ivanch/litmus-hub
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

updates for portal dashboards (#485)

Browse Source 
Signed-off-by: ishangupta-ds <ishan@chaosnative.com>
This commit is contained in:
Ishan Gupta
2021-07-14 20:14:15 +05:30
committed by GitHub
parent b3e542499f
commit 844a1453ee
3 changed files with 826 additions and 761 deletions
Download Patch File Download Diff File Expand all files Collapse all files

349
monitoring/dashboards/litmus-portal/node_metrics.json
View File

@@ -1,144 +1,209 @@
{ {
"dashboardID": "generic_node_metrics", "dashboardID": "generic_node_metrics",
"name": "Node metrics", "name": "Node metrics",
"information": "This dashboard visualizes Node level CPU, memory, disk and IO utilization metrics interleaved with chaos events.", "information": "This dashboard visualizes Node level CPU, memory, disk and IO utilization metrics interleaved with chaos events.",
"chaosEventQueryTemplate": "litmuschaos_awaited_experiments{job=\"chaos-exporter\"}", "chaosEventQueryTemplate": "litmuschaos_awaited_experiments{job=\"chaos-exporter\", chaos_injection_time!=\"\"}",
"chaosVerdictQueryTemplate": "litmuschaos_experiment_verdict{job=\"chaos-exporter\"}", "chaosVerdictQueryTemplate": "litmuschaos_experiment_verdict{job=\"chaos-exporter\", chaosresult_verdict!=\"Awaited\"}",
"panelGroupMap": [ "panelGroupMap": [
{ {
"groupName": "CPU Utilization Metrics", "groupName": "CPU Utilization Metrics",
"panels": ["Chaos-Node-CPU Utilization"] "panels": ["Chaos-Node-CPU Utilization"]
}, },
{ {
"groupName": "Memory Utilization Metrics", "groupName": "Memory Utilization Metrics",
"panels": ["Chaos-Node-Memory Utilization"] "panels": ["Chaos-Node-Memory Utilization"]
}, },
{ {
"groupName": "Disk Usage Metrics", "groupName": "Disk Usage Metrics",
"panels": [ "panels": [
"Chaos-Node-Disk I/O Usage R/W", "Chaos-Node-Disk I/O Usage R/W",
"Chaos-Node-Disk I/O Usage Times" "Chaos-Node-Disk I/O Usage Times"
] ]
}, },
{ {
"groupName": "Network Usage Metrics", "groupName": "Network Usage Metrics",
"panels": [ "panels": [
"Chaos-Node-Network Traffic Bytes", "Chaos-Node-Network Traffic Bits",
"Chaos-Node-Network Traffic Packets" "Chaos-Node-Network Traffic Packets"
] ]
} }
], ],
"panelGroups": [ "panelGroups": [
{ {
"panel_group_name": "CPU Usage Metrics", "panel_group_name": "CPU Utilization Metrics",
"panels": [ "panels": [
{ {
"panel_name": "Chaos-Node-CPU Utilization", "panel_name": "Chaos-Node-CPU Utilization",
"panel_options": { "panel_options": {
"points": false, "points": false,
"grids": true, "grids": true,
"left_axis": true "left_axis": true
}, },
"y_axis_left": "Cores", "y_axis_left": "Cores",
"y_axis_right": "CHAOS", "y_axis_right": "CHAOS",
"x_axis_down": "Time", "x_axis_down": "Time",
"unit": "%", "unit": "%",
"prom_queries": [ "prom_queries": [
{ {
"prom_query_name": "instance:node_cpu_utilisation:rate1m*100", "prom_query_name": "instance:node_cpu_utilisation:rate1m*100",
"legend": "{{instance}}", "legend": "{{instance}}",
"resolution": "1/2", "resolution": "1/2",
"minstep": "5", "minstep": "5",
"line": true, "line": true,
"close_area": false "close_area": false
} }
] ]
} }
] ]
}, },
{ {
"panel_group_name": "Memory Usage Metrics", "panel_group_name": "Memory Utilization Metrics",
"panels": [ "panels": [
{ {
"panel_name": "Chaos-Node-Memory Utilization", "panel_name": "Chaos-Node-Memory Utilization",
"panel_options": { "panel_options": {
"points": false, "points": false,
"grids": true, "grids": true,
"left_axis": true "left_axis": true
}, },
"y_axis_left": "Memory", "y_axis_left": "Memory",
"y_axis_right": "CHAOS", "y_axis_right": "CHAOS",
"x_axis_down": "Time", "x_axis_down": "Time",
"unit": "%", "unit": "%",
"prom_queries": [ "prom_queries": [
{ {
"prom_query_name": "instance:node_memory_utilisation:ratio*100", "prom_query_name": "instance:node_memory_utilisation:ratio*100",
