From e93b0a7744886e106c7f483fc12224baa902bbf9 Mon Sep 17 00:00:00 2001
From: Brad Stein <Brad.Stein@gmail.com>
Date: Wed, 28 Jan 2026 20:29:13 -0300
Subject: [PATCH] snapshot: annotate windows and units

---
 ariadne/services/cluster_state.py | 31 ++++++++++++++++++++++---------
 1 file changed, 22 insertions(+), 9 deletions(-)

diff --git a/ariadne/services/cluster_state.py b/ariadne/services/cluster_state.py
index 502537d..8683221 100644
--- a/ariadne/services/cluster_state.py
+++ b/ariadne/services/cluster_state.py
@@ -16,6 +16,8 @@ from ..utils.logging import get_logger
 logger = get_logger(__name__)
 
 _VALUE_PAIR_LEN = 2
+_RATE_WINDOW = "5m"
+_RESTARTS_WINDOW = "1h"
 _WORKLOAD_LABEL_KEYS = (
     "app.kubernetes.io/name",
     "app",
@@ -449,7 +451,7 @@ def _hottest_nodes(errors: list[str]) -> dict[str, Any]:
     hottest: dict[str, Any] = {}
     try:
         hottest["cpu"] = _vm_topk(
-            'label_replace(topk(1, avg by (node) (((1 - avg by (instance) (rate(node_cpu_seconds_total{mode="idle"}[5m]))) * 100) '
+            f'label_replace(topk(1, avg by (node) (((1 - avg by (instance) (rate(node_cpu_seconds_total{{mode="idle"}}[{_RATE_WINDOW}]))) * 100) '
             '* on(instance) group_left(node) label_replace(node_uname_info{nodename!=""}, "node", "$1", "nodename", "(.*)"))), "__name__", "$1", "node", "(.*)")',
             "node",
         )
@@ -459,12 +461,12 @@ def _hottest_nodes(errors: list[str]) -> dict[str, Any]:
             "node",
         )
         hottest["net"] = _vm_topk(
-            'label_replace(topk(1, avg by (node) ((sum by (instance) (rate(node_network_receive_bytes_total{device!~"lo"}[5m]) '
-            '+ rate(node_network_transmit_bytes_total{device!~"lo"}[5m]))) * on(instance) group_left(node) label_replace(node_uname_info{nodename!=""}, "node", "$1", "nodename", "(.*)"))), "__name__", "$1", "node", "(.*)")',
+            f'label_replace(topk(1, avg by (node) ((sum by (instance) (rate(node_network_receive_bytes_total{{device!~"lo"}}[{_RATE_WINDOW}]) '
+            f'+ rate(node_network_transmit_bytes_total{{device!~"lo"}}[{_RATE_WINDOW}]))) * on(instance) group_left(node) label_replace(node_uname_info{nodename!=""}, "node", "$1", "nodename", "(.*)"))), "__name__", "$1", "node", "(.*)")',
             "node",
         )
         hottest["io"] = _vm_topk(
-            'label_replace(topk(1, avg by (node) ((sum by (instance) (rate(node_disk_read_bytes_total[5m]) + rate(node_disk_written_bytes_total[5m]))) '
+            f'label_replace(topk(1, avg by (node) ((sum by (instance) (rate(node_disk_read_bytes_total[{_RATE_WINDOW}]) + rate(node_disk_written_bytes_total[{_RATE_WINDOW}]))) '
             '* on(instance) group_left(node) label_replace(node_uname_info{nodename!=""}, "node", "$1", "nodename", "(.*)"))), "__name__", "$1", "node", "(.*)")',
             "node",
         )
@@ -477,7 +479,7 @@ def _node_usage(errors: list[str]) -> dict[str, Any]:
     usage: dict[str, Any] = {}
     try:
         usage["cpu"] = _vm_node_metric(
-            'avg by (node) (((1 - avg by (instance) (rate(node_cpu_seconds_total{mode="idle"}[5m]))) * 100) '
+            f'avg by (node) (((1 - avg by (instance) (rate(node_cpu_seconds_total{{mode="idle"}}[{_RATE_WINDOW}]))) * 100) '
             '* on(instance) group_left(node) label_replace(node_uname_info{nodename!=""}, "node", "$1", "nodename", "(.*)"))',
             "node",
         )
@@ -487,13 +489,13 @@ def _node_usage(errors: list[str]) -> dict[str, Any]:
             "node",
         )
         usage["net"] = _vm_node_metric(
-            'avg by (node) ((sum by (instance) (rate(node_network_receive_bytes_total{device!~"lo"}[5m]) '
-            '+ rate(node_network_transmit_bytes_total{device!~"lo"}[5m]))) * on(instance) group_left(node) '
+            f'avg by (node) ((sum by (instance) (rate(node_network_receive_bytes_total{{device!~"lo"}}[{_RATE_WINDOW}]) '
+            f'+ rate(node_network_transmit_bytes_total{{device!~"lo"}}[{_RATE_WINDOW}]))) * on(instance) group_left(node) '
             'label_replace(node_uname_info{nodename!=""}, "node", "$1", "nodename", "(.*)"))',
             "node",
         )
         usage["io"] = _vm_node_metric(
-            'avg by (node) ((sum by (instance) (rate(node_disk_read_bytes_total[5m]) + rate(node_disk_written_bytes_total[5m]))) '
+            f'avg by (node) ((sum by (instance) (rate(node_disk_read_bytes_total[{_RATE_WINDOW}]) + rate(node_disk_written_bytes_total[{_RATE_WINDOW}]))) '
             '* on(instance) group_left(node) label_replace(node_uname_info{nodename!=""}, "node", "$1", "nodename", "(.*)"))',
             "node",
         )
@@ -514,13 +516,24 @@ def _summarize_metrics(errors: list[str]) -> dict[str, Any]:
         metrics["pods_failed"] = _vm_scalar("sum(kube_pod_status_phase{phase=\"Failed\"})")
         metrics["pods_succeeded"] = _vm_scalar("sum(kube_pod_status_phase{phase=\"Succeeded\"})")
         metrics["top_restarts_1h"] = _vm_vector(
-            "topk(5, sum by (namespace,pod) (increase(kube_pod_container_status_restarts_total[1h])))"
+            f"topk(5, sum by (namespace,pod) (increase(kube_pod_container_status_restarts_total[{_RESTARTS_WINDOW}])))"
         )
     except Exception as exc:
         errors.append(f"vm: {exc}")
     metrics["postgres_connections"] = _postgres_connections(errors)
     metrics["hottest_nodes"] = _hottest_nodes(errors)
     metrics["node_usage"] = _node_usage(errors)
+    metrics["units"] = {
+        "cpu": "percent",
+        "ram": "percent",
+        "net": "bytes_per_sec",
+        "io": "bytes_per_sec",
+        "restarts": "count",
+    }
+    metrics["windows"] = {
+        "rates": _RATE_WINDOW,
+        "restarts": _RESTARTS_WINDOW,
+    }
     return metrics