Clean up code around Prometheus metrics, more DRY.
With this change, all definitions and functionality for a given metric are in a single place only instead of being distributed all over the file. This makes it easier to inspect the code for correctness and adding/changing metrics.
This commit is contained in:
Julius Volz
Julius Volz
parent
b28b26ca30
commit
fa17966229
@ -23,61 +23,52 @@ import (
|
|||||||
"github.com/prometheus/client_golang/prometheus"
|
"github.com/prometheus/client_golang/prometheus"
|
||||||
)
|
)
|
||||||
|
|
||||||
|
// This will usually be manager.Manager, but can be swapped out for testing.
|
||||||
type subcontainersInfoProvider interface {
|
type subcontainersInfoProvider interface {
|
||||||
// Get information about all subcontainers of the specified container (includes self).
|
// Get information about all subcontainers of the specified container (includes self).
|
||||||
SubcontainersInfo(containerName string, query *info.ContainerInfoRequest) ([]*info.ContainerInfo, error)
|
SubcontainersInfo(containerName string, query *info.ContainerInfoRequest) ([]*info.ContainerInfo, error)
|
||||||
}
|
}
|
||||||
|
|
||||||
type prometheusMetric struct {
|
// metricValue describes a single metric value for a given set of label values
|
||||||
valueType prometheus.ValueType
|
// within a parent containerMetric.
|
||||||
value float64
|
type metricValue struct {
|
||||||
labels []string
|
value float64
|
||||||
|
labels []string
|
||||||
|
}
|
||||||
|
|
||||||
|
type metricValues []metricValue
|
||||||
|
|
||||||
|
// fsValues is a helper method for assembling per-filesystem stats.
|
||||||
|
func fsValues(fsStats []info.FsStats, valueFn func(*info.FsStats) float64) metricValues {
|
||||||
|
values := make(metricValues, 0, len(fsStats))
|
||||||
|
for _, stat := range fsStats {
|
||||||
|
values = append(values, metricValue{
|
||||||
|
value: valueFn(&stat),
|
||||||
|
labels: []string{stat.Device},
|
||||||
|
})
|
||||||
|
}
|
||||||
|
return values
|
||||||
|
}
|
||||||
|
|
||||||
|
// A containerMetric describes a multi-dimensional metric used for exposing
|
||||||
|
// a certain type of container statistic.
|
||||||
|
type containerMetric struct {
|
||||||
|
name string
|
||||||
|
help string
|
||||||
|
valueType prometheus.ValueType
|
||||||
|
extraLabels []string
|
||||||
|
getValues func(s *info.ContainerStats) metricValues
|
||||||
|
}
|
||||||
|
|
||||||
|
func (cm *containerMetric) desc() *prometheus.Desc {
|
||||||
|
return prometheus.NewDesc(cm.name, cm.help, append([]string{"name", "id"}, cm.extraLabels...), nil)
|
||||||
}
|
}
|
||||||
|
|
||||||
// PrometheusCollector implements prometheus.Collector.
|
// PrometheusCollector implements prometheus.Collector.
