kubernetes_deployment_startup: add vLLM workload support - #6867
Open
ashishsuneja wants to merge 3 commits into
Open
kubernetes_deployment_startup: add vLLM workload support#6867ashishsuneja wants to merge 3 commits into
ashishsuneja wants to merge 3 commits into
Conversation
ashishsuneja
pushed a commit
to ashishsuneja/PerfKitBenchmarker
that referenced
this pull request
Jul 13, 2026
PR 3 of 3 — Baseline/Optimized Scenario + CPU Startup Boost via VPA. Completes the 4-configuration benchmark matrix: Config 1: GKE w/o cpuboost (scenario=baseline, cloud=GCP) Config 2: AWS w/o cpuboost (scenario=baseline, cloud=AWS) Config 3: AKS w/o cpuboost (scenario=baseline, cloud=Azure) Config 4: GKE w/ cpuboost (scenario=optimized, cloud=GCP) Changes: - kubernetes_deployment_startup_boost_factor flag (default 2). - kubernetes_deployment_startup_vpa_yaml flag (slowjvmstartup_vpa.yaml.j2). - CheckPrerequisites(): raises if scenario=optimized on non-GCP or vLLM. - GetConfig(): sets enable_vpa=True for scenario=optimized. - Prepare(): applies VPA manifest with cpuBoostFactor for optimized. - Run(): adds boost_factor to all sample metadata. - slowjvmstartup_vpa.yaml.j2: VPA with startup boost policy. Stacked on PR GoogleCloudPlatform#6865 (metrics) and PR GoogleCloudPlatform#6867 (vLLM workload).
ashishsuneja
pushed a commit
to ashishsuneja/PerfKitBenchmarker
that referenced
this pull request
Jul 13, 2026
PR 3 of 3 — Baseline/Optimized Scenario + CPU Startup Boost via VPA. Completes the 4-configuration benchmark matrix: Config 1: GKE w/o cpuboost (scenario=baseline, cloud=GCP) Config 2: AWS w/o cpuboost (scenario=baseline, cloud=AWS) Config 3: AKS w/o cpuboost (scenario=baseline, cloud=Azure) Config 4: GKE w/ cpuboost (scenario=optimized, cloud=GCP) Changes: - kubernetes_deployment_startup_boost_factor flag (default 2). - kubernetes_deployment_startup_vpa_yaml flag (slowjvmstartup_vpa.yaml.j2). - CheckPrerequisites(): raises if scenario=optimized on non-GCP or vLLM. - GetConfig(): sets enable_vpa=True for scenario=optimized. - Prepare(): applies VPA manifest with cpuBoostFactor for optimized. - Run(): adds boost_factor to all sample metadata. - slowjvmstartup_vpa.yaml.j2: VPA with startup boost policy. Stacked on PR GoogleCloudPlatform#6865 (metrics) and PR GoogleCloudPlatform#6867 (vLLM workload).
ashishsuneja
force-pushed
the
jvm_app_startup_pr2
branch
from
July 13, 2026 09:30
66c55c3 to
ea80c1e
Compare
ashishsuneja
pushed a commit
to ashishsuneja/PerfKitBenchmarker
that referenced
this pull request
Jul 13, 2026
PR 3 of 3 — Baseline/Optimized Scenario + CPU Startup Boost via VPA. Completes the 4-configuration benchmark matrix: Config 1: GKE w/o cpuboost (scenario=baseline, cloud=GCP) Config 2: AWS w/o cpuboost (scenario=baseline, cloud=AWS) Config 3: AKS w/o cpuboost (scenario=baseline, cloud=Azure) Config 4: GKE w/ cpuboost (scenario=optimized, cloud=GCP) Changes: - kubernetes_deployment_startup_boost_factor flag (default 2). - kubernetes_deployment_startup_vpa_yaml flag (slowjvmstartup_vpa.yaml.j2). - CheckPrerequisites(): raises if scenario=optimized on non-GCP or vLLM. - GetConfig(): sets enable_vpa=True for scenario=optimized. - Prepare(): applies VPA manifest with cpuBoostFactor for optimized. - Run(): adds boost_factor to all sample metadata. - slowjvmstartup_vpa.yaml.j2: VPA with startup boost policy. Stacked on PR GoogleCloudPlatform#6865 (metrics) and PR GoogleCloudPlatform#6867 (vLLM workload).
