[clusteragent/autoscaling] Generate local horizontal recommendations #32882
Conversation
jennchenn
added
team/containers
changelog/no-changelog
qa/done
Uncompressed package size comparisonComparison with ancestor Diff per package
Decision |
Test changes on VMUse this command from test-infra-definitions to manually test this PR changes on a VM: inv aws.create-vm --pipeline-id=53705519 --os-family=ubuntuNote: This applies to commit 7804889 |
Regression DetectorRegression Detector ResultsMetrics dashboard Baseline: edd9ae2 Optimization Goals: ✅ No significant changes detected
|
| perf | experiment | goal | Δ mean % | Δ mean % CI | trials | links |
|---|---|---|---|---|---|---|
| ➖ | quality_gate_logs | % cpu utilization | +3.15 | [+0.06, +6.23] | 1 | Logs |
| ➖ | uds_dogstatsd_to_api_cpu | % cpu utilization | +1.29 | [+0.37, +2.21] | 1 | Logs |
| ➖ | quality_gate_idle_all_features | memory utilization | +0.27 | [+0.22, +0.32] | 1 | Logs bounds checks dashboard |
| ➖ | file_to_blackhole_1000ms_latency_linear_load | egress throughput | +0.13 | [-0.34, +0.59] | 1 | Logs |
| ➖ | file_to_blackhole_0ms_latency_http1 | egress throughput | +0.07 | [-0.84, +0.98] | 1 | Logs |
| ➖ | quality_gate_idle | memory utilization | +0.06 | [+0.02, +0.09] | 1 | Logs bounds checks dashboard |
| ➖ | file_to_blackhole_500ms_latency | egress throughput | +0.05 | [-0.73, +0.83] | 1 | Logs |
| ➖ | file_to_blackhole_100ms_latency | egress throughput | +0.02 | [-0.69, +0.74] | 1 | Logs |
| ➖ | file_to_blackhole_300ms_latency | egress throughput | -0.01 | [-0.64, +0.63] | 1 | Logs |
| ➖ | tcp_dd_logs_filter_exclude | ingress throughput | -0.01 | [-0.03, +0.01] | 1 | Logs |
| ➖ | uds_dogstatsd_to_api | ingress throughput | -0.01 | [-0.28, +0.26] | 1 | Logs |
| ➖ | file_to_blackhole_0ms_latency | egress throughput | -0.01 | [-0.93, +0.90] | 1 | Logs |
| ➖ | file_to_blackhole_0ms_latency_http2 | egress throughput | -0.01 | [-0.89, +0.86] | 1 | Logs |
| ➖ | file_tree | memory utilization | -0.05 | [-0.10, -0.00] | 1 | Logs |
| ➖ | file_to_blackhole_1000ms_latency | egress throughput | -0.85 | [-1.65, -0.05] | 1 | Logs |
| ➖ | tcp_syslog_to_blackhole | ingress throughput | -1.84 | [-2.02, -1.65] | 1 | Logs |
Bounds Checks: ✅ Passed
| perf | experiment | bounds_check_name | replicates_passed | links |
|---|---|---|---|---|
| ✅ | file_to_blackhole_0ms_latency | lost_bytes | 10/10 | |
| ✅ | file_to_blackhole_0ms_latency | memory_usage | 10/10 | |
| ✅ | file_to_blackhole_0ms_latency_http1 | lost_bytes | 10/10 | |
| ✅ | file_to_blackhole_0ms_latency_http1 | memory_usage | 10/10 | |
| ✅ | file_to_blackhole_0ms_latency_http2 | lost_bytes | 10/10 | |
| ✅ | file_to_blackhole_0ms_latency_http2 | memory_usage | 10/10 | |
| ✅ | file_to_blackhole_1000ms_latency | memory_usage | 10/10 | |
| ✅ | file_to_blackhole_1000ms_latency_linear_load | memory_usage | 10/10 | |
| ✅ | file_to_blackhole_100ms_latency | lost_bytes | 10/10 | |
| ✅ | file_to_blackhole_100ms_latency | memory_usage | 10/10 | |
| ✅ | file_to_blackhole_300ms_latency | lost_bytes | 10/10 | |
| ✅ | file_to_blackhole_300ms_latency | memory_usage | 10/10 | |
| ✅ | file_to_blackhole_500ms_latency | lost_bytes | 10/10 | |
| ✅ | file_to_blackhole_500ms_latency | memory_usage | 10/10 | |
| ✅ | quality_gate_idle | intake_connections | 10/10 | bounds checks dashboard |
| ✅ | quality_gate_idle | memory_usage | 10/10 | bounds checks dashboard |
| ✅ | quality_gate_idle_all_features | intake_connections | 10/10 | bounds checks dashboard |
| ✅ | quality_gate_idle_all_features | memory_usage | 10/10 | bounds checks dashboard |
| ✅ | quality_gate_logs | intake_connections | 10/10 | |
| ✅ | quality_gate_logs | lost_bytes | 10/10 | |
| ✅ | quality_gate_logs | memory_usage | 10/10 |
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:
-
Its estimated |Δ mean %| ≥ 5.00%, indicating the change is big enough to merit a closer look.