|
||||||
type PrometheusCollector struct {
|
type PrometheusCollector struct {
|
||||||
infoProvider subcontainersInfoProvider
|
infoProvider subcontainersInfoProvider
|
||||||
|
errors prometheus.Gauge
|
||||||
errors prometheus.Gauge
|
containerMetrics []containerMetric
|
||||||
lastSeen *prometheus.Desc
|
|
||||||
|
|
||||||
cpuUsageUserSeconds *prometheus.Desc
|
|
||||||
cpuUsageSystemSeconds *prometheus.Desc
|
|
||||||
cpuUsageSecondsPerCPU *prometheus.Desc
|
|
||||||
|
|
||||||
memoryUsageBytes *prometheus.Desc
|
|
||||||
memoryWorkingSet *prometheus.Desc
|
|
||||||
memoryFailures *prometheus.Desc
|
|
||||||
|
|
||||||
fsLimit *prometheus.Desc
|
|
||||||
fsUsage *prometheus.Desc
|
|
||||||
fsReads *prometheus.Desc
|
|
||||||
fsReadsSectors *prometheus.Desc
|
|
||||||
fsReadsMerged *prometheus.Desc
|
|
||||||
fsReadTime *prometheus.Desc
|
|
||||||
|
|
||||||
fsWrites *prometheus.Desc
|
|
||||||
fsWritesSectors *prometheus.Desc
|
|
||||||
fsWritesMerged *prometheus.Desc
|
|
||||||
fsWriteTime *prometheus.Desc
|
|
||||||
|
|
||||||
fsIoInProgress *prometheus.Desc
|
|
||||||
fsIoTime *prometheus.Desc
|
|
||||||
|
|
||||||
fsWeightedIoTime *prometheus.Desc
|
|
||||||
|
|
||||||
networkRxBytes *prometheus.Desc
|
|
||||||
networkRxPackets *prometheus.Desc
|
|
||||||
networkRxErrors *prometheus.Desc
|
|
||||||
networkRxDropped *prometheus.Desc
|
|
||||||
networkTxBytes *prometheus.Desc
|
|
||||||
networkTxPackets *prometheus.Desc
|
|
||||||
networkTxErrors *prometheus.Desc
|
|
||||||
networkTxDropped *prometheus.Desc
|
|
||||||
|
|
||||||
tasks *prometheus.Desc
|
|
||||||
|
|
||||||
descs []*prometheus.Desc
|
|
||||||
}
|
}
|
||||||
|
|
||||||
// NewPrometheusCollector returns a new PrometheusCollector.
|
// NewPrometheusCollector returns a new PrometheusCollector.
|
||||||
@ -89,188 +80,299 @@ func NewPrometheusCollector(infoProvider subcontainersInfoProvider) *PrometheusC
|
|||||||
Name: "scrape_error",
|
Name: "scrape_error",
|
||||||
Help: "1 if there was an error while getting container metrics, 0 otherwise",
|
Help: "1 if there was an error while getting container metrics, 0 otherwise",
|
||||||
}),
|
}),
|
||||||
lastSeen: prometheus.NewDesc(
|
containerMetrics: []containerMetric{
|
||||||
"container_last_seen",
|
{
|
||||||
"Last time a container was seen by the exporter",
|
name: "container_last_seen",
|
||||||
[]string{"name", "id"},
|
help: "Last time a container was seen by the exporter",
|
||||||
nil),
|
valueType: prometheus.GaugeValue,
|
||||||
cpuUsageUserSeconds: prometheus.NewDesc(
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
"container_cpu_user_seconds_total",
|
return metricValues{{value: float64(time.Now().Unix())}}
|
||||||
"Cumulative user cpu time consumed in seconds.",
|
},
|
||||||
[]string{"name", "id"},
|
}, {
|
||||||
nil),
|
name: "container_cpu_user_seconds_total",
|
||||||
cpuUsageSystemSeconds: prometheus.NewDesc(
|
help: "Cumulative user cpu time consumed in seconds.",
|
||||||
"container_cpu_system_seconds_total",
|
valueType: prometheus.CounterValue,
|
||||||
"Cumulative system cpu time consumed in seconds.",
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
[]string{"name", "id"},
|
return metricValues{{value: float64(s.Cpu.Usage.User) / float64(time.Second)}}
|
||||||
nil),
|
},
|
||||||
cpuUsageSecondsPerCPU: prometheus.NewDesc(
|
}, {
|
||||||
"container_cpu_usage_seconds_total",
|
name: "container_cpu_system_seconds_total",
|
||||||
"Cumulative cpu time consumed per cpu in seconds.",
|
help: "Cumulative system cpu time consumed in seconds.",
|
||||||
[]string{"name", "id", "cpu"},
|
valueType: prometheus.CounterValue,
|
||||||
nil),
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
memoryUsageBytes: prometheus.NewDesc(
|
return metricValues{{value: float64(s.Cpu.Usage.System) / float64(time.Second)}}
|
||||||
"container_memory_usage_bytes",
|
},
|
||||||
"Current memory usage in bytes.",
|
}, {
|
||||||
