fix(dashboards): unhide HA instance and fix CPU panels - #24
Merged
Merged
Conversation
… sizing The churn, slow-insert and slow-query panels pinned pod_name=suse-observability-victoria-metrics-0-0, so HA instance 1 stayed invisible even after #223 made it report metrics. Break them down by pod_name across all VictoriaMetrics pods instead. Both CPU Throttling queries returned no series: they divide by container_cpu_elapsed_periods, which is 0 when a container has no CPU limit, so the ratio was NaN. Collapse them into one query grouped on container, which covers all four components rather than splitting across kube_app_component and kube_app_name, and guard the denominator. Throttling cannot exist without a CPU limit, so add a CPU Pressure panel on container_cpu_partial_stall, which reports contention regardless. That is the signal needed when comparing CPU requests. Add request and limit reference lines to both CPU usage panels so headroom is readable.
VioletCranberry
changed the title
rb3ckers
reviewed
Sep 22, 2026
Contributor
|
I would also be interested in these 3 queue metrics:
|
viliakov
reviewed
Sep 22, 2026
Add the queue signals the sizing work needs, with no existing panel: VictoriaMetrics pending rows split by storage and indexdb, vmagent persistent-queue write time alongside pending bytes, collector metrics export queue occupancy, and receiver memory-limiter saturation. Collector queue is plotted as occupancy against queue_capacity rather than raw depth, since capacity differs per exporter and profile. The receiver has no queue, so its memory limiter stands in: reaching the limit is what makes it reject incoming data. Scope every container selector with kube_app_instance, since container names alone are not specific enough. kube_app_name cannot serve here - receiver and vmagent report it as suse-observability while the collector and VictoriaMetrics report their own names. Anchor the pod_name selectors on the suse-observability prefix, which the chart fixes via fullnameOverride rather than deriving from the release name.
rb3ckers
approved these changes
Sep 22, 2026
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
4 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.
Why
This dashboard is the evidence base for the metrics-path sizing work (#178, #206), and three defects made it unusable for that.
pod_name=suse-observability-victoria-metrics-0-0. #223 fixed the chart so instance 1 reports metrics; these panels still hid them.container_cpu_elapsed_periods, which is0for a container with no CPU limit, making the ratio NaN. Verified against a live namespace: zero series. This is the primary signal the CPU-reduction work in #206 depends on.What
Panels break down
by (pod_name)over all VictoriaMetrics pods. Throttling is one query grouped oncontainer, covering all four components where the old pair split acrosskube_app_componentandkube_app_name, with a> 0guard on the denominator.Throttling genuinely cannot exist without a CPU limit, so a new CPU Pressure panel plots
container_cpu_partial_stall, which reports contention either way. Both CPU usage panels gained request and limit lines.Testing
sts stackpack validatepasses, and test-deployed to nightly-champagne as0.0.8-cli-test.10000— the instance upgraded cleanly from 0.0.7.Be aware neither of those gates is strong: I reintroduced a known-bad
unit: secondsinto a throwaway copy and validate still reported success, so it is structural only. Separately, thevalidatejob in this repo's CI is a no-op — the pinned validator image printsusage: WebAppand exits 0. Worth its own ticket; it means schema problems here are caught by nothing but review.Queries were checked against live data: slow inserts 2.0%, churn 559 series/s, requests 0.25/1.25/1.5/5.5 cores, PSI stall highest on vmagent at 23.8ms.
Reviewer notes
unit: s, notseconds- the latter appears nowhere else in any.sty.data-ingestion.styhas the identical blank throttling panel, left untouched.Refs #225