[clusteragent/autoscaling] Defer workloadmeta pod collection until first DPA#51084
[clusteragent/autoscaling] Defer workloadmeta pod collection until first DPA#51084davidor wants to merge 4 commits into
Conversation
dd-octo-sts
Bot
added
internal
|
Go Package Import DifferencesBaseline: faf049a
|
davidor
added
qa/done
Files inventory check summaryFile checks results against ancestor faf049aa: Results for datadog-agent_7.81.0~devel.git.367.a4b207f.pipeline.116212102-1_amd64.deb:No change detected |
Static quality checks❌ Please find below the results from static quality gates Error
Gate failure full details
Static quality gates prevent the PR to merge! Successful checksInfo
30 successful checks with minimal change (< 2 KiB)
|
Regression DetectorRegression Detector ResultsMetrics dashboard Baseline: faf049a Optimization Goals: ✅ No significant changes detected
|
| perf | experiment | goal | Δ mean % | Δ mean % CI | trials | links |
|---|---|---|---|---|---|---|
| ➖ | docker_containers_cpu | % cpu utilization | -0.54 | [-3.46, +2.39] | 1 | Logs |
Fine details of change detection per experiment
| perf | experiment | goal | Δ mean % | Δ mean % CI | trials | links |
|---|---|---|---|---|---|---|
| ➖ | otlp_ingest_logs | memory utilization | +0.56 | [+0.45, +0.68] | 1 | Logs |
| ➖ | quality_gate_idle | memory utilization | +0.35 | [+0.30, +0.40] | 1 | Logs bounds checks dashboard |
| ➖ | uds_dogstatsd_20mb_12k_contexts_20_senders | memory utilization | +0.34 | [+0.29, +0.39] | 1 | Logs |
| ➖ | quality_gate_idle_all_features | memory utilization | +0.24 | [+0.20, +0.27] | 1 | Logs bounds checks dashboard |
| ➖ | ddot_metrics_sum_cumulativetodelta_exporter | memory utilization | +0.21 | [-0.03, +0.44] | 1 | Logs |
| ➖ | ddot_metrics | memory utilization | +0.05 | [-0.15, +0.25] | 1 | Logs |
| ➖ | file_to_blackhole_0ms_latency | egress throughput | +0.04 | [-0.46, +0.53] | 1 | Logs |
| ➖ | tcp_dd_logs_filter_exclude | ingress throughput | +0.01 | [-0.08, +0.11] | 1 | Logs |
| ➖ | ddot_metrics_sum_delta | memory utilization | +0.01 | [-0.18, +0.19] | 1 | Logs |
| ➖ | ddot_logs | memory utilization | +0.00 | [-0.06, +0.07] | 1 | Logs |
| ➖ | uds_dogstatsd_to_api_v3 | ingress throughput | -0.00 | [-0.19, +0.19] | 1 | Logs |
| ➖ | file_to_blackhole_100ms_latency | egress throughput | -0.01 | [-0.15, +0.13] | 1 | Logs |
| ➖ | file_to_blackhole_500ms_latency | egress throughput | -0.02 | [-0.43, +0.39] | 1 | Logs |
| ➖ | uds_dogstatsd_to_api | ingress throughput | -0.03 | [-0.24, +0.17] | 1 | Logs |
| ➖ | file_to_blackhole_1000ms_latency | egress throughput | -0.07 | [-0.52, +0.38] | 1 | Logs |
| ➖ | file_tree | memory utilization | -0.30 | [-0.35, -0.25] | 1 | Logs |
| ➖ | docker_containers_memory | memory utilization | -0.31 | [-0.41, -0.21] | 1 | Logs |
| ➖ | ddot_metrics_sum_cumulative | memory utilization | -0.47 | [-0.63, -0.31] | 1 | Logs |
| ➖ | docker_containers_cpu | % cpu utilization | -0.54 | [-3.46, +2.39] | 1 | Logs |
| ➖ | otlp_ingest_metrics | memory utilization | -0.56 | [-0.72, -0.41] | 1 | Logs |
| ➖ | tcp_syslog_to_blackhole | ingress throughput | -0.61 | [-0.81, -0.41] | 1 | Logs |
| ➖ | quality_gate_metrics_logs | memory utilization | -0.76 | [-1.01, -0.52] | 1 | Logs bounds checks dashboard |
| ➖ | quality_gate_logs | % cpu utilization | -1.81 | [-2.85, -0.77] | 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 | 726 ≥ 26 | |
| ✅ | docker_containers_memory | memory_usage | 10/10 | 246.09MiB ≤ 370MiB | |
| ✅ | docker_containers_memory | simple_check_run | 10/10 | 689 ≥ 26 | |
| ✅ | file_to_blackhole_0ms_latency | memory_usage | 10/10 | 0.16GiB ≤ 1.20GiB | |
| ✅ | file_to_blackhole_0ms_latency | missed_bytes | 10/10 | 0B = 0B | |
| ✅ | file_to_blackhole_1000ms_latency | memory_usage | 10/10 | 0.20GiB ≤ 1.20GiB | |
| ✅ | file_to_blackhole_1000ms_latency | missed_bytes | 10/10 | 0B = 0B | |
| ✅ | file_to_blackhole_100ms_latency | memory_usage | 10/10 | 0.17GiB ≤ 1.20GiB | |
| ✅ | file_to_blackhole_100ms_latency | missed_bytes | 10/10 | 0B = 0B | |
| ✅ | file_to_blackhole_500ms_latency | memory_usage | 10/10 | 0.18GiB ≤ 1.20GiB | |
| ✅ | file_to_blackhole_500ms_latency | missed_bytes | 10/10 | 0B = 0B | |
| ✅ | quality_gate_idle | intake_connections | 10/10 | 3 ≤ 4 | bounds checks dashboard |
| ✅ | quality_gate_idle | memory_usage | 10/10 | 146.36MiB ≤ 147MiB | bounds checks dashboard |
| ✅ | quality_gate_idle | total_bytes_received | 10/10 | 745.64KiB ≤ 819.20KiB | bounds checks dashboard |
| ✅ | quality_gate_idle_all_features | intake_connections | 10/10 | 3 ≤ 4 | bounds checks dashboard |
| ✅ | quality_gate_idle_all_features | memory_usage | 10/10 | 479.33MiB ≤ 495MiB | bounds checks dashboard |
| ✅ | quality_gate_idle_all_features | total_bytes_received | 10/10 | 1.13MiB ≤ 1.25MiB | bounds checks dashboard |
| ✅ | quality_gate_logs | intake_connections | 10/10 | 4 ≤ 6 | bounds checks dashboard |
| ✅ | quality_gate_logs | memory_usage | 10/10 | 176.04MiB ≤ 195MiB | bounds checks dashboard |