[]string{"name", "id"},
|
name: "container_cpu_usage_seconds_total",
|
||||||
nil),
|
help: "Cumulative cpu time consumed per cpu in seconds.",
|
||||||
memoryWorkingSet: prometheus.NewDesc(
|
valueType: prometheus.CounterValue,
|
||||||
"container_memory_working_set_bytes",
|
extraLabels: []string{"cpu"},
|
||||||
"Current working set in bytes.",
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
[]string{"name", "id"},
|
values := make(metricValues, 0, len(s.Cpu.Usage.PerCpu))
|
||||||
nil),
|
for i, value := range s.Cpu.Usage.PerCpu {
|
||||||
memoryFailures: prometheus.NewDesc(
|
values = append(values, metricValue{
|
||||||
"container_memory_failures_total",
|
value: float64(value) / float64(time.Second),
|
||||||
"Cumulative count of memory allocation failures.",
|
labels: []string{fmt.Sprintf("cpu%02d", i)},
|
||||||
[]string{"type", "scope", "name", "id"},
|
})
|
||||||
nil),
|
}
|
||||||
|
return values
|
||||||
fsLimit: prometheus.NewDesc(
|
},
|
||||||
"container_fs_limit_bytes",
|
}, {
|
||||||
"Number of bytes that can be consumed by the container on this filesystem.",
|
name: "container_memory_usage_bytes",
|
||||||
[]string{"name", "id", "device"},
|
help: "Current memory usage in bytes.",
|
||||||
nil),
|
valueType: prometheus.GaugeValue,
|
||||||
fsUsage: prometheus.NewDesc(
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
"container_fs_usage_bytes",
|
return metricValues{{value: float64(s.Memory.Usage)}}
|
||||||
"Number of bytes that are consumed by the container on this filesystem.",
|
},
|
||||||
[]string{"name", "id", "device"},
|
}, {
|
||||||
nil),
|
name: "container_memory_working_set_bytes",
|
||||||
fsReads: prometheus.NewDesc(
|
help: "Current working set in bytes.",
|
||||||
"container_fs_reads_total",
|
valueType: prometheus.GaugeValue,
|
||||||
"Cumulative count of reads completed",
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
[]string{"name", "id", "device"},
|
return metricValues{{value: float64(s.Memory.WorkingSet)}}
|
||||||
nil),
|
},
|
||||||
fsReadsSectors: prometheus.NewDesc(
|
}, {
|
||||||
"container_fs_sector_reads_total",
|
name: "container_memory_failures_total",
|
||||||
"Cumulative count of sector reads completed",
|
help: "Cumulative count of memory allocation failures.",
|
||||||
[]string{"name", "id", "device"},
|
valueType: prometheus.CounterValue,
|
||||||
nil),
|
extraLabels: []string{"type", "scope"},
|
||||||
fsReadsMerged: prometheus.NewDesc(
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
"container_fs_reads_merged_total",
|
return metricValues{
|
||||||
"Cumulative count of reads merged",
|
{
|
||||||
[]string{"name", "id", "device"},
|
value: float64(s.Memory.ContainerData.Pgfault),
|
||||||
nil),
|
labels: []string{"pgfault", "container"},
|
||||||
fsReadTime: prometheus.NewDesc(
|
},
|
||||||
"container_fs_read_seconds_total",
|
{
|
||||||
"Cumulative count of seconds spent reading",
|
value: float64(s.Memory.ContainerData.Pgmajfault),
|
||||||
[]string{"name", "id", "device"},
|
labels: []string{"pgmajfault", "container"},
|
||||||
nil),
|
},
|
||||||
fsWrites: prometheus.NewDesc(
|
{
|
||||||
"container_fs_writes_total",
|
value: float64(s.Memory.HierarchicalData.Pgfault),
|
||||||
"Cumulative count of writes completed",
|
labels: []string{"pgfault", "hierarchy"},
|
||||||
[]string{"name", "id", "device"},
|
},
|
||||||
nil),
|
{
|
||||||
fsWritesSectors: prometheus.NewDesc(
|
value: float64(s.Memory.HierarchicalData.Pgmajfault),
|
||||||
"container_fs_sector_writes_total",
|
labels: []string{"pgmajfault", "hierarchy"},
|
||||||
"Cumulative count of sector writes completed",
|
},
|
||||||
[]string{"name", "id", "device"},
|
}
|
||||||
nil),
|
},
|
||||||
fsWritesMerged: prometheus.NewDesc(
|
}, {
|
||||||
"container_fs_writes_merged_total",
|
name: "container_fs_limit_bytes",
|
||||||
"Cumulative count of writes merged",
|
help: "Number of bytes that can be consumed by the container on this filesystem.",
