loki.source.docker: deduplicate matching container IDs (grafana#6055)

Signed-off-by: Paschalis Tsilias <[email protected]>
Co-authored-by: Clayton Cornell <[email protected]>

CHANGELOG.md

@@ -105,11 +105,15 @@ v0.39.0 (2024-01-09)
 - Add `max_cache_size` to `prometheus.relabel` to allow configurability instead of hard coded 100,000. (@mattdurham)
 
 - Add support for `http_sd_config` within a `scrape_config` for prometheus to flow config conversion. (@erikbaranowski)
+
 - `discovery.lightsail` now supports additional parameters for configuring HTTP client settings. (@ptodev)
 - Add `sample_age_limit` to remote_write config to drop samples older than a specified duration. (@marctc)
 
 - Handle paths in the Kubelet URL for `discovery.kubelet`. (@petewall)
 
+- `loki.source.docker` now deduplicates targets which report the same container
+  ID. (@tpaschalis)
+
 ### Bugfixes
 
 - Update `pyroscope.ebpf` to fix a logical bug causing to profile to many kthreads instead of regular processes https://github.com/grafana/pyroscope/pull/2778 (@korniltsev)

Makefile

@@ -174,7 +174,7 @@ lint: agentlint
 # more without -race for packages that have known race detection issues.
 test:
 	$(GO_ENV) go test $(GO_FLAGS) -race $(shell go list ./... | grep -v /integration-tests/)
-	$(GO_ENV) go test $(GO_FLAGS) ./pkg/integrations/node_exporter ./pkg/logs ./pkg/operator ./pkg/util/k8s ./component/otelcol/processor/tail_sampling ./component/loki/source/file
+	$(GO_ENV) go test $(GO_FLAGS) ./pkg/integrations/node_exporter ./pkg/logs ./pkg/operator ./pkg/util/k8s ./component/otelcol/processor/tail_sampling ./component/loki/source/file ./component/loki/source/docker
 
 test-packages:
 	docker pull $(BUILD_IMAGE)

component/loki/source/docker/docker.go

@@ -215,13 +215,18 @@ func (c *Component) Update(args component.Arguments) error {
 
 	// Convert input targets into targets to give to tailer.
 	targets := make([]*dt.Target, 0, len(newArgs.Targets))
+	seenTargets := make(map[string]struct{}, len(newArgs.Targets))
 
 	for _, target := range newArgs.Targets {
 		containerID, ok := target[dockerLabelContainerID]
 		if !ok {
 			level.Debug(c.opts.Logger).Log("msg", "docker target did not include container ID label:"+dockerLabelContainerID)
 			continue
 		}
+		if _, seen := seenTargets[containerID]; seen {
+			continue
+		}
+		seenTargets[containerID] = struct{}{}
 
 		var labels = make(model.LabelSet)
 		for k, v := range target {

component/loki/source/docker/docker_test.go

@@ -1,13 +1,17 @@
+//go:build !race
+
 package docker
 
 import (
 	"context"
 	"testing"
 	"time"
 
+	"github.com/grafana/agent/component"
 	"github.com/grafana/agent/pkg/flow/componenttest"
 	"github.com/grafana/agent/pkg/util"
 	"github.com/grafana/river"
+	"github.com/prometheus/client_golang/prometheus"
 	"github.com/stretchr/testify/require"
 )
 
@@ -33,3 +37,39 @@ func Test(t *testing.T) {
 
 	require.NoError(t, ctrl.WaitRunning(time.Minute))
 }
+
+func TestDuplicateTargets(t *testing.T) {
+	// Use host that works on all platforms (including Windows).
+	var cfg = `
+		host       = "tcp://127.0.0.1:9376"
+		targets    = [
+			{__meta_docker_container_id = "foo", __meta_docker_port_private = "8080"},
+			{__meta_docker_container_id = "foo", __meta_docker_port_private = "8081"},
+		]
+		forward_to = []
+	`
+
+	var args Arguments
+	err := river.Unmarshal([]byte(cfg), &args)
+	require.NoError(t, err)
+
+	ctrl, err := componenttest.NewControllerFromID(util.TestLogger(t), "loki.source.docker")
+	require.NoError(t, err)
+
+	go func() {
+		err := ctrl.Run(context.Background(), args)
+		require.NoError(t, err)
+	}()
+
+	require.NoError(t, ctrl.WaitRunning(time.Minute))
+
+	cmp, err := New(component.Options{
+		ID:         "loki.source.docker.test",
+		Logger:     util.TestFlowLogger(t),
+		Registerer: prometheus.NewRegistry(),
+		DataPath:   t.TempDir(),
+	}, args)
+	require.NoError(t, err)
+
+	require.Len(t, cmp.manager.tasks, 1)
+}

docs/sources/flow/reference/components/loki.source.docker.md

@@ -131,6 +131,14 @@ fully qualified name) to store its _positions file_. The positions file
 stores the read offsets so that if there is a component or Agent restart,
 `loki.source.docker` can pick up tailing from the same spot.
 
+If the target's argument contains multiple entries with the same container
+ID (for example as a result of `discovery.docker` picking up multiple exposed
+ports or networks), `loki.source.docker` will deduplicate them, and only keep
+the first of each container ID instances, based on the
+`__meta_docker_container_id` label.  As such, the Docker daemon is queried
+for each container ID only once, and only one target will be available in the
+component's debug info.
+
 ## Example
 
 This example collects log entries from the files specified in the `targets`