Add support for podman metrics in docker module (#41889) (#41966)

* Add support for podman metrics (cherry picked from commit 1fefdbb) Co-authored-by: Michael Katsoulis <[email protected]>
elastic · Dec 10, 2024 · 0ba4be3 · 0ba4be3
1 parent 8f24f9e
commit 0ba4be3
Show file tree

Hide file tree

Showing 14 changed files with 76 additions and 16 deletions.
diff --git a/metricbeat/docs/modules/docker.asciidoc b/metricbeat/docs/modules/docker.asciidoc
@@ -22,6 +22,9 @@ The Docker module is currently tested on Linux and Mac with the community
 edition engine, versions 1.11 and 17.09.0-ce. It is not tested on Windows,
 but it should also work there.
 
+The Docker module supports collection of metrics from Podman's Docker-compatible API.
+It has been tested on Linux and Mac with Podman Rest API v2.0.0 and above.
+
 [float]
 === Module-specific configuration notes
 
@@ -30,6 +33,9 @@ It is strongly recommended that you run Docker metricsets with a
 Docker API already takes up to 2 seconds. Specifying less than 3 seconds will
 result in requests that timeout, and no data will be reported for those
 requests.
+In the case of Podman, the configuration parameter `podman` should be set to `true`. 
+This enables streaming of container stats output, which allows for more accurate 
+CPU percentage calculations when using Podman.
 
 
 :edit_url:
@@ -62,6 +68,9 @@ metricbeat.modules:
   # If set to true, replace dots in labels with `_`.
   #labels.dedot: false
 
+  # Docker module supports metrics collection from podman's docker compatible API. In case of podman set to true.
+  # podman: false
+
   # Skip metrics for certain device major numbers in docker/diskio. 
   # Necessary on systems with software RAID, device mappers, 
   # or other configurations where virtual disks will sum metrics from other disks.

diff --git a/metricbeat/metricbeat.reference.yml b/metricbeat/metricbeat.reference.yml
@@ -258,6 +258,9 @@ metricbeat.modules:
   # If set to true, replace dots in labels with `_`.
   #labels.dedot: false
 
+  # Docker module supports metrics collection from podman's docker compatible API. In case of podman set to true.
+  # podman: false
+
   # Skip metrics for certain device major numbers in docker/diskio. 
   # Necessary on systems with software RAID, device mappers, 
   # or other configurations where virtual disks will sum metrics from other disks.

diff --git a/metricbeat/module/docker/_meta/config.reference.yml b/metricbeat/module/docker/_meta/config.reference.yml
@@ -17,6 +17,9 @@
   # If set to true, replace dots in labels with `_`.
   #labels.dedot: false
 
+  # Docker module supports metrics collection from podman's docker compatible API. In case of podman set to true.
+  # podman: false
+
   # Skip metrics for certain device major numbers in docker/diskio. 
   # Necessary on systems with software RAID, device mappers, 
   # or other configurations where virtual disks will sum metrics from other disks.

diff --git a/metricbeat/module/docker/_meta/config.yml b/metricbeat/module/docker/_meta/config.yml
@@ -15,6 +15,9 @@
   # If set to true, replace dots in labels with `_`.
   #labels.dedot: false
 
+  # Docker module supports metrics collection from podman's Docker-compatible API. In case of podman set to true.
+  # podman: false
+
   # Skip metrics for certain device major numbers in docker/diskio. 
   # Necessary on systems with software RAID, device mappers, 
   # or other configurations where virtual disks will sum metrics from other disks.

diff --git a/metricbeat/module/docker/_meta/docs.asciidoc b/metricbeat/module/docker/_meta/docs.asciidoc
@@ -11,6 +11,9 @@ The Docker module is currently tested on Linux and Mac with the community
 edition engine, versions 1.11 and 17.09.0-ce. It is not tested on Windows,
 but it should also work there.
 
+The Docker module supports collection of metrics from Podman's Docker-compatible API.
+It has been tested on Linux and Mac with Podman Rest API v2.0.0 and above.
+
 [float]
 === Module-specific configuration notes
 
@@ -19,3 +22,6 @@ It is strongly recommended that you run Docker metricsets with a
 Docker API already takes up to 2 seconds. Specifying less than 3 seconds will
 result in requests that timeout, and no data will be reported for those
 requests.
+In the case of Podman, the configuration parameter `podman` should be set to `true`. 
+This enables streaming of container stats output, which allows for more accurate 
+CPU percentage calculations when using Podman.
diff --git a/metricbeat/module/docker/config.go b/metricbeat/module/docker/config.go
@@ -19,14 +19,16 @@ package docker
 