|
||||||
[]string{"name", "id", "device"},
|
valueType: prometheus.GaugeValue,
|
||||||
nil),
|
extraLabels: []string{"device"},
|
||||||
fsWriteTime: prometheus.NewDesc(
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
"container_fs_write_seconds_total",
|
return fsValues(s.Filesystem, func(fs *info.FsStats) float64 {
|
||||||
"Cumulative count of seconds spent writing",
|
return float64(fs.Limit)
|
||||||
[]string{"name", "id", "device"},
|
})
|
||||||
nil),
|
},
|
||||||
fsIoInProgress: prometheus.NewDesc(
|
}, {
|
||||||
"container_fs_io_current",
|
name: "container_fs_usage_bytes",
|
||||||
"Number of I/Os currently in progress",
|
help: "Number of bytes that are consumed by the container on this filesystem.",
|
||||||
[]string{"name", "id", "device"},
|
valueType: prometheus.GaugeValue,
|
||||||
nil),
|
extraLabels: []string{"device"},
|
||||||
fsIoTime: prometheus.NewDesc(
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
"container_fs_io_time_seconds_total",
|
return fsValues(s.Filesystem, func(fs *info.FsStats) float64 {
|
||||||
"Cumulative count of seconds spent doing I/Os",
|
return float64(fs.Usage)
|
||||||
[]string{"name", "id", "device"},
|
})
|
||||||
nil),
|
},
|
||||||
fsWeightedIoTime: prometheus.NewDesc(
|
}, {
|
||||||
"container_fs_io_time_weighted_seconds_total",
|
name: "container_fs_reads_total",
|
||||||
"Cumulative weighted I/O time in seconds",
|
help: "Cumulative count of reads completed",
|
||||||
[]string{"name", "id", "device"},
|
valueType: prometheus.CounterValue,
|
||||||
nil),
|
extraLabels: []string{"device"},
|
||||||
networkRxBytes: prometheus.NewDesc(
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
"container_network_receive_bytes_total",
|
return fsValues(s.Filesystem, func(fs *info.FsStats) float64 {
|
||||||
"Cumulative count of bytes received",
|
return float64(fs.ReadsCompleted)
|
||||||
[]string{"name", "id"},
|
})
|
||||||
nil),
|
},
|
||||||
networkRxPackets: prometheus.NewDesc(
|
}, {
|
||||||
"container_network_receive_packets_total",
|
name: "container_fs_sector_reads_total",
|
||||||
"Cumulative count of packets received",
|
help: "Cumulative count of sector reads completed",
|
||||||
[]string{"name", "id"},
|
valueType: prometheus.CounterValue,
|
||||||
nil),
|
extraLabels: []string{"device"},
|
||||||
networkRxDropped: prometheus.NewDesc(
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
"container_network_receive_packets_dropped_total",
|
return fsValues(s.Filesystem, func(fs *info.FsStats) float64 {
|
||||||
"Cumulative count of packets dropped while receiving",
|
return float64(fs.SectorsRead)
|
||||||
[]string{"name", "id"},
|
})
|
||||||
nil),
|
},
|
||||||
networkRxErrors: prometheus.NewDesc(
|
}, {
|
||||||
"container_network_receive_errors_total",
|
name: "container_fs_reads_merged_total",
|
||||||
"Cumulative count of errors encountered while receiving",
|
help: "Cumulative count of reads merged",
|
||||||
[]string{"name", "id"},
|
valueType: prometheus.CounterValue,
|
||||||
nil),
|
extraLabels: []string{"device"},
|
||||||
networkTxBytes: prometheus.NewDesc(
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
"container_network_transmit_bytes_total",
|
return fsValues(s.Filesystem, func(fs *info.FsStats) float64 {
|
||||||
"Cumulative count of bytes transmitted",
|
return float64(fs.ReadsMerged)
|
||||||
[]string{"name", "id"},
|
})
|
||||||
nil),
|
},
|
||||||
networkTxPackets: prometheus.NewDesc(
|
}, {
|
||||||
"container_network_transmit_packets_total",
|
name: "container_fs_read_seconds_total",
|
||||||
"Cumulative count of packets transmitted",
|
help: "Cumulative count of seconds spent reading",
|
||||||
[]string{"name", "id"},
|
valueType: prometheus.CounterValue,
|
||||||
nil),
|
extraLabels: []string{"device"},
|
||||||
networkTxDropped: prometheus.NewDesc(
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
"container_network_transmit_packets_dropped_total",
|
return fsValues(s.Filesystem, func(fs *info.FsStats) float64 {
|
||||||
"Cumulative count of packets dropped while transmitting",
|
return float64(fs.ReadTime) / float64(time.Second)
|
||||||
[]string{"name", "id"},
|
})
|
||||||
nil),
|
},
|
||||||
networkTxErrors: prometheus.NewDesc(
|
}, {
|
||||||
"container_network_transmit_errors_total",
|
name: "container_fs_writes_total",
|
||||||
"Cumulative count of errors encountered while transmitting",
|
help: "Cumulative count of writes completed",
|
||||||
[]string{"name", "id"},
|
valueType: prometheus.CounterValue,
|
||||||
nil),
|
extraLabels: []string{"device"},
|
||||||
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
tasks: prometheus.NewDesc(
|
return fsValues(s.Filesystem, func(fs *info.FsStats) float64 {
|
||||||
"container_tasks_state",
|
return float64(fs.WritesCompleted)
|
||||||
"Number of tasks in given state",
|
})
|
||||||
[]string{"state", "name", "id"},
|
},
|
||||||
nil),
|
}, {
|
||||||
}
|
name: "container_fs_sector_writes_total",
|
||||||
c.descs = []*prometheus.Desc{
|
help: "Cumulative count of sector writes completed",
|
||||||
c.lastSeen,
|
valueType: prometheus.CounterValue,
|
||||||
|
extraLabels: []string{"device"},
|
||||||
c.cpuUsageUserSeconds,
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
c.cpuUsageSystemSeconds,
|
return fsValues(s.Filesystem, func(fs *info.FsStats) float64 {
|
||||||
|
return float64(fs.SectorsWritten)
|
||||||
c.memoryUsageBytes,
|
})
|
||||||
c.memoryWorkingSet,
|
},
|
||||||
c.memoryFailures,
|
}, {
|
||||||
|
name: "container_fs_writes_merged_total",
|
||||||
c.fsLimit,
|
help: "Cumulative count of writes merged",
|
||||||
c.fsUsage,
|
valueType: prometheus.CounterValue,
|
||||||
c.fsReads,
|
extraLabels: []string{"device"},
|
||||||
c.fsReadsSectors,
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
c.fsReadsMerged,
|
return fsValues(s.Filesystem, func(fs *info.FsStats) float64 {
|
||||||
c.fsReadTime,
|
return float64(fs.WritesMerged)
|
||||||
c.fsWrites,
|
})
|
||||||
c.fsWritesSectors,
|
},
|
||||||
c.fsWritesMerged,
|
}, {
|
||||||
c.fsWriteTime,
|
name: "container_fs_write_seconds_total",
|
||||||
c.fsIoInProgress,
|
help: "Cumulative count of seconds spent writing",
|
||||||
c.fsIoTime,
|
valueType: prometheus.CounterValue,
|
||||||
c.fsWeightedIoTime,
|
extraLabels: []string{"device"},
|
||||||
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
c.networkRxBytes,
|
return fsValues(s.Filesystem, func(fs *info.FsStats) float64 {
|
||||||
c.networkRxPackets,
|
return float64(fs.WriteTime) / float64(time.Second)
|
||||||
c.networkRxErrors,
|
})
|
||||||
c.networkRxDropped,
|
},
|
||||||
c.networkTxBytes,
|
}, {
|
||||||
c.networkTxPackets,
|
name: "container_fs_io_current",
|
||||||
c.networkTxErrors,
|
help: "Number of I/Os currently in progress",
|
||||||
c.networkTxDropped,
|
valueType: prometheus.GaugeValue,
|
||||||
|
extraLabels: []string{"device"},
|
||||||
c.tasks,
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
|
return fsValues(s.Filesystem, func(fs *info.FsStats) float64 {
|
||||||
|
return float64(fs.IoInProgress)
|
||||||
|
})
|
||||||
|
},
|
||||||
|
}, {
|
||||||
|
name: "container_fs_io_time_seconds_total",
|
||||||
|
help: "Cumulative count of seconds spent doing I/Os",
|
||||||
|
valueType: prometheus.CounterValue,
|
||||||
|
extraLabels: []string{"device"},
|
||||||
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
|
return fsValues(s.Filesystem, func(fs *info.FsStats) float64 {
|
||||||
|
return float64(float64(fs.IoTime) / float64(time.Second))
|
||||||
|
})
|
||||||
|
},
|
||||||
|
}, {
|
||||||
|
name: "container_fs_io_time_weighted_seconds_total",
|
||||||
|
help: "Cumulative weighted I/O time in seconds",
|
||||||
|
valueType: prometheus.CounterValue,
|
||||||
|
extraLabels: []string{"device"},
|
||||||
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
|
