bstein/ariadne
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

snapshot: add pod top metrics with node

Browse Source
This commit is contained in:
Brad Stein 2026-01-29 10:14:34 -03:00
parent d4a632d2e4
commit 29575baeb0
1 changed files with 12 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
ariadne/services/cluster_state.py
View File

@ -1373,11 +1373,21 @@ def _summarize_metrics(errors: list[str]) -> dict[str, Any]:
f'topk(5, sum by (namespace,pod) (rate(container_cpu_usage_seconds_total{{namespace!=""}}[{_RATE_WINDOW}])))' f'topk(5, sum by (namespace,pod) (rate(container_cpu_usage_seconds_total{{namespace!=""}}[{_RATE_WINDOW}])))'
) )
) )
metrics["pod_cpu_top_node"] = _filter_namespace_vector(
_vm_vector(
f'topk(5, sum by (node,namespace,pod) (rate(container_cpu_usage_seconds_total{{namespace!=""}}[{_RATE_WINDOW}]) * on (namespace,pod) group_left(node) kube_pod_info))'
)
)
metrics["pod_mem_top"] = _filter_namespace_vector( metrics["pod_mem_top"] = _filter_namespace_vector(
_vm_vector( _vm_vector(
"topk(5, sum by (namespace,pod) (container_memory_working_set_bytes{namespace!=\"\"}))" "topk(5, sum by (namespace,pod) (container_memory_working_set_bytes{namespace!=\"\"}))"
) )
) )
metrics["pod_mem_top_node"] = _filter_namespace_vector(
_vm_vector(
"topk(5, sum by (node,namespace,pod) (container_memory_working_set_bytes{namespace!=\"\"} * on (namespace,pod) group_left(node) kube_pod_info))"
)
)
metrics["job_failures_24h"] = _vm_vector( metrics["job_failures_24h"] = _vm_vector(
"topk(5, sum by (namespace,job_name) (increase(kube_job_status_failed[24h])))" "topk(5, sum by (namespace,job_name) (increase(kube_job_status_failed[24h])))"
) )
@ -1436,6 +1446,8 @@ def _summarize_metrics(errors: list[str]) -> dict[str, Any]:
"restarts": "count", "restarts": "count",
"pod_cpu": "cores", "pod_cpu": "cores",
"pod_mem": "bytes", "pod_mem": "bytes",
"pod_cpu_top_node": "cores",
"pod_mem_top_node": "bytes",
"job_failures_24h": "count", "job_failures_24h": "count",
"namespace_cpu": "cores", "namespace_cpu": "cores",
"namespace_mem": "bytes", "namespace_mem": "bytes",