 // Config contains the config needed for the docker
 type Config struct {
-	TLS   *TLSConfig `config:"ssl"`
-	DeDot bool       `config:"labels.dedot"`
+	TLS    *TLSConfig `config:"ssl"`
+	DeDot  bool       `config:"labels.dedot"`
+	Podman bool       `config:"podman"`
 }
 
 // DefaultConfig returns default module config
 func DefaultConfig() Config {
 	return Config{
-		DeDot: true,
+		DeDot:  true,
+		Podman: false,
 	}
 }
 

diff --git a/metricbeat/module/docker/cpu/cpu.go b/metricbeat/module/docker/cpu/cpu.go
@@ -40,6 +40,7 @@ type MetricSet struct {
 	cpuService   *CPUService
 	dockerClient *client.Client
 	dedot        bool
+	podman       bool
 }
 
 // New creates a new instance of the docker cpu MetricSet.
@@ -68,12 +69,13 @@ func New(base mb.BaseMetricSet) (mb.MetricSet, error) {
 		dockerClient:  client,
 		cpuService:    &CPUService{Cores: cpuConfig.Cores},
 		dedot:         config.DeDot,
+		podman:        config.Podman,
 	}, nil
 }
 
 // Fetch returns a list of docker CPU stats.
 func (m *MetricSet) Fetch(r mb.ReporterV2) error {
-	stats, err := docker.FetchStats(m.dockerClient, m.Module().Config().Timeout)
+	stats, err := docker.FetchStats(m.dockerClient, m.Module().Config().Timeout, m.podman, m.Logger())
 	if err != nil {
 		return fmt.Errorf("failed to get docker stats: %w", err)
 	}

diff --git a/metricbeat/module/docker/diskio/diskio.go b/metricbeat/module/docker/diskio/diskio.go
@@ -89,7 +89,7 @@ func New(base mb.BaseMetricSet) (mb.MetricSet, error) {
 
 // Fetch creates list of events with diskio stats for all containers.
 func (m *MetricSet) Fetch(r mb.ReporterV2) error {
-	stats, err := docker.FetchStats(m.dockerClient, m.Module().Config().Timeout)
+	stats, err := docker.FetchStats(m.dockerClient, m.Module().Config().Timeout, false, m.Logger())
 	if err != nil {
 		return fmt.Errorf("failed to get docker stats: %w", err)
 	}

diff --git a/metricbeat/module/docker/docker.go b/metricbeat/module/docker/docker.go
@@ -34,6 +34,7 @@ import (
 	"github.com/elastic/beats/v7/metricbeat/mb"
 	"github.com/elastic/beats/v7/metricbeat/mb/parse"
 	"github.com/elastic/elastic-agent-autodiscover/docker"
+	"github.com/elastic/elastic-agent-libs/logp"
 )
 
 // HostParser is a TCP host parser function for docker tcp host addresses
@@ -91,7 +92,7 @@ func NewDockerClient(endpoint string, config Config) (*client.Client, error) {
 }
 
 // FetchStats returns a list of running containers with all related stats inside
-func FetchStats(client *client.Client, timeout time.Duration) ([]Stat, error) {
+func FetchStats(client *client.Client, timeout time.Duration, stream bool, logger *logp.Logger) ([]Stat, error) {
 	ctx, cancel := context.WithTimeout(context.Background(), timeout)
 	defer cancel()
 	containers, err := client.ContainerList(ctx, container.ListOptions{})
@@ -108,7 +109,7 @@ func FetchStats(client *client.Client, timeout time.Duration) ([]Stat, error) {
 	for _, container := range containers {
 		go func(container types.Container) {
 			defer wg.Done()
-			statsQueue <- exportContainerStats(ctx, client, &container)
+			statsQueue <- exportContainerStats(ctx, client, &container, stream, logger)
 		}(container)
 	}
 
@@ -133,18 +134,41 @@ func FetchStats(client *client.Client, timeout time.Duration) ([]Stat, error) {
 // This is currently very inefficient as docker calculates the average for each request,
 // means each request will take at least 2s: https://github.com/docker/docker/blob/master/cli/command/container/stats_helpers.go#L148
 // Getting all stats at once is implemented here: https://github.com/docker/docker/pull/25361
-func exportContainerStats(ctx context.Context, client *client.Client, container *types.Container) Stat {
+// In case stream is true, we use get a stream of results for container stats. From the stream we keep the second result.
+// This is needed for podman use case where in case stream is false, no precpu stats are returned. The precpu stats
+// are required for the cpu percentage calculation. We keep the second  result as in the first result, the stats are not correct.
+func exportContainerStats(ctx context.Context, client *client.Client, container *types.Container, stream bool, logger *logp.Logger) Stat {
 	var event Stat
 	event.Container = container
-
-	containerStats, err := client.ContainerStats(ctx, container.ID, false)
+	containerStats, err := client.ContainerStats(ctx, container.ID, stream)
 	if err != nil {
+		logger.Debugf("Failed fetching container stats: %v", err)
 		return event
 	}
-
 	defer containerStats.Body.Close()
-	decoder := json.NewDecoder(containerStats.Body)
-	decoder.Decode(&event.Stats)
 
