[v9.1.x] Metrics: Fixed grafana_database_conn_* metrics, and added new go_sql_stats_* metrics as eventual replacement (#54540)
Signed-off-by: Dave Henderson <dave.henderson@grafana.com> Signed-off-by: Dave Henderson <dave.henderson@grafana.com>
This commit is contained in:
Dave Henderson
@@ -11,6 +11,7 @@ import (
|
||||
"sync"
|
||||
"time"
|
||||
|
||||
"github.com/dlmiddlecote/sqlstats"
|
||||
"github.com/go-sql-driver/mysql"
|
||||
_ "github.com/lib/pq"
|
||||
"github.com/prometheus/client_golang/prometheus"
|
||||
@@ -54,17 +55,6 @@ type SQLStore struct {
|
||||
skipEnsureDefaultOrgAndUser bool
|
||||
migrations registry.DatabaseMigrator
|
||||
tracer tracing.Tracer
|
||||
metrics struct {
|
||||
maxOpenConnections prometheus.Gauge
|
||||
openConnections prometheus.Gauge
|
||||
inUse prometheus.Gauge
|
||||
idle prometheus.Gauge
|
||||
waitCount prometheus.Counter
|
||||
waitDuration prometheus.Counter
|
||||
maxIdleClosed prometheus.Counter
|
||||
maxIdleTimeClosed prometheus.Counter
|
||||
maxLifetimeClosed prometheus.Counter
|
||||
}
|
||||
}
|
||||
|
||||
func ProvideService(cfg *setting.Cfg, cacheService *localcache.CacheService, migrations registry.DatabaseMigrator, bus bus.Bus, tracer tracing.Tracer) (*SQLStore, error) {
|
||||
@@ -86,9 +76,13 @@ func ProvideService(cfg *setting.Cfg, cacheService *localcache.CacheService, mig
|
||||
}
|
||||
s.tracer = tracer
|
||||
|
||||
s.initMetrics()
|
||||
// initialize and register metrics wrapper around the *sql.DB
|
||||
db := s.engine.DB().DB
|
||||
|
||||
prometheus.MustRegister(s)
|
||||
// register the go_sql_stats_connections_* metrics
|
||||
prometheus.MustRegister(sqlstats.NewStatsCollector("grafana", db))
|
||||
// TODO: deprecate/remove these metrics
|
||||
prometheus.MustRegister(newSQLStoreMetrics(db))
|
||||
|
||||
return s, nil
|
||||
}
|
||||
@@ -442,118 +436,6 @@ func (ss *SQLStore) readConfig() error {
|
||||
return nil
|
||||
}
|
||||
|
||||
// initMetrics initializes the database connection metrics
|
||||
func (ss *SQLStore) initMetrics() {
|
||||
namespace := "grafana"
|
||||
subsystem := "database"
|
||||
|
||||
ss.metrics.maxOpenConnections = prometheus.NewGauge(prometheus.GaugeOpts{
|
||||
Namespace: namespace,
|
||||
Subsystem: subsystem,
|
||||
Name: "conn_max_open",
|
||||
Help: "Maximum number of open connections to the database",
|
||||
})
|
||||
|
||||
ss.metrics.openConnections = prometheus.NewGauge(prometheus.GaugeOpts{
|
||||
Namespace: namespace,
|
||||
Subsystem: subsystem,
|
||||
Name: "conn_open",
|
||||
Help: "The number of established connections both in use and idle",
|
||||
})
|
||||
|
||||
ss.metrics.inUse = prometheus.NewGauge(prometheus.GaugeOpts{
|
||||
Namespace: namespace,
|
||||
Subsystem: subsystem,
|
||||
Name: "conn_in_use",
|
||||
Help: "The number of connections currently in use",
|
||||
})
|
||||
|
||||
ss.metrics.idle = prometheus.NewGauge(prometheus.GaugeOpts{
|
||||
Namespace: namespace,
|
||||
Subsystem: subsystem,
|
||||
Name: "conn_idle",
|
||||
Help: "The number of idle connections",
|
||||
})
|
||||
|
||||
ss.metrics.waitCount = prometheus.NewCounter(prometheus.CounterOpts{
|
||||
Namespace: namespace,
|
||||
Subsystem: subsystem,
|
||||
Name: "conn_wait_count_total",
|
||||
Help: "The total number of connections waited for",
|
||||
})
|
||||
|
||||
ss.metrics.waitDuration = prometheus.NewCounter(prometheus.CounterOpts{
|
||||
Namespace: namespace,
|
||||
Subsystem: subsystem,
|
||||
Name: "conn_wait_duration_seconds",
|
||||
Help: "The total time blocked waiting for a new connection",
|
||||
})
|
||||
|
||||
ss.metrics.maxIdleClosed = prometheus.NewCounter(prometheus.CounterOpts{
|
||||
Namespace: namespace,
|
||||
Subsystem: subsystem,
|
||||
Name: "conn_max_idle_closed_total",
|
||||
Help: "The total number of connections closed due to SetMaxIdleConns",
|
||||
})
|
||||
|
||||
ss.metrics.maxIdleTimeClosed = prometheus.NewCounter(prometheus.CounterOpts{
|
||||
Namespace: namespace,
|
||||
Subsystem: subsystem,
|
||||
Name: "conn_max_idle_closed_seconds",
|
||||
Help: "The total number of connections closed due to SetConnMaxIdleTime",
|
||||
})
|
||||
|
||||
ss.metrics.maxLifetimeClosed = prometheus.NewCounter(prometheus.CounterOpts{
|
||||
Namespace: namespace,
|
||||
Subsystem: subsystem,
|
||||
Name: "conn_max_lifetime_closed_total",
|
||||
Help: "The total number of connections closed due to SetConnMaxLifetime",
|
||||
})
|
||||
}
|
||||
|
||||
// collectDBStats instruments connections stats from the database.