| ✅ | quality_gate_logs | missed_bytes | 10/10 | 0B = 0B | bounds checks dashboard |
| ✅ | quality_gate_logs | total_bytes_received | 10/10 | 263.97MiB ≤ 292MiB | bounds checks dashboard |
| ✅ | quality_gate_metrics_logs | cpu_usage | 10/10 | 346.88 ≤ 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 | 383.27MiB ≤ 430MiB | bounds checks dashboard |
| ✅ | quality_gate_metrics_logs | missed_bytes | 10/10 | 0B = 0B | bounds checks dashboard |
| ✅ | quality_gate_metrics_logs | total_bytes_received | 10/10 | 0.94GiB ≤ 1.04GiB | 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:
-
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_logs, bounds check missed_bytes: 10/10 replicas passed. Gate passed.
- quality_gate_logs, bounds check total_bytes_received: 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_idle_all_features, bounds check intake_connections: 10/10 replicas passed. Gate passed.
- quality_gate_idle_all_features, bounds check total_bytes_received: 10/10 replicas passed. Gate passed.
- quality_gate_idle_all_features, 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, bounds check memory_usage: 10/10 replicas passed. Gate passed.
- quality_gate_idle, bounds check total_bytes_received: 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 missed_bytes: 10/10 replicas passed. Gate passed.
- quality_gate_metrics_logs, bounds check total_bytes_received: 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.
|
@codex review |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: d014c9817c
ℹ️ About Codex in GitHub
Codex has been enabled to automatically 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 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| reflector, store := storeBuilder(ctx, wlmetaStore, c.config, client) | ||
| if shouldHavePodStore(c.config) { | ||
| autoscalingEnabled := c.config.GetBool("autoscaling.workload.enabled") | ||
| lazyStart := !podsRequiredAtStartup(c.config) && autoscalingEnabled |
There was a problem hiding this comment.
When autoscaling.workload.enabled is set together with cluster_agent.kube_metadata_collection.resources containing pods, this branch treats workload autoscaling as the only pod-store reason and defers the pod reflector until a DPA appears. However resourcesWithExplicitMetadataCollectionEnabled skips pods because it expects the dedicated pod store to provide them, so clusters with pod metadata explicitly requested but no current DPA stop collecting that metadata until autoscaling is first used. Please make explicit pod metadata collection a startup-time pod requirement before enabling the lazy path.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
The explicit collection part is an existing bug. It should be addressed separately.
davidor
force-pushed
the
davidor/contp-1632-autoscaling-lazy-pod-collection
branch
from
d014c98 to
a4b207f
Compare
|
Rebased on top of main to fix CI |
|
Implementation LGTM, but we need to gate the controller scaling logic as now the PodWatcher may be delayed more than before (was already not great before). Just before this code https://github.com/DataDog/datadog-agent/blob/main/pkg/clusteragent/autoscaling/workload/controller.go#L439-L440 we should check for PodWatcher synced. It should not be a blocking wait, more like check if ready, if not requeue 30s. We should also log when we start waiting and when wait is over. |
5 participants
What does this PR do?
The goal of this PR is to reduce the memory cost of enabling workload autoscaling when it's not in use.
Right now, when autoscaling is enabled, the kubeapiserver workloadmeta collector starts a pod reflector. In large clusters, this can use a lot of memory. This happens even if no DPA is deployed.
We want to avoid this memory usage when no DPAs are deployed.
This will let us enable workload autoscaling by default at a much lower cost when it's not in use. Users who want it will be able to create DPAs directly, without having to enable the option in the Cluster Agent.
This PR does not flip the
autoscaling.workload.enableddefault to true. That will be done in a separate PR so it can be reverted independently if needed.Note that this PR only gates the pod collection part instead of the whole autoscaling stack. That approach was tried in another PR (#50305) but it proved to be tricky. The problem is that parts of autoscaling need to be running to create DPAs in some cases (for example, when they come from remote config, or from profile-labelled workloads). So this PR reduces the cost of enabling autoscaling when there are no DPAs, but doesn't completely remove it. Some parts still run, like the metadata-only informers for deployments and statefulsets.
Describe how you validated your changes
Unit tests plus tests on a local kind cluster.
For the kind tests, I used kwok to simulate a large number of pods so the memory impact of the pod reflector would be measurable. I ran 5 scenarios: