Add system-node-critical priorityClass to health-monitoring-agent DaemonSets - #451
Open
huaziyao wants to merge 1 commit into
Open
Add system-node-critical priorityClass to health-monitoring-agent DaemonSets#451huaziyao wants to merge 1 commit into
huaziyao wants to merge 1 commit into
Conversation
…monSets Under node DiskPressure, HMA was evicted moments before a GPU failure went undetected and unremediated (P438078245). Setting priorityClassName to the built-in system-node-critical makes HMA the last pod evicted (after training and other monitoring pods), preserving health monitoring during node pressure. Mirrors what EKSNodeMonitoringAgent already does. Applied to both the health-monitoring-agent (NVIDIA) and health-monitoring-agent-non-nvidia (Trainium/Inferentia) DaemonSets. Configurable via .Values.priorityClassName, defaulting to system-node-critical.
huaziyao
requested a deployment
to
manual-approval
August 21, 2026 22:43 — with 
GitHub Actions
Waiting
GitHub Actions
Waiting
Contributor
|
LGTM |
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
2 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What's changing and why?
Adds a
priorityClassName(defaulting to the Kubernetes built-insystem-node-critical) to both health-monitoring-agent DaemonSets:health-monitoring-agent(NVIDIA / GPU)health-monitoring-agent-non-nvidia(Trainium / Inferentia)The value is configurable via
.Values.priorityClassNameand falls back tosystem-node-criticalwhen unset.Why: Under node
DiskPressure, HMA was evicted moments before a GPU failure occurred — so the failure went undetected and unremediated.system-node-critical(priority value2000001000) makes HMA among the last pods evicted under node pressure, after training and other monitoring pods, preserving health monitoring exactly when it matters most. This mirrors what EKSNodeMonitoringAgent already does. Becausesystem-node-criticalis a built-in PriorityClass, no additional PriorityClass object needs to be created.Before / After UX
0. Under node pressure (e.g.DiskPressure), they can be evicted early — potentially before or during a hardware failure — leaving the node unmonitored.2000001000(system-node-critical) by default, so the kubelet evicts them last, keeping health monitoring alive during node pressure. Operators can override or disable via.Values.priorityClassName.How was this change tested?
Validated through render, lint, dry-run, and a live apply/observe/revert on a real cluster:
helm template— rendered both DaemonSets and confirmedpriorityClassName: system-node-criticalappears in each pod spec; verified--set priorityClassName=<x>propagates correctly, with fallback tosystem-node-criticalwhen unset.helm lint— passes clean.kubectl apply --server-side --dry-runaccepted with no schema/admission errors.priority: 2000001000(up from0); reverting returned them topriority: 0with no priorityClassName, again via a clean rollout. No DaemonSet disruption during either transition.Are unit tests added?
No — this is a Helm chart template/values change with no application code. Coverage is provided by the
helm template/helm lintrender checks above.Are integration tests added?
No new automated integration tests. Verified manually via the live apply/observe/revert described above.
Reviewer Guidelines
One of the following must be true: