Skip to content

fix(metrics): resolve high-cardinality leak in memory_allocated_bytes #2142

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
hami-robot merged 3 commits into from
Jul 29, 2026
Merged

fix(metrics): resolve high-cardinality leak in memory_allocated_bytes #2142

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
4d93beb
fix(metrics): resolve high-cardinality leak in memory_allocated_bytes
Aryanbhargava18 Jul 27, 2026
4760bda
fix: remove dynamic state variables from metric labels
Aryanbhargava18 Jul 27, 2026
f3747aa
fix: restore static labels to avoid dashboard breakage and deduplicat…
Aryanbhargava18 Jul 28, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 6 additions & 6 deletions cmd/scheduler/metrics.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -80,7 +80,7 @@ func (cc ClusterManagerCollector) Collect(ch chan<- prometheus.Metric) {
nodeGPUOverview := prometheus.NewDesc(
"hami_node_gpu_overview",
"GPU overview on a certain node",
[]string{"node", "device_uuid", "device_index", "device_cores", "shared_containers", "device_memory_limit", "device_type"}, nil,
[]string{"node", "device_uuid", "device_index", "device_cores", "device_memory_limit", "device_type"}, nil,
)
nodeGPUMemoryPercentage := prometheus.NewDesc(
"hami_node_gpu_memory_allocated_ratio",
Expand Down Expand Up @@ -136,7 +136,7 @@ func (cc ClusterManagerCollector) Collect(ch chan<- prometheus.Metric) {
legacyOverview = prometheus.NewDesc(
"nodeGPUOverview",
"GPU overview on a certain node",
[]string{"nodeid", "deviceuuid", "deviceidx", "devicecores", "sharedcontainers", "devicememorylimit", "devicetype"}, nil,
[]string{"nodeid", "deviceuuid", "deviceidx", "devicecores", "devicememorylimit", "devicetype"}, nil,
)
legacyMemoryPercentage = prometheus.NewDesc(
"nodeGPUMemoryPercentage",
Expand Down Expand Up @@ -208,7 +208,7 @@ func (cc ClusterManagerCollector) Collect(ch chan<- prometheus.Metric) {
nodevGPUMemoryAllocatedDesc,
prometheus.GaugeValue,
float64(devs.Device.Usedmem)*float64(1024)*float64(1024),
nodeID, devs.Device.ID, fmt.Sprint(devs.Device.Index), fmt.Sprint(devs.Device.Usedcores), devs.Device.Type,
nodeID, devs.Device.ID, fmt.Sprint(devs.Device.Index), fmt.Sprint(devs.Device.Totalcore), devs.Device.Type,
)
ch <- prometheus.MustNewConstMetric(
nodevGPUSharedNumDesc,
Expand All @@ -226,7 +226,7 @@ func (cc ClusterManagerCollector) Collect(ch chan<- prometheus.Metric) {
nodeGPUOverview,
prometheus.GaugeValue,
float64(devs.Device.Usedmem)*float64(1024)*float64(1024),
nodeID, devs.Device.ID, fmt.Sprint(devs.Device.Index), fmt.Sprint(devs.Device.Usedcores), fmt.Sprint(devs.Device.Used), fmt.Sprint(devs.Device.Totalmem), devs.Device.Type,
nodeID, devs.Device.ID, fmt.Sprint(devs.Device.Index), fmt.Sprint(devs.Device.Totalcore), fmt.Sprint(devs.Device.Totalmem), devs.Device.Type,
)
ch <- prometheus.MustNewConstMetric(
nodeGPUMemoryPercentage,
Expand All @@ -252,7 +252,7 @@ func (cc ClusterManagerCollector) Collect(ch chan<- prometheus.Metric) {
legacyMemoryAllocatedDesc,
prometheus.GaugeValue,
float64(devs.Device.Usedmem)*float64(1024)*float64(1024),
nodeID, devs.Device.ID, fmt.Sprint(devs.Device.Index), fmt.Sprint(devs.Device.Usedcores), devs.Device.Type,
nodeID, devs.Device.ID, fmt.Sprint(devs.Device.Index), fmt.Sprint(devs.Device.Totalcore), devs.Device.Type,
)
ch <- prometheus.MustNewConstMetric(
legacySharedNumDesc,
Expand All @@ -270,7 +270,7 @@ func (cc ClusterManagerCollector) Collect(ch chan<- prometheus.Metric) {
legacyOverview,
prometheus.GaugeValue,
float64(devs.Device.Usedmem)*float64(1024)*float64(1024),
nodeID, devs.Device.ID, fmt.Sprint(devs.Device.Index), fmt.Sprint(devs.Device.Usedcores), fmt.Sprint(devs.Device.Used), fmt.Sprint(devs.Device.Totalmem), devs.Device.Type,
nodeID, devs.Device.ID, fmt.Sprint(devs.Device.Index), fmt.Sprint(devs.Device.Totalcore), fmt.Sprint(devs.Device.Totalmem), devs.Device.Type,
)
ch <- prometheus.MustNewConstMetric(
legacyMemoryPercentage,
Expand Down
4 changes: 3 additions & 1 deletion skill/hami-vgpu-metrics-summary/SKILL.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -245,10 +245,12 @@ These are emitted in **bytes**, even though internal scheduler state stores memo
These are emitted directly from quota manager `q.Used` and should be interpreted as **MiB-like GPU memory units**, not bytes.

**Overview labels**
- `hami_node_gpu_overview.device_cores`
- `hami_node_gpu_overview.device_memory_limit`
- legacy `nodeGPUOverview.devicecores`
- legacy `nodeGPUOverview.devicememorylimit`

These label values are also in **MiB-like units**.
These label values are static. Memory limits are in **MiB-like units**, and cores are integers.

Use these conversions when needed:

Expand Down
Loading