|
||||
func (ss *SQLStore) collectDBstats() {
|
||||
dbstats := ss.engine.DB().Stats()
|
||||
ss.metrics.maxOpenConnections.Set(float64(dbstats.MaxOpenConnections))
|
||||
ss.metrics.openConnections.Set(float64(dbstats.MaxOpenConnections))
|
||||
ss.metrics.inUse.Set(float64(dbstats.InUse))
|
||||
ss.metrics.idle.Set(float64(dbstats.Idle))
|
||||
|
||||
ss.metrics.waitCount.Add(float64(dbstats.WaitCount))
|
||||
ss.metrics.waitDuration.Add(float64(dbstats.WaitDuration / time.Second))
|
||||
ss.metrics.maxIdleClosed.Add(float64(dbstats.MaxIdleClosed))
|
||||
ss.metrics.maxIdleTimeClosed.Add(float64(dbstats.MaxIdleTimeClosed))
|
||||
ss.metrics.maxLifetimeClosed.Add(float64(dbstats.MaxLifetimeClosed))
|
||||
}
|
||||
|
||||
// Collect implements Prometheus.Collector.
|
||||
func (ss *SQLStore) Collect(ch chan<- prometheus.Metric) {
|
||||
ss.collectDBstats()
|
||||
|
||||
ss.metrics.maxOpenConnections.Collect(ch)
|
||||
ss.metrics.openConnections.Collect(ch)
|
||||
ss.metrics.inUse.Collect(ch)
|
||||
ss.metrics.idle.Collect(ch)
|
||||
ss.metrics.waitCount.Collect(ch)
|
||||
ss.metrics.waitDuration.Collect(ch)
|
||||
ss.metrics.maxIdleClosed.Collect(ch)
|
||||
ss.metrics.maxIdleTimeClosed.Collect(ch)
|
||||
ss.metrics.maxLifetimeClosed.Collect(ch)
|
||||
}
|
||||
|
||||
// Describe implements Prometheus.Collector.
|
||||
func (ss *SQLStore) Describe(ch chan<- *prometheus.Desc) {
|
||||
ss.metrics.maxOpenConnections.Describe(ch)
|
||||
ss.metrics.openConnections.Describe(ch)
|
||||
ss.metrics.inUse.Describe(ch)
|
||||
ss.metrics.idle.Describe(ch)
|
||||
ss.metrics.waitCount.Describe(ch)
|
||||
ss.metrics.waitDuration.Describe(ch)
|
||||
ss.metrics.maxIdleClosed.Describe(ch)
|
||||
ss.metrics.maxIdleTimeClosed.Describe(ch)
|
||||
ss.metrics.maxLifetimeClosed.Describe(ch)
|
||||
}
|
||||
|
||||
// ITestDB is an interface of arguments for testing db
|
||||
type ITestDB interface {
|
||||
Helper()
|
||||
|
||||
@@ -0,0 +1,144 @@
|
||||
package sqlstore
|
||||
|
||||
import (
|
||||
"github.com/dlmiddlecote/sqlstats"
|
||||
"github.com/prometheus/client_golang/prometheus"
|
||||
)
|
||||
|
||||
type sqlStoreMetrics struct {
|
||||
db sqlstats.StatsGetter
|
||||
|
||||
// gauges
|
||||
maxOpenConnections *prometheus.Desc
|
||||
openConnections *prometheus.Desc
|
||||
inUse *prometheus.Desc
|
||||
idle *prometheus.Desc
|
||||
|
||||
// counters
|
||||
waitCount *prometheus.Desc
|
||||
waitDuration *prometheus.Desc
|
||||
maxIdleClosed *prometheus.Desc
|
||||
maxIdleTimeClosed *prometheus.Desc
|
||||
maxLifetimeClosed *prometheus.Desc
|
||||
}
|
||||
|
||||
func newSQLStoreMetrics(db sqlstats.StatsGetter) *sqlStoreMetrics {
|
||||
ns := "grafana"
|
||||
sub := "database"
|
||||
|
||||
return &sqlStoreMetrics{
|
||||
db: db,
|
||||
maxOpenConnections: prometheus.NewDesc(
|
||||
prometheus.BuildFQName(ns, sub, "conn_max_open"),
|
||||
"Maximum number of open connections to the database",
|
||||
nil, nil,
|
||||
),
|
||||
openConnections: prometheus.NewDesc(
|
||||
prometheus.BuildFQName(ns, sub, "conn_open"),