return fsValues(s.Filesystem, func(fs *info.FsStats) float64 {
|
||||||
|
return float64(fs.WeightedIoTime) / float64(time.Second)
|
||||||
|
})
|
||||||
|
},
|
||||||
|
}, {
|
||||||
|
name: "container_network_receive_bytes_total",
|
||||||
|
help: "Cumulative count of bytes received",
|
||||||
|
valueType: prometheus.CounterValue,
|
||||||
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
|
return metricValues{{value: float64(s.Network.RxBytes)}}
|
||||||
|
},
|
||||||
|
}, {
|
||||||
|
name: "container_network_receive_packets_total",
|
||||||
|
help: "Cumulative count of packets received",
|
||||||
|
valueType: prometheus.CounterValue,
|
||||||
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
|
return metricValues{{value: float64(s.Network.RxPackets)}}
|
||||||
|
},
|
||||||
|
}, {
|
||||||
|
name: "container_network_receive_packets_dropped_total",
|
||||||
|
help: "Cumulative count of packets dropped while receiving",
|
||||||
|
valueType: prometheus.CounterValue,
|
||||||
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
|
return metricValues{{value: float64(s.Network.RxDropped)}}
|
||||||
|
},
|
||||||
|
}, {
|
||||||
|
name: "container_network_receive_errors_total",
|
||||||
|
help: "Cumulative count of errors encountered while receiving",
|
||||||
|
valueType: prometheus.CounterValue,
|
||||||
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
|
return metricValues{{value: float64(s.Network.RxErrors)}}
|
||||||
|
},
|
||||||
|
}, {
|
||||||
|
name: "container_network_transmit_bytes_total",
|
||||||
|
help: "Cumulative count of bytes transmitted",
|
||||||
|
valueType: prometheus.CounterValue,
|
||||||
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
|
return metricValues{{value: float64(s.Network.TxBytes)}}
|
||||||
|
},
|
||||||
|
}, {
|
||||||
|
name: "container_network_transmit_packets_total",
|
||||||
|
help: "Cumulative count of packets transmitted",
|
||||||
|
valueType: prometheus.CounterValue,
|
||||||
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
|
return metricValues{{value: float64(s.Network.TxPackets)}}
|
||||||
|
},
|
||||||
|
}, {
|
||||||
|
name: "container_network_transmit_packets_dropped_total",
|
||||||
|
help: "Cumulative count of packets dropped while transmitting",
|
||||||
|
valueType: prometheus.CounterValue,
|
||||||
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
|
return metricValues{{value: float64(s.Network.TxDropped)}}
|
||||||
|
},
|
||||||
|
}, {
|
||||||
|
name: "container_network_transmit_errors_total",
|
||||||
|
help: "Cumulative count of errors encountered while transmitting",
|
||||||
|
valueType: prometheus.CounterValue,
|
||||||
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
|
return metricValues{{value: float64(s.Network.TxErrors)}}
|
||||||
|
},
|
||||||
|
}, {
|
||||||
|
name: "container_tasks_state",
|
||||||
|
help: "Number of tasks in given state",
|
||||||
|
extraLabels: []string{"state"},
|
||||||
|
valueType: prometheus.GaugeValue,
|
||||||
|
getValues: func(s *info.ContainerStats) metricValues {
|
||||||
|
return metricValues{
|
||||||
|
{
|
||||||
|
value: float64(s.TaskStats.NrSleeping),
|
||||||
|
labels: []string{"sleeping"},
|
||||||
|
},
|
||||||
|
{
|
||||||
|
value: float64(s.TaskStats.NrRunning),
|
||||||
|
labels: []string{"running"},
|
||||||
|
},
|
||||||
|
{
|
||||||
|
value: float64(s.TaskStats.NrStopped),
|
||||||
|
labels: []string{"stopped"},
|
||||||
|
},
|
||||||
|
{
|
||||||
|
value: float64(s.TaskStats.NrUninterruptible),
|
||||||
|
labels: []string{"uninterruptible"},
|
||||||
|
},
|
||||||
|
{
|
||||||
|
value: float64(s.TaskStats.NrIoWait),
|
||||||
|
labels: []string{"iowaiting"},
|
||||||
|
},
|
||||||
|
}
|
||||||
|
},
|
||||||
|
},
|
||||||
|
},
|
||||||
}
|
}
|
||||||
return c
|
return c
|
||||||
}
|
}
|
||||||
@ -279,8 +381,8 @@ func NewPrometheusCollector(infoProvider subcontainersInfoProvider) *PrometheusC
|
|||||||
// implements prometheus.PrometheusCollector.