"legend": "{{instance}}", "legend": "{{instance}}",
"resolution": "1/2", "resolution": "1/2",
"minstep": "5", "minstep": "5",
"line": true, "line": true,
"close_area": false "close_area": false
} }
] ]
} }
] ]
}, },
{ {
"panel_group_name": "Disk Usage Metrics", "panel_group_name": "Disk Usage Metrics",
"panels": [ "panels": [
{ {
"panel_name": "Chaos-Node-Disk I/O Usage R/W", "panel_name": "Chaos-Node-Disk I/O Usage R/W",
"panel_options": { "panel_options": {
"points": false, "points": false,
"grids": true, "grids": true,
"left_axis": true "left_axis": true
}, },
"y_axis_left": "bytes read (-) / write (+)", "y_axis_left": "bytes read (-) / write (+)",
"y_axis_right": "CHAOS", "y_axis_right": "CHAOS",
"x_axis_down": "Time", "x_axis_down": "Time",
"unit": "KiB", "unit": "KiB",
"prom_queries": [ "prom_queries": [
{ {
"prom_query_name": "node_disk_read_bytes_total", "prom_query_name": "node_disk_read_bytes_total",
"legend": "{{instance}} - {{device}} - Successfully read bytes", "legend": "{{instance}} - {{device}} - Successfully read bytes",
"resolution": "1/2", "resolution": "1/2",
"minstep": "5", "minstep": "5",
"line": true, "line": true,
"close_area": false "close_area": false
}, },
{ {
"prom_query_name": "node_disk_written_bytes_total", "prom_query_name": "node_disk_written_bytes_total",
"legend": "{{instance}} - {{device}} - Successfully written bytes", "legend": "{{instance}} - {{device}} - Successfully written bytes",
"resolution": "1/2", "resolution": "1/2",
"minstep": "5", "minstep": "5",
"line": true, "line": true,
"close_area": false "close_area": false
} }
] ]
}, },
{ {
"panel_name": "Chaos-Node-Disk I/O Usage Times", "panel_name": "Chaos-Node-Disk I/O Usage Times",
"panel_options": { "panel_options": {
"points": false, "points": false,
"grids": true, "grids": true,
"left_axis": true "left_axis": true
}, },
"y_axis_left": "time", "y_axis_left": "time",
"y_axis_right": "CHAOS", "y_axis_right": "CHAOS",
"x_axis_down": "Time", "x_axis_down": "Time",
"unit": "ms", "unit": "ms",
"prom_queries": [ "prom_queries": [
{ {
"prom_query_name": "node_disk_io_time_seconds_total", "prom_query_name": "node_disk_io_time_seconds_total",
"legend": "{{instance}} - {{device}} - Time spent doing I/Os", "legend": "{{instance}} - {{device}} - Time spent doing I/Os",
"resolution": "1/2", "resolution": "1/2",
"minstep": "5", "minstep": "5",
"line": true, "line": true,
"close_area": false "close_area": false
} }
] ]
} }
] ]
} },
] {
"panel_group_name": "Network Usage Metrics",
"panels": [
{
"panel_name": "Chaos-Node-Network Traffic Bits",
"panel_options": {
"points": false,
"grids": true,
"left_axis": true
},
"y_axis_left": "bits out (-) / in (+)",
"y_axis_right": "CHAOS",
"x_axis_down": "Time",
"unit": "b/s",
"prom_queries": [
{
"prom_query_name": "node_network_receive_bytes_total*8",
"legend": "{{instance}} - {{device}} - Receive",
"resolution": "1/2",
"minstep": "5",
"line": true,
"close_area": false
},
{
"prom_query_name": "node_network_transmit_bytes_total*8",
"legend": "{{instance}} - {{device}} - Transmit",
"resolution": "1/2",
"minstep": "5",
"line": true,
"close_area": false
}
]
},
{
"panel_name": "Chaos-Node-Network Traffic Packets",
"panel_options": {
"points": false,
"grids": true,
"left_axis": true
},
"y_axis_left": "packets out (-) / in (+)",
"y_axis_right": "CHAOS",
"x_axis_down": "Time",
"unit": "p/s",
"prom_queries": [
{
"prom_query_name": "node_network_receive_packets_total",
"legend": "{{instance}} - {{device}} - Receive",
"resolution": "1/2",
"minstep": "5",
"line": true,
"close_area": false
},
{
"prom_query_name": "node_network_transmit_packets_total",
"legend": "{{instance}} - {{device}} - Transmit",
"resolution": "1/2",
"minstep": "5",
"line": true,
"close_area": false
}
]
}
]
}
]
} }