|
||||
"The number of established connections both in use and idle",
|
||||
nil, nil,
|
||||
),
|
||||
inUse: prometheus.NewDesc(
|
||||
prometheus.BuildFQName(ns, sub, "conn_in_use"),
|
||||
"The number of connections currently in use",
|
||||
nil, nil,
|
||||
),
|
||||
idle: prometheus.NewDesc(
|
||||
prometheus.BuildFQName(ns, sub, "conn_idle"),
|
||||
"The number of idle connections",
|
||||
nil, nil,
|
||||
),
|
||||
|
||||
waitCount: prometheus.NewDesc(
|
||||
prometheus.BuildFQName(ns, sub, "conn_wait_count_total"),
|
||||
"The total number of connections waited for",
|
||||
nil, nil,
|
||||
),
|
||||
waitDuration: prometheus.NewDesc(
|
||||
prometheus.BuildFQName(ns, sub, "conn_wait_duration_seconds"),
|
||||
"The total time blocked waiting for a new connection",
|
||||
nil, nil,
|
||||
),
|
||||
maxIdleClosed: prometheus.NewDesc(
|
||||
prometheus.BuildFQName(ns, sub, "conn_max_idle_closed_total"),
|
||||
"The total number of connections closed due to SetMaxIdleConns",
|
||||
nil, nil,
|
||||
),
|
||||
maxIdleTimeClosed: prometheus.NewDesc(
|
||||
prometheus.BuildFQName(ns, sub, "conn_max_idle_closed_seconds"),
|
||||
"The total number of connections closed due to SetConnMaxIdleTime",
|
||||
nil, nil,
|
||||
),
|
||||
maxLifetimeClosed: prometheus.NewDesc(
|
||||
prometheus.BuildFQName(ns, sub, "conn_max_lifetime_closed_total"),
|
||||
"The total number of connections closed due to SetConnMaxLifetime",
|
||||
nil, nil,
|
||||
),
|
||||
}
|
||||
}
|
||||
|
||||
// Collect implements Prometheus.Collector.
|
||||
func (m *sqlStoreMetrics) Collect(ch chan<- prometheus.Metric) {
|
||||
stats := m.db.Stats()
|
||||
|
||||
ch <- prometheus.MustNewConstMetric(
|
||||
m.maxOpenConnections,
|
||||
prometheus.GaugeValue,
|
||||
float64(stats.MaxOpenConnections),
|
||||
)
|
||||
ch <- prometheus.MustNewConstMetric(
|
||||
m.openConnections,
|
||||
prometheus.GaugeValue,
|
||||
float64(stats.OpenConnections),
|
||||
)
|
||||
ch <- prometheus.MustNewConstMetric(
|
||||
m.inUse,
|
||||
prometheus.GaugeValue,
|
||||
float64(stats.InUse),
|
||||
)
|
||||
ch <- prometheus.MustNewConstMetric(
|
||||
m.idle,
|
||||
prometheus.GaugeValue,
|
||||
float64(stats.Idle),
|
||||
)
|
||||
|
||||
ch <- prometheus.MustNewConstMetric(
|
||||
m.waitCount,
|
||||
prometheus.CounterValue,
|
||||
float64(stats.WaitCount),
|
||||
)
|
||||
ch <- prometheus.MustNewConstMetric(
|
||||
m.waitDuration,
|
||||
prometheus.CounterValue,
|
||||
stats.WaitDuration.Seconds(),
|
||||
)
|
||||
ch <- prometheus.MustNewConstMetric(
|
||||
m.maxIdleClosed,
|
||||
prometheus.CounterValue,
|
||||
float64(stats.MaxIdleClosed),
|
||||
)
|
||||
ch <- prometheus.MustNewConstMetric(
|
||||
m.maxIdleTimeClosed,
|
||||
prometheus.CounterValue,
|
||||
float64(stats.MaxIdleTimeClosed),
|
||||
)
|
||||
ch <- prometheus.MustNewConstMetric(
|
||||
m.maxLifetimeClosed,
|
||||
prometheus.CounterValue,
|
||||
float64(stats.MaxLifetimeClosed),
|
||||
)
|
||||
}
|
||||
|
||||
// Describe implements Prometheus.Collector.
