metrics: rename pids_v2 to pids

dicussed in #3726

Signed-off-by: Akihiro Suda <akihiro.suda.cz@hco.ntt.co.jp>

metrics/cgroups/v1/blkio.go

@@ -0,0 +1,132 @@
+// +build linux
+
+/*
+   Copyright The containerd Authors.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
+*/
+
+package v1
+
+import (
+	"strconv"
+
+	v1 "github.com/containerd/containerd/metrics/types/v1"
+	metrics "github.com/docker/go-metrics"
+	"github.com/prometheus/client_golang/prometheus"
+)
+
+var blkioMetrics = []*metric{
+	{
+		name:   "blkio_io_merged_recursive",
+		help:   "The blkio io merged recursive",
+		unit:   metrics.Total,
+		vt:     prometheus.GaugeValue,
+		labels: []string{"op", "device", "major", "minor"},
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Blkio == nil {
+				return nil
+			}
+			return blkioValues(stats.Blkio.IoMergedRecursive)
+		},
+	},
+	{
+		name:   "blkio_io_queued_recursive",
+		help:   "The blkio io queued recursive",
+		unit:   metrics.Total,
+		vt:     prometheus.GaugeValue,
+		labels: []string{"op", "device", "major", "minor"},
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Blkio == nil {
+				return nil
+			}
+			return blkioValues(stats.Blkio.IoQueuedRecursive)
+		},
+	},
+	{
+		name:   "blkio_io_service_bytes_recursive",
+		help:   "The blkio io service bytes recursive",
+		unit:   metrics.Bytes,
+		vt:     prometheus.GaugeValue,
+		labels: []string{"op", "device", "major", "minor"},
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Blkio == nil {
+				return nil
+			}
+			return blkioValues(stats.Blkio.IoServiceBytesRecursive)
+		},
+	},
+	{
+		name:   "blkio_io_service_time_recursive",
+		help:   "The blkio io service time recursive",
+		unit:   metrics.Total,
+		vt:     prometheus.GaugeValue,
+		labels: []string{"op", "device", "major", "minor"},
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Blkio == nil {
+				return nil
+			}
+			return blkioValues(stats.Blkio.IoServiceTimeRecursive)
+		},
+	},
+	{
+		name:   "blkio_io_serviced_recursive",
+		help:   "The blkio io serviced recursive",
+		unit:   metrics.Total,
+		vt:     prometheus.GaugeValue,
+		labels: []string{"op", "device", "major", "minor"},
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Blkio == nil {
+				return nil
+			}
+			return blkioValues(stats.Blkio.IoServicedRecursive)
+		},
+	},
+	{
+		name:   "blkio_io_time_recursive",
+		help:   "The blkio io time recursive",
+		unit:   metrics.Total,
+		vt:     prometheus.GaugeValue,
+		labels: []string{"op", "device", "major", "minor"},
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Blkio == nil {
+				return nil
+			}
+			return blkioValues(stats.Blkio.IoTimeRecursive)
+		},
+	},
+	{
+		name:   "blkio_sectors_recursive",
+		help:   "The blkio sectors recursive",
+		unit:   metrics.Total,
+		vt:     prometheus.GaugeValue,
+		labels: []string{"op", "device", "major", "minor"},
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Blkio == nil {
+				return nil
+			}
+			return blkioValues(stats.Blkio.SectorsRecursive)
+		},
+	},
+}
+
+func blkioValues(l []*v1.BlkIOEntry) []value {
+	var out []value
+	for _, e := range l {
+		out = append(out, value{
+			v: float64(e.Value),
+			l: []string{e.Op, e.Device, strconv.FormatUint(e.Major, 10), strconv.FormatUint(e.Minor, 10)},
+		})
+	}
+	return out
+}

metrics/cgroups/v1/cgroups.go

@@ -0,0 +1,93 @@
+// +build linux
+
+/*
+   Copyright The containerd Authors.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
+*/
+
+package v1
+
+import (
+	"context"
+
+	"github.com/containerd/cgroups"
+	eventstypes "github.com/containerd/containerd/api/events"
+	"github.com/containerd/containerd/errdefs"
+	"github.com/containerd/containerd/events"
+	"github.com/containerd/containerd/log"
+	"github.com/containerd/containerd/namespaces"
+	"github.com/containerd/containerd/runtime"
+	"github.com/containerd/containerd/runtime/v1/linux"
+	metrics "github.com/docker/go-metrics"
+	"github.com/sirupsen/logrus"
+)
+
+// NewTaskMonitor returns a new cgroups monitor
+func NewTaskMonitor(ctx context.Context, publisher events.Publisher, ns *metrics.Namespace) (runtime.TaskMonitor, error) {
+	collector := newCollector(ns)
+	oom, err := newOOMCollector(ns)
+	if err != nil {
+		return nil, err
+	}
+	return &cgroupsMonitor{
+		collector: collector,
+		oom:       oom,
+		context:   ctx,
+		publisher: publisher,
+	}, nil
+}
+
+type cgroupsMonitor struct {
+	collector *collector
+	oom       *oomCollector
+	context   context.Context
+	publisher events.Publisher
+}
+
+func (m *cgroupsMonitor) Monitor(c runtime.Task) error {
+	if err := m.collector.Add(c); err != nil {
+		return err
+	}
+	t, ok := c.(*linux.Task)
+	if !ok {
+		return nil
+	}
+	cg, err := t.Cgroup()
+	if err != nil {
+		if errdefs.IsNotFound(err) {
+			return nil
+		}
+		return err
+	}
+	err = m.oom.Add(c.ID(), c.Namespace(), cg, m.trigger)
+	if err == cgroups.ErrMemoryNotSupported {
+		logrus.WithError(err).Warn("OOM monitoring failed")
+		return nil
+	}
+	return err
+}
+
+func (m *cgroupsMonitor) Stop(c runtime.Task) error {
+	m.collector.Remove(c)
+	return nil
+}
+
+func (m *cgroupsMonitor) trigger(id, namespace string, cg cgroups.Cgroup) {
+	ctx := namespaces.WithNamespace(m.context, namespace)
+	if err := m.publisher.Publish(ctx, runtime.TaskOOMEventTopic, &eventstypes.TaskOOM{
+		ContainerID: id,
+	}); err != nil {
+		log.G(m.context).WithError(err).Error("post OOM event")
+	}
+}

metrics/cgroups/v1/cpu.go

@@ -0,0 +1,146 @@
+// +build linux
+
+/*
+   Copyright The containerd Authors.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
+*/
+
+package v1
+
+import (
+	"strconv"
+
+	v1 "github.com/containerd/containerd/metrics/types/v1"
+	metrics "github.com/docker/go-metrics"
+	"github.com/prometheus/client_golang/prometheus"
+)
+
+var cpuMetrics = []*metric{
+	{
+		name: "cpu_total",
+		help: "The total cpu time",
+		unit: metrics.Nanoseconds,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.CPU == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.CPU.Usage.Total),
+				},
+			}
+		},
+	},
+	{
+		name: "cpu_kernel",
+		help: "The total kernel cpu time",
+		unit: metrics.Nanoseconds,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.CPU == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.CPU.Usage.Kernel),
+				},
+			}
+		},
+	},
+	{
+		name: "cpu_user",
+		help: "The total user cpu time",
+		unit: metrics.Nanoseconds,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.CPU == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.CPU.Usage.User),
+				},
+			}
+		},
+	},
+	{
+		name:   "per_cpu",
+		help:   "The total cpu time per cpu",
+		unit:   metrics.Nanoseconds,
+		vt:     prometheus.GaugeValue,
+		labels: []string{"cpu"},
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.CPU == nil {
+				return nil
+			}
+			var out []value
+			for i, v := range stats.CPU.Usage.PerCPU {
+				out = append(out, value{
+					v: float64(v),
+					l: []string{strconv.Itoa(i)},
+				})
+			}
+			return out
+		},
+	},
+	{
+		name: "cpu_throttle_periods",
+		help: "The total cpu throttle periods",
+		unit: metrics.Total,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.CPU == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.CPU.Throttling.Periods),
+				},
+			}
+		},
+	},
+	{
+		name: "cpu_throttled_periods",
+		help: "The total cpu throttled periods",
+		unit: metrics.Total,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.CPU == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.CPU.Throttling.ThrottledPeriods),
+				},
+			}
+		},
+	},
+	{
+		name: "cpu_throttled_time",
+		help: "The total cpu throttled time",
+		unit: metrics.Nanoseconds,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.CPU == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.CPU.Throttling.ThrottledTime),
+				},
+			}
+		},
+	},
+}