-
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.
-
Its configuration does not mark it "erratic".
CI Pass/Fail Decision
✅ Passed. All Quality Gates passed.
- quality_gate_idle, bounds check intake_connections: 10/10 replicas passed. Gate passed.
- quality_gate_idle, bounds check memory_usage: 10/10 replicas passed. Gate passed.
- quality_gate_idle_all_features, bounds check memory_usage: 10/10 replicas passed. Gate passed.
- quality_gate_idle_all_features, bounds check intake_connections: 10/10 replicas passed. Gate passed.
- quality_gate_logs, bounds check intake_connections: 10/10 replicas passed. Gate passed.
- quality_gate_logs, bounds check memory_usage: 10/10 replicas passed. Gate passed.
- quality_gate_logs, bounds check lost_bytes: 10/10 replicas passed. Gate passed.
Go Package Import DifferencesBaseline: edd9ae2
|
| @@ -246,6 +246,7 @@ func parsePodContainers( | |||
| if containerSpec != nil { | |||
| env = extractEnvFromSpec(containerSpec.Env) | |||
| resources = extractResources(containerSpec) | |||
| podContainer.Resources = resources | |||
There was a problem hiding this comment.
Not sure this is required in case of Kubelet collector
| c := workloadmeta.OrchestratorContainer{ | ||
| Name: container.Name, | ||
| } | ||
| if cpuReq, found := container.Resources.Requests[corev1.ResourceCPU]; found { |
There was a problem hiding this comment.
Code is duplicated with already existing Kubelet container parsing. Perhaps we can do something to share the code.
| containersList = append(containersList, c) | ||
| } | ||
|
|
||
| // If this list is empty, we want to return nil instead of an empty list |
There was a problem hiding this comment.
Is there a reason for that? From the code I don't even think it's possible except if the POD has 0 container (which is not possible).
There was a problem hiding this comment.
Added this because of a test case failing, but you're right the test case defined a pod with 0 containers - removed!
| const ( | ||
| staleDataThresholdSeconds = 180 // 3 minutes | ||
| containerCPUUsageMetricName = "container.cpu.usage" | ||
| containerMemoryUsageMetricName = "container.memory.usage" |
There was a problem hiding this comment.
Note for later: we may want to tweak that. For horizontal scaling, perhaps working_set could be make sense if we want to be closer to OOM limits.
| } | ||
| } | ||
|
|
||
| func newResourceRecommenderSettings(target datadoghq.DatadogPodAutoscalerTarget) (*resourceRecommenderSettings, error) { |
There was a problem hiding this comment.
I think this part will change as we may want a specific fallback target. Another point is that it's also validating targets.
It would perhaps not expect this to be done there, as validation could be done way earlier, but if it's done there, then I would expect more relevant error messages for users.
| ) | ||
|
|
||
| type resourceRecommenderSettings struct { | ||
| MetricName string |
There was a problem hiding this comment.
Same here, do we need these fields to be public?
| } | ||
|
|
||
| // ReinitLoadstore reinitializes the loadstore for the local recommender | ||
| func (l *Recommender) ReinitLoadstore(ctx context.Context) error { |
There was a problem hiding this comment.
I'm not sure I understand why we need this? Is there any issue with initializing this in the new function?
There was a problem hiding this comment.
Changed this bit of logic; original motivation was that we need to check the store is not nil before calling store.GetMetricsRaw. This check is now done in the process loop directly (fixes some of the other comments above as well)
| return nil, fmt.Errorf("Invalid target type: %s", target.Type) | ||
| } | ||
|
|
||
| func getOptionsFromPodResource(target *datadoghq.DatadogPodAutoscalerResourceTarget) (*resourceRecommenderSettings, error) { |
There was a problem hiding this comment.
I think we need to split the code for the metrics/calculation and the settings/setup part in different files.
| ID string | ||
| Name string | ||
| Image ContainerImage | ||
| Resources ContainerResources |
There was a problem hiding this comment.
This is all because the metric we send is absolute and not relative right?
There was a problem hiding this comment.
Discussed briefly online - yes (currently using container.cpu.usage from the agent so to calculate utilization we require container requests)
|
|
||
| //go:build kubeapiserver | ||
|
|
||
| package shared |
There was a problem hiding this comment.
nit: usually more named common than shared
…luster agent image
…-local-recommender
…-local-recommender
What does this PR do?
Initial implementation of local recommendation engine. This PR:
localand pass pod information directly into the recommendation logic if that makes more senseautoscaling.workload.local.horizontal_scaling_recommended_replicastelemetry metric to track local recommendationsMotivation
Support local fallback.
Describe how you validated your changes
Possible Drawbacks / Trade-offs
Additional Notes
This PR does not include: