containers.go

package definitions

import (
	"fmt"

	"github.com/sourcegraph/sourcegraph/monitoring/definitions/shared"
	"github.com/sourcegraph/sourcegraph/monitoring/monitoring"
)

func Containers() *monitoring.Dashboard {
	var (
		// HACK:
		// TODO: This is no longer true, we can clean this up.
		// Image names are defined in enterprise package
		// github.com/sourcegraph/sourcegraph/dev/ci/images
		// Hence we can't use the exported names in OSS here.
		// Also, the exported names do not cover edge cases such as `pgsql`, `codeintel-db`, and `codeinsights-db`.
		// We cannot use "wildcard" to cover all running containers:
		// On Kubernetes, prometheus could scrape containers from other namespaces
		// On docker-compose, prometheus could scrape non-sourcegraph containers running on the same host.
		// Therefore, we need to explicitly define the container names and track changes using Code Monitor
		// https://k8s.sgdev.org/code-monitoring/Q29kZU1vbml0b3I6MTQ=
		// Whenever we're notified, we need to:
		// - review what's changed in the commits
		// - check if the commit contains changes to the container name query in each dashboard definition
		// - update this container name query accordingly
		containerNameQuery = shared.CadvisorContainerNameMatcher("(frontend|sourcegraph-frontend|gitserver|pgsql|codeintel-db|codeinsights|precise-code-intel-worker|prometheus|redis-cache|redis-store|redis-exporter|repo-updater|searcher|symbols|syntect-server|worker|zoekt-indexserver|zoekt-webserver|indexed-search|grafana|blobstore|jaeger)")
	)

	return &monitoring.Dashboard{
		Name:                     "containers",
		Title:                    "Global Containers Resource Usage",
		Description:              "Container usage and provisioning indicators of all services.",
		NoSourcegraphDebugServer: true,
		Groups: []monitoring.Group{
			{
				Title: "Containers (not available on server)",
				// This chart is extremely noisy on k8s, so we hide it by default.
				Hidden: true,
				Rows: []monitoring.Row{
					{
						monitoring.Observable{
							Name:        "container_memory_usage",
							Description: "container memory usage of all services",
							Query:       fmt.Sprintf(`cadvisor_container_memory_usage_percentage_total{%s}`, containerNameQuery),
							NoAlert:     true,
							Panel:       monitoring.Panel().With(monitoring.PanelOptions.LegendOnRight()).LegendFormat("{{name}}").Unit(monitoring.Percentage).Interval(100).Max(100).Min(0),
							Owner:       monitoring.ObservableOwnerInfraOrg,
							Interpretation: `
								This value indicates the memory usage of all containers.
							`,
						},
					},
					{
						monitoring.Observable{
							Name:        "container_cpu_usage",
							Description: "container cpu usage total (1m average) across all cores by instance",
							Query:       fmt.Sprintf(`cadvisor_container_cpu_usage_percentage_total{%s}`, containerNameQuery),
							NoAlert:     true,
							Panel:       monitoring.Panel().With(monitoring.PanelOptions.LegendOnRight()).LegendFormat("{{name}}").Unit(monitoring.Percentage).Interval(100).Max(100).Min(0),
							Owner:       monitoring.ObservableOwnerInfraOrg,
							Interpretation: `
								This value indicates the CPU usage of all containers.
							`,
						},
					},
				},
			},
			{
				Title:  "Containers: Resource utilization (not available on server)",
				Hidden: false,
				Rows: []monitoring.Row{
					{
						monitoring.Observable{
							Name:        "container_memory_usage_provisioning",
							Description: "container memory usage (5m maximum) of services that exceed 80% memory limit",
							Query:       fmt.Sprintf(`max_over_time(cadvisor_container_memory_usage_percentage_total{%s}[5m]) >= 80`, containerNameQuery),
							NoAlert:     true,
							Panel:       monitoring.Panel().With(monitoring.PanelOptions.LegendOnRight()).LegendFormat("{{name}}").Unit(monitoring.Percentage).Interval(100).Max(100).Min(0),
							Owner:       monitoring.ObservableOwnerInfraOrg,
							Interpretation: `
								Containers that exceed 80% memory limit. The value indicates potential underprovisioned resources.
							`,
						},
					},
					{
						monitoring.Observable{
							Name:        "container_cpu_usage_provisioning",
							Description: "container cpu usage total (5m maximum) across all cores of services that exceed 80% cpu limit",
							Query:       fmt.Sprintf(`max_over_time(cadvisor_container_cpu_usage_percentage_total{%s}[5m]) >= 80`, containerNameQuery),
							NoAlert:     true,
							Panel:       monitoring.Panel().With(monitoring.PanelOptions.LegendOnRight()).LegendFormat("{{name}}").Unit(monitoring.Percentage).Interval(100).Max(100).Min(0),
							Owner:       monitoring.ObservableOwnerInfraOrg,
							Interpretation: `
								Containers that exceed 80% CPU limit. The value indicates potential underprovisioned resources.
							`,
						},
					},
					{
						monitoring.Observable{
							Name:        "container_oomkill_events_total",
							Description: "container OOMKILL events total",
							Query:       fmt.Sprintf(`max by (name) (container_oom_events_total{%s}) >= 1`, containerNameQuery),
							NoAlert:     true,
							Panel:       monitoring.Panel().With(monitoring.PanelOptions.LegendOnRight()).LegendFormat("{{name}}"),
							Owner:       monitoring.ObservableOwnerInfraOrg,
							Interpretation: `
								This value indicates the total number of times the container main process or child processes were terminated by OOM killer.
								When it occurs frequently, it is an indicator of underprovisioning.
							`,
						},
					},
					{
						monitoring.Observable{
							Name:        "container_missing",
							Description: "container missing",
							// inspired by https://awesome-prometheus-alerts.grep.to/rules#docker-containers
							Query:   fmt.Sprintf(`count by(name) ((time() - container_last_seen{%s}) > 60)`, containerNameQuery),
							NoAlert: true,
							Panel:   monitoring.Panel().With(monitoring.PanelOptions.LegendOnRight()).LegendFormat("{{name}}"),
							Owner:   monitoring.ObservableOwnerInfraOrg,
							Interpretation: `
								This value is the number of times a container has not been seen for more than one minute. If you observe this
								value change independent of deployment events (such as an upgrade), it could indicate pods are being OOM killed or terminated for some other reasons.
							`,
						},
					},
				},
			},
		},
	}
}