--- editable: false alerts: height: 100 panels: - type: alert title: "Root disk used %" max: 100 expr: '(node_filesystem_size{mountpoint="/"} - node_filesystem_avail{mountpoint="/"})/node_filesystem_size{mountpoint="/"}*100 > 90)' - type: alert title: "Availability" max: 2 expr: 'up < 0.5' - type: alert title: "CPU" max: 100 expr: '100 - (avg by (instance)(rate(node_cpu{mode="idle"}[5m]))*100) > 85' - type: alert title: "Used Memory, %" max: 100 expr: '(node_memory_MemTotal - node_memory_Cached - node_memory_Buffers - node_memory_MemFree)/node_memory_MemTotal*100 > 80' balancer: height: 150 panels: - type: custom title: "Out (eth1)" format: "Bps" expr: 'rate(node_network_transmit_bytes{device="eth1",job="balancer"}[1m])' - type: custom title: "In (eth1)" format: "Bps" expr: 'rate(node_network_receive_bytes{device="eth1",job="balancer"}[1m])' - type: custom title: "eth1" format: "Bps" fill: 8 stack: true exprs: - {expr: 'sum(rate(node_network_transmit_bytes{device="eth1",job="balancer"}[1m]))', title: "Out"} - {expr: 'sum(rate(node_network_receive_bytes{device="eth1",job="balancer"}[1m]))', title: "In"} - type: custom title: "Requests" format: "ops" expr: 'sum by (instance)(rate(http_requests_total{job="balancer"}[1m]))' - type: custom title: "Requests" format: "ops" fill: 8 stack: true exprs: - {expr: 'sum(rate(http_requests_total{code="500",job="balancer"}[1m]))', title: "500", color: red} - {expr: 'sum(rate(http_requests_total{code="404",job="balancer"}[1m]))', title: "404"} - {expr: 'sum(rate(http_requests_total{code="302",job="balancer"}[1m]))', title: "302"} - {expr: 'sum(rate(http_requests_total{code="200",job="balancer"}[1m]))', title: "200"} front: height: 150 panels: - type: custom title: "Requests" format: "ops" expr: 'sum by (instance)(rate(http_requests_total{job="front"}[1m]))' - type: custom title: "Requests" format: "ops" fill: 8 stack: true exprs: - {expr: 'sum(rate(http_requests_total{code="500",job="front"}[1m]))', title: "500", color: red} - {expr: 'sum(rate(http_requests_total{code="404",job="front"}[1m]))', title: "404"} - {expr: 'sum(rate(http_requests_total{code="302",job="front"}[1m]))', title: "302"} - {expr: 'sum(rate(http_requests_total{code="200",job="front"}[1m]))', title: "200"} - type: custom title: "CPU" max: 100 expr: '100 - (avg by (instance)(rate(node_cpu{mode="idle",job="front"}[5m]))*100)' - type: custom title: "Used Memory, %" max: 100 expr: '(node_memory_MemTotal{job="front"} - node_memory_Cached{job="front"} - node_memory_Buffers{job="front"} - node_memory_MemFree{job="front"})/node_memory_MemTotal{job="front"}*100'