metrics/cgroups/v1/hugetlb.go

@@ -0,0 +1,88 @@
+// +build linux
+
+/*
+   Copyright The containerd Authors.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
+*/
+
+package v1
+
+import (
+	v1 "github.com/containerd/containerd/metrics/types/v1"
+	metrics "github.com/docker/go-metrics"
+	"github.com/prometheus/client_golang/prometheus"
+)
+
+var hugetlbMetrics = []*metric{
+	{
+		name:   "hugetlb_usage",
+		help:   "The hugetlb usage",
+		unit:   metrics.Bytes,
+		vt:     prometheus.GaugeValue,
+		labels: []string{"page"},
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Hugetlb == nil {
+				return nil
+			}
+			var out []value
+			for _, v := range stats.Hugetlb {
+				out = append(out, value{
+					v: float64(v.Usage),
+					l: []string{v.Pagesize},
+				})
+			}
+			return out
+		},
+	},
+	{
+		name:   "hugetlb_failcnt",
+		help:   "The hugetlb failcnt",
+		unit:   metrics.Total,
+		vt:     prometheus.GaugeValue,
+		labels: []string{"page"},
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Hugetlb == nil {
+				return nil
+			}
+			var out []value
+			for _, v := range stats.Hugetlb {
+				out = append(out, value{
+					v: float64(v.Failcnt),
+					l: []string{v.Pagesize},
+				})
+			}
+			return out
+		},
+	},
+	{
+		name:   "hugetlb_max",
+		help:   "The hugetlb maximum usage",
+		unit:   metrics.Bytes,
+		vt:     prometheus.GaugeValue,
+		labels: []string{"page"},
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Hugetlb == nil {
+				return nil
+			}
+			var out []value
+			for _, v := range stats.Hugetlb {
+				out = append(out, value{
+					v: float64(v.Max),
+					l: []string{v.Pagesize},
+				})
+			}
+			return out
+		},
+	},
+}