|
||||
func (m *sqlStoreMetrics) Describe(ch chan<- *prometheus.Desc) {
|
||||
ch <- m.maxOpenConnections
|
||||
ch <- m.openConnections
|
||||
ch <- m.inUse
|
||||
ch <- m.idle
|
||||
|
||||
ch <- m.waitCount
|
||||
ch <- m.waitDuration
|
||||
ch <- m.maxIdleClosed
|
||||
ch <- m.maxIdleTimeClosed
|
||||
ch <- m.maxLifetimeClosed
|
||||
}
|
||||
@@ -0,0 +1,68 @@
|
||||
package sqlstore
|
||||
|
||||
import (
|
||||
"database/sql"
|
||||
"strings"
|
||||
"testing"
|
||||
"time"
|
||||
|
||||
"github.com/dlmiddlecote/sqlstats"
|
||||
"github.com/prometheus/client_golang/prometheus/testutil"
|
||||
"github.com/stretchr/testify/require"
|
||||
)
|
||||
|
||||
func TestSQLStore_Metrics(t *testing.T) {
|
||||
stats := sql.DBStats{
|
||||
MaxOpenConnections: 9,
|
||||
OpenConnections: 8,
|
||||
InUse: 4,
|
||||
Idle: 4,
|
||||
WaitCount: 5,
|
||||
WaitDuration: 6 * time.Second,
|
||||
MaxIdleClosed: 7,
|
||||
MaxIdleTimeClosed: 8,
|
||||
MaxLifetimeClosed: 9,
|
||||
}
|
||||
|
||||
m := newSQLStoreMetrics(&fakeStatsGetter{stats: stats})
|
||||
|
||||
require.NoError(t, testutil.CollectAndCompare(m, strings.NewReader(`
|
||||
# HELP grafana_database_conn_idle The number of idle connections
|
||||
# TYPE grafana_database_conn_idle gauge
|
||||
grafana_database_conn_idle 4
|
||||
# HELP grafana_database_conn_in_use The number of connections currently in use
|
||||
# TYPE grafana_database_conn_in_use gauge
|
||||
grafana_database_conn_in_use 4
|
||||
# HELP grafana_database_conn_max_idle_closed_seconds The total number of connections closed due to SetConnMaxIdleTime
|
||||
# TYPE grafana_database_conn_max_idle_closed_seconds counter
|
||||
grafana_database_conn_max_idle_closed_seconds 8
|
||||
# HELP grafana_database_conn_max_idle_closed_total The total number of connections closed due to SetMaxIdleConns
|
||||
# TYPE grafana_database_conn_max_idle_closed_total counter
|
||||
grafana_database_conn_max_idle_closed_total 7
|
||||
# HELP grafana_database_conn_max_lifetime_closed_total The total number of connections closed due to SetConnMaxLifetime
|
||||
# TYPE grafana_database_conn_max_lifetime_closed_total counter
|
||||
grafana_database_conn_max_lifetime_closed_total 9
|
||||
# HELP grafana_database_conn_max_open Maximum number of open connections to the database
|
||||
# TYPE grafana_database_conn_max_open gauge
|
||||
grafana_database_conn_max_open 9
|
||||
# HELP grafana_database_conn_open The number of established connections both in use and idle
|
||||
# TYPE grafana_database_conn_open gauge
|
||||
grafana_database_conn_open 8
|
||||
# HELP grafana_database_conn_wait_count_total The total number of connections waited for
|
||||
# TYPE grafana_database_conn_wait_count_total counter
|
||||
grafana_database_conn_wait_count_total 5
|
||||
# HELP grafana_database_conn_wait_duration_seconds The total time blocked waiting for a new connection
|
||||
# TYPE grafana_database_conn_wait_duration_seconds counter
|
||||
grafana_database_conn_wait_duration_seconds 6
|
||||
`)))
|
||||
}
|
||||
|
||||
type fakeStatsGetter struct {
|
||||
stats sql.DBStats
|
||||
}
|
||||
|
||||
var _ sqlstats.StatsGetter = (*fakeStatsGetter)(nil)
|
||||
|
||||
func (f *fakeStatsGetter) Stats() sql.DBStats {
|
||||
return f.stats
|
||||
}
|
||||
Reference in New Issue
Block a user