|
// implements prometheus.PrometheusCollector.
|
||||||
func (c *PrometheusCollector) Describe(ch chan<- *prometheus.Desc) {
|
func (c *PrometheusCollector) Describe(ch chan<- *prometheus.Desc) {
|
||||||
c.errors.Describe(ch)
|
c.errors.Describe(ch)
|
||||||
for _, d := range c.descs {
|
for _, cm := range c.containerMetrics {
|
||||||
ch <- d
|
ch <- cm.desc()
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
@ -301,68 +403,10 @@ func (c *PrometheusCollector) Collect(ch chan<- prometheus.Metric) {
|
|||||||
}
|
}
|
||||||
stats := container.Stats[0]
|
stats := container.Stats[0]
|
||||||
|
|
||||||
for desc, metrics := range map[*prometheus.Desc][]prometheusMetric{
|
for _, cm := range c.containerMetrics {
|
||||||
c.cpuUsageUserSeconds: {{valueType: prometheus.CounterValue, value: float64(stats.Cpu.Usage.User) / float64(time.Second)}},
|
desc := cm.desc()
|
||||||
c.cpuUsageSystemSeconds: {{valueType: prometheus.CounterValue, value: float64(stats.Cpu.Usage.System) / float64(time.Second)}},
|
for _, metricValue := range cm.getValues(stats) {
|
||||||
|
ch <- prometheus.MustNewConstMetric(desc, cm.valueType, float64(metricValue.value), append([]string{name, id}, metricValue.labels...)...)
|
||||||
c.memoryFailures: {
|
|
||||||
{valueType: prometheus.CounterValue, labels: []string{"pgfault", "container"}, value: float64(stats.Memory.ContainerData.Pgfault)},
|
|
||||||
{valueType: prometheus.CounterValue, labels: []string{"pgmajfault", "container"}, value: float64(stats.Memory.ContainerData.Pgmajfault)},
|
|
||||||
{valueType: prometheus.CounterValue, labels: []string{"pgfault", "hierarchy"}, value: float64(stats.Memory.HierarchicalData.Pgfault)},
|
|
||||||
{valueType: prometheus.CounterValue, labels: []string{"pgmajfault", "hierarchy"}, value: float64(stats.Memory.HierarchicalData.Pgmajfault)},
|
|
||||||
},
|
|
||||||
c.tasks: {
|
|
||||||
{valueType: prometheus.GaugeValue, labels: []string{"sleeping"}, value: float64(stats.TaskStats.NrSleeping)},
|
|
||||||
{valueType: prometheus.GaugeValue, labels: []string{"running"}, value: float64(stats.TaskStats.NrRunning)},
|
|
||||||
{valueType: prometheus.GaugeValue, labels: []string{"stopped"}, value: float64(stats.TaskStats.NrStopped)},
|
|
||||||
{valueType: prometheus.GaugeValue, labels: []string{"uninterruptible"}, value: float64(stats.TaskStats.NrUninterruptible)},
|
|
||||||
{valueType: prometheus.GaugeValue, labels: []string{"iowaiting"}, value: float64(stats.TaskStats.NrIoWait)},
|
|
||||||
},
|
|
||||||
|
|
||||||
c.lastSeen: {{valueType: prometheus.GaugeValue, value: float64(time.Now().Unix())}},
|
|
||||||
|
|
||||||
c.memoryUsageBytes: {{valueType: prometheus.GaugeValue, value: float64(stats.Memory.Usage)}},
|
|
||||||
c.memoryWorkingSet: {{valueType: prometheus.GaugeValue, value: float64(stats.Memory.WorkingSet)}},
|
|
||||||
|
|
||||||
c.networkRxBytes: {{valueType: prometheus.CounterValue, value: float64(stats.Network.RxBytes)}},
|
|
||||||