metrics/cgroups/v1/memory.go

@@ -0,0 +1,796 @@
+// +build linux
+
+/*
+   Copyright The containerd Authors.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
+*/
+
+package v1
+
+import (
+	v1 "github.com/containerd/containerd/metrics/types/v1"
+	metrics "github.com/docker/go-metrics"
+	"github.com/prometheus/client_golang/prometheus"
+)
+
+var memoryMetrics = []*metric{
+	{
+		name: "memory_cache",
+		help: "The cache amount used",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.Cache),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_rss",
+		help: "The rss amount used",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.RSS),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_rss_huge",
+		help: "The rss_huge amount used",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.RSSHuge),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_mapped_file",
+		help: "The mapped_file amount used",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.MappedFile),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_dirty",
+		help: "The dirty amount",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.Dirty),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_writeback",
+		help: "The writeback amount",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.Writeback),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_pgpgin",
+		help: "The pgpgin amount",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.PgPgIn),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_pgpgout",
+		help: "The pgpgout amount",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.PgPgOut),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_pgfault",
+		help: "The pgfault amount",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.PgFault),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_pgmajfault",
+		help: "The pgmajfault amount",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.PgMajFault),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_inactive_anon",
+		help: "The inactive_anon amount",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.InactiveAnon),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_active_anon",
+		help: "The active_anon amount",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.ActiveAnon),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_inactive_file",
+		help: "The inactive_file amount",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.InactiveFile),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_active_file",
+		help: "The active_file amount",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.ActiveFile),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_unevictable",
+		help: "The unevictable amount",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.Unevictable),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_hierarchical_memory_limit",
+		help: "The hierarchical_memory_limit amount",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.HierarchicalMemoryLimit),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_hierarchical_memsw_limit",
+		help: "The hierarchical_memsw_limit amount",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.HierarchicalSwapLimit),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_total_cache",
+		help: "The total_cache amount used",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.TotalCache),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_total_rss",
+		help: "The total_rss amount used",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.TotalRSS),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_total_rss_huge",
+		help: "The total_rss_huge amount used",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.TotalRSSHuge),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_total_mapped_file",
+		help: "The total_mapped_file amount used",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.TotalMappedFile),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_total_dirty",
+		help: "The total_dirty amount",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.TotalDirty),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_total_writeback",
+		help: "The total_writeback amount",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.TotalWriteback),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_total_pgpgin",
+		help: "The total_pgpgin amount",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.TotalPgPgIn),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_total_pgpgout",
+		help: "The total_pgpgout amount",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.TotalPgPgOut),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_total_pgfault",
+		help: "The total_pgfault amount",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.TotalPgFault),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_total_pgmajfault",
+		help: "The total_pgmajfault amount",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.TotalPgMajFault),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_total_inactive_anon",
+		help: "The total_inactive_anon amount",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.TotalInactiveAnon),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_total_active_anon",
+		help: "The total_active_anon amount",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.TotalActiveAnon),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_total_inactive_file",
+		help: "The total_inactive_file amount",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.TotalInactiveFile),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_total_active_file",
+		help: "The total_active_file amount",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.TotalActiveFile),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_total_unevictable",
+		help: "The total_unevictable amount",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.TotalUnevictable),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_usage_failcnt",
+		help: "The usage failcnt",
+		unit: metrics.Total,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.Usage.Failcnt),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_usage_limit",
+		help: "The memory limit",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.Usage.Limit),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_usage_max",
+		help: "The memory maximum usage",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.Usage.Max),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_usage_usage",
+		help: "The memory usage",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.Usage.Usage),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_swap_failcnt",
+		help: "The swap failcnt",
+		unit: metrics.Total,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.Swap.Failcnt),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_swap_limit",
+		help: "The swap limit",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.Swap.Limit),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_swap_max",
+		help: "The swap maximum usage",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.Swap.Max),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_swap_usage",
+		help: "The swap usage",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.Swap.Usage),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_kernel_failcnt",
+		help: "The kernel failcnt",
+		unit: metrics.Total,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.Kernel.Failcnt),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_kernel_limit",
+		help: "The kernel limit",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.Kernel.Limit),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_kernel_max",
+		help: "The kernel maximum usage",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.Kernel.Max),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_kernel_usage",
+		help: "The kernel usage",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.Kernel.Usage),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_kerneltcp_failcnt",
+		help: "The kerneltcp failcnt",
+		unit: metrics.Total,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.KernelTCP.Failcnt),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_kerneltcp_limit",
+		help: "The kerneltcp limit",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.KernelTCP.Limit),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_kerneltcp_max",
+		help: "The kerneltcp maximum usage",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.KernelTCP.Max),
+				},
+			}
+		},
+	},
+	{
+		name: "memory_kerneltcp_usage",
+		help: "The kerneltcp usage",
+		unit: metrics.Bytes,
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Memory == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Memory.KernelTCP.Usage),
+				},
+			}
+		},
+	},
+}

