nomad/e2e/metrics/metrics_test.go

// Copyright (c) HashiCorp, Inc.
// SPDX-License-Identifier: BUSL-1.1

package metrics

import (
	"context"
	"fmt"
	"testing"
	"time"

	nomadapi "github.com/hashicorp/nomad/api"
	"github.com/hashicorp/nomad/e2e/e2eutil"
	"github.com/hashicorp/nomad/e2e/v3/cluster3"
	"github.com/hashicorp/nomad/e2e/v3/jobs3"
	promapi "github.com/prometheus/client_golang/api"
	promapi1 "github.com/prometheus/client_golang/api/prometheus/v1"
	promodel "github.com/prometheus/common/model"
	"github.com/shoenig/test"
	"github.com/shoenig/test/must"
)

type metric struct {
	name   string
	filter string
	key    string
	value  float64
	sum    bool
}

func (m *metric) String() string {
	return fmt.Sprintf("%s[%s]=%v", m.name, m.key, m.value)
}

func (m *metric) Query() string {
	query := fmt.Sprintf("%s{%s=%q}", m.name, m.filter, m.key)
	if m.sum {
		query = "sum(" + query + ")"
	}
	return query
}

func TestMetrics(t *testing.T) {
	// Run via the e2e suite. Requires AWS attributes.

	// Wait for the cluster to be ready.
	cluster3.Establish(t,
		cluster3.Leader(),
		cluster3.LinuxClients(1),
	)

	t.Log("tweaking podman registry auth files ...")
	_, cleanupSetup := jobs3.Submit(t, "./input/setup.hcl")
	t.Cleanup(cleanupSetup)

	t.Log("running metrics job cpustress ...")
	jobCPU, cleanupCPU := jobs3.Submit(t, "./input/cpustress.hcl")
	t.Cleanup(cleanupCPU)

	t.Log("running metrics job nomadagent ...")
	jobHP, cleanupHP := jobs3.Submit(t, "./input/nomadagent.hcl")
	t.Cleanup(cleanupHP)

	t.Log("running metrics job prometheus ...")
	_, cleanupProm := jobs3.Submit(t, "./input/prometheus.hcl", jobs3.Timeout(60*time.Second))
	t.Cleanup(cleanupProm)

	t.Log("running metrics job pythonhttp ...")
	jobPy, cleanupPy := jobs3.Submit(t, "./input/pythonhttp.hcl")
	t.Cleanup(cleanupPy)

	t.Log("running metrics job caddy ...")
	_, cleanupCaddy := jobs3.Submit(t, "./input/caddy.hcl")
	t.Cleanup(cleanupCaddy)

	t.Log("let the metrics collect for a bit (10s) ...")
	time.Sleep(10 * time.Second)

	t.Log("measuring alloc metrics ...")
	testAllocMetrics(t, []*metric{{
		name:   "nomad_client_allocs_memory_usage",
		filter: "exported_job",
		key:    jobHP.JobID(),
	}, {
		name:   "nomad_client_allocs_cpu_user",
		filter: "exported_job",
		key:    jobCPU.JobID(),
	}, {
		name:   "nomad_client_allocs_cpu_allocated",
		filter: "exported_job",
		key:    jobPy.JobID(),
	}})

	t.Log("measuring client metrics ...")
	testClientMetrics(t, []*metric{{
		name: "nomad_client_allocated_memory",
	}, {
		name: "nomad_client_host_cpu_user",
		sum:  true, // metric is per core
	}, {
		name: "nomad_client_host_memory_used",
	}, {
		name: "nomad_client_uptime",
	}})

}

func testAllocMetrics(t *testing.T, metrics []*metric) {
	// query metrics and update values
	query(t, metrics)

	// assert each metric has a positive value
	positives(t, metrics)
}

func testClientMetrics(t *testing.T, metrics []*metric) {
	nodes, _, err := e2eutil.NomadClient(t).Nodes().List(&nomadapi.QueryOptions{
		Filter: fmt.Sprintf("Attributes[%q] == %q", "kernel.name", "linux"),
	})
	must.NoError(t, err)

	// permute each metric per node
	results := make([]*metric, 0, len(nodes)*len(metrics))
	for _, node := range nodes {
		for _, m := range metrics {
			results = append(results, &metric{
				name:   m.name,
				filter: "node_id",
				key:    node.ID,
				sum:    m.sum,
			})
		}
	}

	// query metrics and update values
	query(t, results)

	// assert each metric has a positive value
	positives(t, results)
}

func query(t *testing.T, metrics []*metric) {
	services := e2eutil.NomadClient(t).Services()
	regs, _, err := services.Get("caddy", &nomadapi.QueryOptions{
		Filter: `Tags contains "expose"`,
	})
	must.NoError(t, err, must.Sprint("unable to query nomad for caddy service"))
	must.Len(t, 1, regs, must.Sprint("expected one caddy instance"))

	prom := regs[0] // tag[0] is public aws address
	address := fmt.Sprintf("http://%s:%d", prom.Tags[0], prom.Port)
	opts := promapi.Config{Address: address}
	t.Log("expose prometheus http address", address)

	client, err := promapi.NewClient(opts)
	must.NoError(t, err, must.Sprint("unable to create prometheus api client"))

	api1 := promapi1.NewAPI(client)

	// waiting for prometheus can take anywhere from 10 seconds to like 2 minutes,
	// so we keep polling the api for a long while until we get a result
	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Minute)
	defer cancel()

	for _, m := range metrics {

		// for each metric we keep polling the prometheus api until we get a result, or the overall
		// long wait context finally times out
		//
		// individual queries get a 10 second timeout
		for {

			q := m.Query()
			t.Log("query for metric", q)
			result, warnings, err := api1.Query(ctx, q, time.Now(), promapi1.WithTimeout(10*time.Second))
			must.NoError(t, err, must.Sprintf("unable to query %q", q))
			must.SliceEmpty(t, warnings, must.Sprintf("got warnings %v", warnings))

			// extract the actual value
			vector, ok := result.(promodel.Vector)
			must.True(t, ok, must.Sprint("unable to convert metric to vector"))

			// if we got an empty result we need to keep trying
			if len(vector) == 0 {
				t.Log("-> empty vector, will try again in 5 seconds")
				time.Sleep(5 * time.Second)
				continue
			} else {
				sample := vector[len(vector)-1]
				m.value = float64(sample.Value)
				break
			}
		}
	}
}

func positives(t *testing.T, metrics []*metric) {
	// just ensure each metric value is positive
	for _, m := range metrics {
		test.Positive(t, m.value, test.Sprintf(
			"%s should have been positive",
			m.String(),
		))
	}
}