142
monitoring/dashboards/litmus-portal/pod_metrics.json
View File

@@ -1,73 +1,73 @@
{ {
"dashboardID": "generic_pod_metrics", "dashboardID": "generic_pod_metrics",
"name": "Pod metrics", "name": "Pod metrics",
"information": "This dashboard visualizes Pod level CPU and memory usage metrics interleaved with chaos events.", "information": "This dashboard visualizes Pod level CPU and memory usage metrics interleaved with chaos events.",
"chaosEventQueryTemplate": "litmuschaos_awaited_experiments{job=\"chaos-exporter\"}", "chaosEventQueryTemplate": "litmuschaos_awaited_experiments{job=\"chaos-exporter\", chaos_injection_time!=\"\"}",
"chaosVerdictQueryTemplate": "litmuschaos_experiment_verdict{job=\"chaos-exporter\"}", "chaosVerdictQueryTemplate": "litmuschaos_experiment_verdict{job=\"chaos-exporter\", chaosresult_verdict!=\"Awaited\"}",
"panelGroupMap": [ "panelGroupMap": [
{ {
"groupName": "CPU Usage Metrics", "groupName": "CPU Usage Metrics",
"panels": ["Chaos-Pod-CPU Usage"] "panels": ["Chaos-Pod-CPU Usage"]
}, },
{ {
"groupName": "Memory Usage Metrics", "groupName": "Memory Usage Metrics",
"panels": ["Chaos-Pod-Memory Usage"] "panels": ["Chaos-Pod-Memory Usage"]
} }
], ],
"panelGroups": [ "panelGroups": [
{ {
"panel_group_name": "CPU Usage Metrics", "panel_group_name": "CPU Usage Metrics",
"panels": [ "panels": [
{ {
"panel_name": "Chaos-Pod-CPU Usage", "panel_name": "Chaos-Pod-CPU Usage",
"panel_options": { "panel_options": {
"points": false, "points": false,
"grids": true, "grids": true,
"left_axis": true "left_axis": true
}, },
"y_axis_left": "Cores", "y_axis_left": "Cores",
"y_axis_right": "CHAOS", "y_axis_right": "CHAOS",
"x_axis_down": "Time", "x_axis_down": "Time",
"unit": "", "unit": "",
"prom_queries": [ "prom_queries": [
{ {
"prom_query_name": "sum(rate(container_cpu_usage_seconds_total{container!=\"POD\",pod!=\"\"}[5m])) by (pod)", "prom_query_name": "sum(rate(container_cpu_usage_seconds_total{container!=\"POD\",pod!=\"\"}[5m])) by (pod)",
"legend": "{{pod}}", "legend": "{{pod}}",
"resolution": "1/2", "resolution": "1/2",
"minstep": "5", "minstep": "5",
"line": true, "line": true,
"close_area": false "close_area": false
} }
] ]
} }
] ]
}, },
{ {
"panel_group_name": "Memory Usage Metrics", "panel_group_name": "Memory Usage Metrics",
"panels": [ "panels": [
{ {
"panel_name": "Chaos-Pod-Memory Usage", "panel_name": "Chaos-Pod-Memory Usage",
"panel_options": { "panel_options": {
"points": false, "points": false,
"grids": true, "grids": true,
"left_axis": true "left_axis": true
}, },
"y_axis_left": "Memory", "y_axis_left": "Memory",
"y_axis_right": "CHAOS", "y_axis_right": "CHAOS",
"x_axis_down": "Time", "x_axis_down": "Time",
"unit": "GiB", "unit": "GiB",
"prom_queries": [ "prom_queries": [
{ {
"prom_query_name": "sum(container_memory_usage_bytes{container!=\"POD\",container!=\"\"}) by (pod)", "prom_query_name": "sum(container_memory_usage_bytes{container!=\"POD\",container!=\"\"}) by (pod)",
"legend": "{{pod}}", "legend": "{{pod}}",
"resolution": "1/2", "resolution": "1/2",
"minstep": "5", "minstep": "5",
"line": true, "line": true,
"close_area": false "close_area": false
} }
] ]
} }
] ]
} }
] ]
} }

1096
monitoring/dashboards/litmus-portal/sock-shop.json
View File

File diff suppressed because it is too large Load Diff