metrics/cgroups/v1/metric.go

@@ -0,0 +1,61 @@
+// +build linux
+
+/*
+   Copyright The containerd Authors.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
+*/
+
+package v1
+
+import (
+	v1 "github.com/containerd/containerd/metrics/types/v1"
+	metrics "github.com/docker/go-metrics"
+	"github.com/prometheus/client_golang/prometheus"
+)
+
+type value struct {
+	v float64
+	l []string
+}
+
+type metric struct {
+	name   string
+	help   string
+	unit   metrics.Unit
+	vt     prometheus.ValueType
+	labels []string
+	// getValues returns the value and labels for the data
+	getValues func(stats *v1.Metrics) []value
+}
+
+func (m *metric) desc(ns *metrics.Namespace) *prometheus.Desc {
+	// the namespace label is for containerd namespaces
+	return ns.NewDesc(m.name, m.help, m.unit, append([]string{"container_id", "namespace"}, m.labels...)...)
+}
+
+func (m *metric) collect(id, namespace string, stats *v1.Metrics, ns *metrics.Namespace, ch chan<- prometheus.Metric, block bool) {
+	values := m.getValues(stats)
+	for _, v := range values {
+		// block signals to block on the sending the metrics so none are missed
+		if block {
+			ch <- prometheus.MustNewConstMetric(m.desc(ns), m.vt, v.v, append([]string{id, namespace}, v.l...)...)
+			continue
+		}
+		// non-blocking metrics can be dropped if the chan is full
+		select {
+		case ch <- prometheus.MustNewConstMetric(m.desc(ns), m.vt, v.v, append([]string{id, namespace}, v.l...)...):
+		default:
+		}
+	}
+}