+	// JSON decoder
+	decoder := json.NewDecoder(containerStats.Body)
+	if !stream {
+		if err := decoder.Decode(&event.Stats); err != nil {
+			logger.Debugf("Failed decoding event: %v", err)
+			return event
+		}
+	} else {
+		// handle stream. Take the second result.
+		count := 0
+		for decoder.More() {
+			if err := decoder.Decode(&event.Stats); err != nil {
+				logger.Debugf("Failed decoding event: %v", err)
+				return event
+			}
+
+			count++
+			// Exit after the second result
+			if count == 2 {
+				break
+			}
+		}
+	}
 	return event
 }
diff --git a/metricbeat/module/docker/memory/memory.go b/metricbeat/module/docker/memory/memory.go
@@ -43,6 +43,7 @@ type MetricSet struct {
 	memoryService *MemoryService
 	dockerClient  *client.Client
 	dedot         bool
+	podman        bool
 	logger        *logp.Logger
 }
 
@@ -64,13 +65,14 @@ func New(base mb.BaseMetricSet) (mb.MetricSet, error) {
 		memoryService: &MemoryService{},
 		dockerClient:  dockerClient,
 		dedot:         config.DeDot,
+		podman:        config.Podman,
 		logger:        logger,
 	}, nil
 }
 
 // Fetch creates a list of memory events for each container.
 func (m *MetricSet) Fetch(r mb.ReporterV2) error {
-	stats, err := docker.FetchStats(m.dockerClient, m.Module().Config().Timeout)
+	stats, err := docker.FetchStats(m.dockerClient, m.Module().Config().Timeout, m.podman, m.Logger())
 	if err != nil {
 		return fmt.Errorf("failed to get docker stats: %w", err)
 	}

diff --git a/metricbeat/module/docker/network/network.go b/metricbeat/module/docker/network/network.go
@@ -66,7 +66,7 @@ func New(base mb.BaseMetricSet) (mb.MetricSet, error) {
 
 // Fetch methods creates a list of network events for each container.
 func (m *MetricSet) Fetch(r mb.ReporterV2) error {
-	stats, err := docker.FetchStats(m.dockerClient, m.Module().Config().Timeout)
+	stats, err := docker.FetchStats(m.dockerClient, m.Module().Config().Timeout, false, m.Logger())
 	if err != nil {
 		return fmt.Errorf("failed to get docker stats: %w", err)
 	}

diff --git a/metricbeat/module/docker/network_summary/network_summary.go b/metricbeat/module/docker/network_summary/network_summary.go
@@ -84,7 +84,7 @@ func New(base mb.BaseMetricSet) (mb.MetricSet, error) {
 // of an error set the Error field of mb.Event or simply call report.Error().
 func (m *MetricSet) Fetch(ctx context.Context, report mb.ReporterV2) error {
 
-	stats, err := docker.FetchStats(m.dockerClient, m.Module().Config().Timeout)
+	stats, err := docker.FetchStats(m.dockerClient, m.Module().Config().Timeout, false, m.Logger())
 	if err != nil {
 		return fmt.Errorf("failed to get docker stats: %w", err)
 	}

diff --git a/metricbeat/modules.d/docker.yml.disabled b/metricbeat/modules.d/docker.yml.disabled
@@ -18,6 +18,9 @@
   # If set to true, replace dots in labels with `_`.
   #labels.dedot: false
 
+  # Docker module supports metrics collection from podman's Docker-compatible API. In case of podman set to true.
+  # podman: false
+
   # Skip metrics for certain device major numbers in docker/diskio. 
   # Necessary on systems with software RAID, device mappers, 
   # or other configurations where virtual disks will sum metrics from other disks.

diff --git a/x-pack/metricbeat/metricbeat.reference.yml b/x-pack/metricbeat/metricbeat.reference.yml
@@ -503,6 +503,9 @@ metricbeat.modules:
   # If set to true, replace dots in labels with `_`.
   #labels.dedot: false
 
+  # Docker module supports metrics collection from podman's docker compatible API. In case of podman set to true.
+  # podman: false
+
   # Skip metrics for certain device major numbers in docker/diskio. 
   # Necessary on systems with software RAID, device mappers, 
   # or other configurations where virtual disks will sum metrics from other disks.