ingress throughput	+0.01	[-0.10, +0.11]	1	Logs
➖	uds_dogstatsd_to_api_v3	ingress throughput	+0.00	[-0.20, +0.20]	1	Logs
➖	uds_dogstatsd_to_api	ingress throughput	-0.00	[-0.21, +0.20]	1	Logs
➖	file_to_blackhole_500ms_latency	egress throughput	-0.01	[-0.40, +0.39]	1	Logs
➖	file_to_blackhole_1000ms_latency	egress throughput	-0.01	[-0.43, +0.42]	1	Logs
➖	otlp_ingest_metrics	memory utilization	-0.01	[-0.17, +0.14]	1	Logs
➖	file_to_blackhole_100ms_latency	egress throughput	-0.02	[-0.10, +0.05]	1	Logs
➖	ddot_metrics_sum_delta	memory utilization	-0.08	[-0.25, +0.10]	1	Logs
➖	ddot_logs	memory utilization	-0.08	[-0.16, +0.00]	1	Logs
➖	uds_dogstatsd_20mb_12k_contexts_20_senders	memory utilization	-0.15	[-0.21, -0.10]	1	Logs
➖	file_tree	memory utilization	-0.16	[-0.21, -0.10]	1	Logs
➖	ddot_metrics_sum_cumulativetodelta_exporter	memory utilization	-0.38	[-0.60, -0.16]	1	Logs
➖	tcp_syslog_to_blackhole	ingress throughput	-0.44	[-0.58, -0.30]	1	Logs
➖	ddot_metrics	memory utilization	-0.50	[-0.67, -0.33]	1	Logs
➖	quality_gate_logs	% cpu utilization	-0.66	[-2.24, +0.92]	1	Logs bounds checks dashboard

Bounds Checks: ✅ Passed

perf	experiment	bounds_check_name	replicates_passed	observed_value	links
✅	docker_containers_cpu	simple_check_run	10/10	709 ≥ 26
✅	docker_containers_memory	memory_usage	10/10	273.68MiB ≤ 370MiB
✅	docker_containers_memory	simple_check_run	10/10	695 ≥ 26
✅	file_to_blackhole_0ms_latency	memory_usage	10/10	0.19GiB ≤ 1.20GiB
✅	file_to_blackhole_0ms_latency	missed_bytes	10/10	0B = 0B
✅	file_to_blackhole_1000ms_latency	memory_usage	10/10	0.23GiB ≤ 1.20GiB
✅	file_to_blackhole_1000ms_latency	missed_bytes	10/10	0B = 0B
✅	file_to_blackhole_100ms_latency	memory_usage	10/10	0.20GiB ≤ 1.20GiB
✅	file_to_blackhole_100ms_latency	missed_bytes	10/10	0B = 0B
✅	file_to_blackhole_500ms_latency	memory_usage	10/10	0.22GiB ≤ 1.20GiB
✅	file_to_blackhole_500ms_latency	missed_bytes	10/10	0B = 0B
✅	quality_gate_idle	intake_connections	10/10	3 = 3	bounds checks dashboard
✅	quality_gate_idle	memory_usage	10/10	173.66MiB ≤ 175MiB	bounds checks dashboard
✅	quality_gate_idle_all_features	intake_connections	10/10	2 ≤ 3	bounds checks dashboard
✅	quality_gate_idle_all_features	memory_usage	10/10	486.85MiB ≤ 550MiB	bounds checks dashboard
✅	quality_gate_logs	intake_connections	10/10	3 ≤ 6	bounds checks dashboard
✅	quality_gate_logs	memory_usage	10/10	203.47MiB ≤ 220MiB	bounds checks dashboard
✅	quality_gate_logs	missed_bytes	10/10	0B = 0B	bounds checks dashboard
✅	quality_gate_metrics_logs	cpu_usage	10/10	345.91 ≤ 2000	bounds checks dashboard
✅	quality_gate_metrics_logs	intake_connections	10/10	4 ≤ 6	bounds checks dashboard
✅	quality_gate_metrics_logs	memory_usage	10/10	418.82MiB ≤ 475MiB	bounds checks dashboard
✅	quality_gate_metrics_logs	missed_bytes	10/10	0B = 0B	bounds checks dashboard

Explanation

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

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

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".

CI Pass/Fail Decision

✅ Passed. All Quality Gates passed.

• quality_gate_idle, bounds check memory_usage: 10/10 replicas passed. Gate passed.
• quality_gate_idle, bounds check intake_connections: 10/10 replicas passed. Gate passed.
• quality_gate_idle_all_features, bounds check intake_connections: 10/10 replicas passed. Gate passed.
• quality_gate_idle_all_features, bounds check memory_usage: 10/10 replicas passed. Gate passed.
• quality_gate_logs, bounds check memory_usage: 10/10 replicas passed. Gate passed.
• quality_gate_logs, bounds check missed_bytes: 10/10 replicas passed. Gate passed.
• quality_gate_logs, bounds check intake_connections: 10/10 replicas passed. Gate passed.
• quality_gate_metrics_logs, bounds check memory_usage: 10/10 replicas passed. Gate passed.
• quality_gate_metrics_logs, bounds check cpu_usage: 10/10 replicas passed. Gate passed.
• quality_gate_metrics_logs, bounds check intake_connections: 10/10 replicas passed. Gate passed.
• quality_gate_metrics_logs, bounds check missed_bytes: 10/10 replicas passed. Gate passed.