metrics/cgroups/v1/metrics.go

@@ -0,0 +1,151 @@
+// +build linux
+
+/*
+   Copyright The containerd Authors.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
+*/
+
+package v1
+
+import (
+	"context"
+	"fmt"
+	"sync"
+
+	"github.com/containerd/cgroups"
+	"github.com/containerd/containerd/log"
+	v1 "github.com/containerd/containerd/metrics/types/v1"
+	"github.com/containerd/containerd/namespaces"
+	"github.com/containerd/containerd/runtime"
+	"github.com/containerd/typeurl"
+	metrics "github.com/docker/go-metrics"
+	"github.com/prometheus/client_golang/prometheus"
+)
+
+// Trigger will be called when an event happens and provides the cgroup
+// where the event originated from
+type Trigger func(string, string, cgroups.Cgroup)
+
+// newCollector registers the collector with the provided namespace and returns it so
+// that cgroups can be added for collection
+func newCollector(ns *metrics.Namespace) *collector {
+	if ns == nil {
+		return &collector{}
+	}
+	// add machine cpus and memory info
+	c := &collector{
+		ns:    ns,
+		tasks: make(map[string]runtime.Task),
+	}
+	c.metrics = append(c.metrics, pidMetrics...)
+	c.metrics = append(c.metrics, cpuMetrics...)
+	c.metrics = append(c.metrics, memoryMetrics...)
+	c.metrics = append(c.metrics, hugetlbMetrics...)
+	c.metrics = append(c.metrics, blkioMetrics...)
+	c.storedMetrics = make(chan prometheus.Metric, 100*len(c.metrics))
+	ns.Add(c)
+	return c
+}
+
+func taskID(id, namespace string) string {
+	return fmt.Sprintf("%s-%s", id, namespace)
+}
+
+// collector provides the ability to collect container stats and export
+// them in the prometheus format
+type collector struct {
+	mu sync.RWMutex
+
+	tasks         map[string]runtime.Task
+	ns            *metrics.Namespace
+	metrics       []*metric
+	storedMetrics chan prometheus.Metric
+}
+
+func (c *collector) Describe(ch chan<- *prometheus.Desc) {
+	for _, m := range c.metrics {
+		ch <- m.desc(c.ns)
+	}
+}
+
+func (c *collector) Collect(ch chan<- prometheus.Metric) {
+	c.mu.RLock()
+	wg := &sync.WaitGroup{}
+	for _, t := range c.tasks {
+		wg.Add(1)
+		go c.collect(t, ch, true, wg)
+	}
+storedLoop:
+	for {
+		// read stored metrics until the channel is flushed
+		select {
+		case m := <-c.storedMetrics:
+			ch <- m
+		default:
+			break storedLoop
+		}
+	}
+	c.mu.RUnlock()
+	wg.Wait()
+}
+
+func (c *collector) collect(t runtime.Task, ch chan<- prometheus.Metric, block bool, wg *sync.WaitGroup) {
+	if wg != nil {
+		defer wg.Done()
+	}
+	ctx := namespaces.WithNamespace(context.Background(), t.Namespace())
+	stats, err := t.Stats(ctx)
+	if err != nil {
+		log.L.WithError(err).Errorf("stat task %s", t.ID())
+		return
+	}
+	data, err := typeurl.UnmarshalAny(stats)
+	if err != nil {
+		log.L.WithError(err).Errorf("unmarshal stats for %s", t.ID())
+		return
+	}
+	s, ok := data.(*v1.Metrics)
+	if !ok {
+		log.L.WithError(err).Errorf("invalid metric type for %s", t.ID())
+		return
+	}
+	for _, m := range c.metrics {
+		m.collect(t.ID(), t.Namespace(), s, c.ns, ch, block)
+	}
+}
+
+// Add adds the provided cgroup and id so that metrics are collected and exported
+func (c *collector) Add(t runtime.Task) error {
+	if c.ns == nil {
+		return nil
+	}
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	id := taskID(t.ID(), t.Namespace())
+	if _, ok := c.tasks[id]; ok {
+		return nil // requests to collect metrics should be idempotent
+	}
+	c.tasks[id] = t
+	return nil
+}
+
+// Remove removes the provided cgroup by id from the collector
+func (c *collector) Remove(t runtime.Task) {
+	if c.ns == nil {
+		return
+	}
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	delete(c.tasks, taskID(t.ID(), t.Namespace()))
+}