c.networkRxPackets: {{valueType: prometheus.CounterValue, value: float64(stats.Network.RxPackets)}},
|
|
||||||
c.networkRxErrors: {{valueType: prometheus.CounterValue, value: float64(stats.Network.RxErrors)}},
|
|
||||||
c.networkRxDropped: {{valueType: prometheus.CounterValue, value: float64(stats.Network.RxDropped)}},
|
|
||||||
c.networkTxBytes: {{valueType: prometheus.CounterValue, value: float64(stats.Network.TxBytes)}},
|
|
||||||
c.networkTxPackets: {{valueType: prometheus.CounterValue, value: float64(stats.Network.TxPackets)}},
|
|
||||||
c.networkTxErrors: {{valueType: prometheus.CounterValue, value: float64(stats.Network.TxErrors)}},
|
|
||||||
c.networkTxDropped: {{valueType: prometheus.CounterValue, value: float64(stats.Network.TxDropped)}},
|
|
||||||
} {
|
|
||||||
for _, m := range metrics {
|
|
||||||
ch <- prometheus.MustNewConstMetric(desc, m.valueType, float64(m.value), append(m.labels, name, id)...)
|
|
||||||
}
|
|
||||||
}
|
|
||||||
|
|
||||||
// Metrics with dynamic labels
|
|
||||||
for i, value := range stats.Cpu.Usage.PerCpu {
|
|
||||||
ch <- prometheus.MustNewConstMetric(c.cpuUsageSecondsPerCPU, prometheus.CounterValue, float64(value)/float64(time.Second), name, id, fmt.Sprintf("cpu%02d", i))
|
|
||||||
}
|
|
||||||
|
|
||||||
for _, stat := range stats.Filesystem {
|
|
||||||
for desc, m := range map[*prometheus.Desc]prometheusMetric{
|
|
||||||
c.fsReads: {valueType: prometheus.CounterValue, value: float64(stat.ReadsCompleted)},
|
|
||||||
c.fsReadsSectors: {valueType: prometheus.CounterValue, value: float64(stat.SectorsRead)},
|
|
||||||
c.fsReadsMerged: {valueType: prometheus.CounterValue, value: float64(stat.ReadsMerged)},
|
|
||||||
c.fsReadTime: {valueType: prometheus.CounterValue, value: float64(stat.ReadTime) / float64(time.Second)},
|
|
||||||
|
|
||||||
c.fsWrites: {valueType: prometheus.CounterValue, value: float64(stat.WritesCompleted)},
|
|
||||||
c.fsWritesSectors: {valueType: prometheus.CounterValue, value: float64(stat.SectorsWritten)},
|
|
||||||
c.fsWritesMerged: {valueType: prometheus.CounterValue, value: float64(stat.WritesMerged)},
|
|
||||||
c.fsWriteTime: {valueType: prometheus.CounterValue, value: float64(stat.WriteTime) / float64(time.Second)},
|
|
||||||
|
|
||||||
c.fsIoTime: {valueType: prometheus.CounterValue, value: float64(stat.IoTime) / float64(time.Second)},
|
|
||||||
c.fsWeightedIoTime: {valueType: prometheus.CounterValue, value: float64(stat.WeightedIoTime) / float64(time.Second)},
|
|
||||||
|
|
||||||
c.fsIoInProgress: {valueType: prometheus.GaugeValue, value: float64(stat.IoInProgress)},
|
|
||||||
c.fsLimit: {valueType: prometheus.GaugeValue, value: float64(stat.Limit)},
|
|
||||||
c.fsUsage: {valueType: prometheus.GaugeValue, value: float64(stat.Usage)},
|
|
||||||
} {
|
|
||||||
ch <- prometheus.MustNewConstMetric(desc, m.valueType, m.value, name, id, stat.Device)
|
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|||||||
Loading…
Reference in New Issue
Block a user