ashishsuneja
pushed a commit
to ashishsuneja/PerfKitBenchmarker
that referenced
this pull request
Jul 13, 2026
PR 3 of 3 — Baseline/Optimized Scenario + CPU Startup Boost via VPA. Completes the 4-configuration benchmark matrix: Config 1: GKE w/o cpuboost (scenario=baseline, cloud=GCP) Config 2: AWS w/o cpuboost (scenario=baseline, cloud=AWS) Config 3: AKS w/o cpuboost (scenario=baseline, cloud=Azure) Config 4: GKE w/ cpuboost (scenario=optimized, cloud=GCP) Changes: - kubernetes_deployment_startup_boost_factor flag (default 2). - kubernetes_deployment_startup_vpa_yaml flag (slowjvmstartup_vpa.yaml.j2). - CheckPrerequisites(): raises if scenario=optimized on non-GCP or vLLM. - GetConfig(): sets enable_vpa=True for scenario=optimized. - Prepare(): applies VPA manifest with cpuBoostFactor for optimized. - Run(): adds boost_factor to all sample metadata. - slowjvmstartup_vpa.yaml.j2: VPA with startup boost policy. Stacked on PR GoogleCloudPlatform#6865 (metrics) and PR GoogleCloudPlatform#6867 (vLLM workload).
…metrics PR 1 of 3 — Metrics & Observability. New metrics: - per_pod_ready_time: one Sample per pod (pod_name in metadata) so callers can compute p50/p90 across replicas. Uses existing kubernetes_conditions.GetStatusConditionsForResourceType() data. - cpu_utilization_peak_millicores / cpu_utilization_mean_millicores / cpu_utilization_reading_count: CPU polled in a background thread via _CpuUtilizationCollector._Observe() following the same pattern as kubernetes_hpa_benchmark.KubernetesMetricsCollector. Existing max_pod_ready_time metric is preserved unchanged. Adds scenario/workload/cloud metadata to all samples for cross-config comparison (PR 3 will set scenario=optimized for GKE CPU Startup Boost). Flag stubs for --kubernetes_deployment_startup_workload and --kubernetes_deployment_startup_scenario registered now so PR 2 and PR 3 do not need flag renames. PR 2 adds vLLM workload support. PR 3 adds VPA CPU Startup Boost scenario (GKE only).
PR 2 of 3 — vLLM Workload Support. Adds a second workload (vLLM CPU inference server) alongside the existing JVM slow-start app: - kubernetes_deployment_startup_vllm_image flag (default public ECR image). - kubernetes_deployment_startup_vllm_yaml flag (vllm.yaml.j2 manifest). - GetConfig(): sets container image from workload flag. - Prepare(): deploys JVM or vLLM manifest based on workload flag. - Run(): uses workload-appropriate deployment name (vllm-startup vs startup). - vllm.yaml.j2: port 8000, /health readiness probe, CPU resource limits. Independent of PR 3 (VPA scenario). No changes to PR 1 metrics.
ashishsuneja
force-pushed
the
jvm_app_startup_pr2
branch
from
July 14, 2026 14:00
6f27c88 to
98e189a
Compare
ashishsuneja
pushed a commit
to ashishsuneja/PerfKitBenchmarker
that referenced
this pull request
Jul 14, 2026
PR 3 of 3 — Baseline/Optimized Scenario + CPU Startup Boost via VPA. Completes the 4-configuration benchmark matrix: Config 1: GKE w/o cpuboost (scenario=baseline, cloud=GCP) Config 2: AWS w/o cpuboost (scenario=baseline, cloud=AWS) Config 3: AKS w/o cpuboost (scenario=baseline, cloud=Azure) Config 4: GKE w/ cpuboost (scenario=optimized, cloud=GCP) Changes: - kubernetes_deployment_startup_boost_factor flag (default 2). - kubernetes_deployment_startup_vpa_yaml flag (slowjvmstartup_vpa.yaml.j2). - CheckPrerequisites(): raises if scenario=optimized on non-GCP or vLLM. - GetConfig(): sets enable_vpa=True for scenario=optimized. - Prepare(): applies VPA manifest with cpuBoostFactor for optimized. - Run(): adds boost_factor to all sample metadata. - slowjvmstartup_vpa.yaml.j2: VPA with startup boost policy. Stacked on PR GoogleCloudPlatform#6865 (metrics) and PR GoogleCloudPlatform#6867 (vLLM workload).
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
PR 2 of 3 — vLLM Workload Support for JVM App Startup Benchmark.
Stacked on PR #6865 (PR 1 — Metrics & Observability). Please review after #6865 merges.
Changes:
--kubernetes_deployment_startup_vllm_imageflag (default: public ECR vLLM CPU image).--kubernetes_deployment_startup_vllm_yamlflag pointing to newvllm.yaml.j2.GetConfig(): routes container image from workload flag.Prepare(): deploys JVM or vLLM manifest based on--workload.Run(): waits on correct deployment name (vllm-startupvsstartup).vllm.yaml.j2: port 8000,/healthreadiness probe, CPU resource limits.Independent of PR 3 — can be reviewed/merged in any order relative to VPA scenario PR.