metrics/cgroups/v1/oom.go

@@ -0,0 +1,158 @@
+// +build linux
+
+/*
+   Copyright The containerd Authors.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
+*/
+
+package v1
+
+import (
+	"sync"
+	"sync/atomic"
+
+	"golang.org/x/sys/unix"
+
+	"github.com/containerd/cgroups"
+	metrics "github.com/docker/go-metrics"
+	"github.com/prometheus/client_golang/prometheus"
+	"github.com/sirupsen/logrus"
+)
+
+func newOOMCollector(ns *metrics.Namespace) (*oomCollector, error) {
+	fd, err := unix.EpollCreate1(unix.EPOLL_CLOEXEC)
+	if err != nil {
+		return nil, err
+	}
+	var desc *prometheus.Desc
+	if ns != nil {
+		desc = ns.NewDesc("memory_oom", "The number of times a container has received an oom event", metrics.Total, "container_id", "namespace")
+	}
+	c := &oomCollector{
+		fd:   fd,
+		desc: desc,
+		set:  make(map[uintptr]*oom),
+	}
+	if ns != nil {
+		ns.Add(c)
+	}
+	go c.start()
+	return c, nil
+}
+
+type oomCollector struct {
+	mu sync.Mutex
+
+	desc *prometheus.Desc
+	fd   int
+	set  map[uintptr]*oom
+}
+
+type oom struct {
+	// count needs to stay the first member of this struct to ensure 64bits
+	// alignment on a 32bits machine (e.g. arm32). This is necessary as we use
+	// the sync/atomic operations on this field.
+	count     int64
+	id        string
+	namespace string
+	c         cgroups.Cgroup
+	triggers  []Trigger
+}
+
+func (o *oomCollector) Add(id, namespace string, cg cgroups.Cgroup, triggers ...Trigger) error {
+	o.mu.Lock()
+	defer o.mu.Unlock()
+	fd, err := cg.OOMEventFD()
+	if err != nil {
+		return err
+	}
+	o.set[fd] = &oom{
+		id:        id,
+		c:         cg,
+		triggers:  triggers,
+		namespace: namespace,
+	}
+	event := unix.EpollEvent{
+		Fd:     int32(fd),
+		Events: unix.EPOLLHUP | unix.EPOLLIN | unix.EPOLLERR,
+	}
+	return unix.EpollCtl(o.fd, unix.EPOLL_CTL_ADD, int(fd), &event)
+}
+
+func (o *oomCollector) Describe(ch chan<- *prometheus.Desc) {
+	ch <- o.desc
+}
+
+func (o *oomCollector) Collect(ch chan<- prometheus.Metric) {
+	o.mu.Lock()
+	defer o.mu.Unlock()
+	for _, t := range o.set {
+		c := atomic.LoadInt64(&t.count)
+		ch <- prometheus.MustNewConstMetric(o.desc, prometheus.CounterValue, float64(c), t.id, t.namespace)
+	}
+}
+
+// Close closes the epoll fd
+func (o *oomCollector) Close() error {
+	return unix.Close(o.fd)
+}
+
+func (o *oomCollector) start() {
+	var events [128]unix.EpollEvent
+	for {
+		n, err := unix.EpollWait(o.fd, events[:], -1)
+		if err != nil {
+			if err == unix.EINTR {
+				continue
+			}
+			logrus.WithError(err).Error("cgroups: epoll wait failed, OOM notifications disabled")
+			return
+		}
+		for i := 0; i < n; i++ {
+			o.process(uintptr(events[i].Fd), events[i].Events)
+		}
+	}
+}
+
+func (o *oomCollector) process(fd uintptr, event uint32) {
+	// make sure to always flush the fd
+	flush(fd)
+
+	o.mu.Lock()
+	info, ok := o.set[fd]
+	if !ok {
+		o.mu.Unlock()
+		return
+	}
+	o.mu.Unlock()
+	// if we received an event but it was caused by the cgroup being deleted and the fd
+	// being closed make sure we close our copy and remove the container from the set
+	if info.c.State() == cgroups.Deleted {
+		o.mu.Lock()
+		delete(o.set, fd)
+		o.mu.Unlock()
+		unix.Close(int(fd))
+		return
+	}
+	atomic.AddInt64(&info.count, 1)
+	for _, t := range info.triggers {
+		t(info.id, info.namespace, info.c)
+	}
+}
+
+func flush(fd uintptr) error {
+	var buf [8]byte
+	_, err := unix.Read(int(fd), buf[:])
+	return err
+}

metrics/cgroups/v1/pids.go

@@ -0,0 +1,60 @@
+// +build linux
+
+/*
+   Copyright The containerd Authors.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
+*/
+
+package v1
+
+import (
+	v1 "github.com/containerd/containerd/metrics/types/v1"
+	metrics "github.com/docker/go-metrics"
+	"github.com/prometheus/client_golang/prometheus"
+)
+
+var pidMetrics = []*metric{
+	{
+		name: "pids",
+		help: "The limit to the number of pids allowed",
+		unit: metrics.Unit("limit"),
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Pids == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Pids.Limit),
+				},
+			}
+		},
+	},
+	{
+		name: "pids",
+		help: "The current number of pids",
+		unit: metrics.Unit("current"),
+		vt:   prometheus.GaugeValue,
+		getValues: func(stats *v1.Metrics) []value {
+			if stats.Pids == nil {
+				return nil
+			}
+			return []value{
+				{
+					v: float64(stats.Pids.Current),
+				},
+			}
+		},
+	},
+}