Merge pull request #27766 from Banana-Cultist/podman-events-oom

Improve OOMKilled Visibility

cmd/podman/system/events.go

@@ -54,6 +54,8 @@ type Event struct {
 	// containerExitCode is for storing the exit code of a container which can
 	// be used for "internal" event notification
 	ContainerExitCode *int `json:",omitempty"`
+	// OOMKilled indicates whether the container was killed by an OOM condition
+	OOMKilled *bool `json:",omitempty"`
 	// ID can be for the container, image, volume, etc
 	ID string `json:",omitempty"`
 	// Image used where applicable
@@ -81,6 +83,7 @@ type Event struct {
 func newEventFromLibpodEvent(e *events.Event) Event {
 	return Event{
 		ContainerExitCode: e.ContainerExitCode,
+		OOMKilled:         e.OOMKilled,
 		ID:                e.ID,
 		Image:             e.Image,
 		Name:              e.Name,

docs/source/markdown/podman-events.1.md

@@ -127,6 +127,7 @@ Format the output to JSON Lines or using the given Go template.
 | .Image                | Name of image being run (string)                                     |
 | .Name                 | Container name (string)                                              |
 | .Network              | Name of network being used (string)                                  |
+| .OOMKilled            | Whether container was killed due to OOM (bool)                       |
 | .PodID                | ID of pod associated with container, if any                          |
 | .Status               | Event status (e.g., create, start, died, ...)                        |
 | .Time                 | Event timestamp (string)                                             |

docs/source/markdown/podman-inspect.1.md.in

@@ -147,6 +147,12 @@ podman container inspect --latest --format {{.EffectiveCaps}}
 [CAP_CHOWN CAP_DAC_OVERRIDE CAP_FSETID CAP_FOWNER CAP_SETGID CAP_SETUID CAP_SETFCAP CAP_SETPCAP CAP_NET_BIND_SERVICE CAP_KILL]
 ```
 
+Inspect the specified container for the `OOMKilled` format specifier
+```
+podman container inspect myContainer --format {{.State.OOMKilled}}
+false
+```
+
 Inspect the specified pod for the `Name` format specifier:
 ```
 # podman inspect myPod --type pod --format "{{.Name}}"

libpod/events.go

@@ -105,9 +105,11 @@ func (c *Container) newContainerExitedEvent(exitCode int32) {
 	intExitCode := int(exitCode)
 	e.ContainerExitCode = &intExitCode
 
-	e.Details = events.Details{
+	attrs := c.Labels()
-		Attributes: c.Labels(),
+	if c.state.OOMKilled {
+		e.OOMKilled = &c.state.OOMKilled
 	}
+	e.Details = events.Details{Attributes: attrs}
 
 	if err := c.runtime.eventer.Write(e); err != nil {
 		logrus.Errorf("Unable to write container exited event: %q", err)

libpod/events/config.go

@@ -24,6 +24,8 @@ type Event struct {
 	// ContainerExitCode is for storing the exit code of a container which can
 	// be used for "internal" event notification
 	ContainerExitCode *int `json:",omitempty"`
+	// OOMKilled indicates whether the container was killed by an OOM condition
+	OOMKilled *bool `json:",omitempty"`
 	// ID can be for the container, image, volume, etc
 	ID string `json:",omitempty"`
 	// Image used where applicable

libpod/events/journal_linux.go

@@ -50,6 +50,9 @@ func (e EventJournalD) Write(ee Event) error {
 		if ee.ContainerExitCode != nil {
 			m["PODMAN_EXIT_CODE"] = strconv.Itoa(*ee.ContainerExitCode)
 		}
+		if ee.OOMKilled != nil {
+			m["PODMAN_OOM_KILLED"] = strconv.FormatBool(*ee.OOMKilled)
+		}
 		if ee.PodID != "" {
 			m["PODMAN_POD_ID"] = ee.PodID
 		}
@@ -250,6 +253,14 @@ func newEventFromJournalEntry(entry *sdjournal.JournalEntry) (*Event, error) {
 				newEvent.ContainerExitCode = &intCode
 			}
 		}
+		if val, ok := entry.Fields["PODMAN_OOM_KILLED"]; ok {
+			oomKilled, err := strconv.ParseBool(val)
+			if err != nil {
+				logrus.Errorf("Parsing event oom killed value %s", val)
+			} else {
+				newEvent.OOMKilled = &oomKilled
+			}
+		}
 		if err := getLabelsFromJournal(entry, &newEvent); err != nil {
 			return nil, err
 		}

pkg/domain/entities/events.go

@@ -29,6 +29,16 @@ func ConvertToLibpodEvent(e Event) *libpodEvents.Event {
 	if err != nil {
 		return nil
 	}
+	var (
+		oomKilled bool
+		hasOOM    bool
+	)
+	if raw, ok := e.Actor.Attributes["oomKilled"]; ok {
+		if parsed, err := strconv.ParseBool(raw); err == nil {
+			oomKilled = parsed
+			hasOOM = true
+		}
+	}
 	image := e.Actor.Attributes["image"]
 	name := e.Actor.Attributes["name"]
 	network := e.Actor.Attributes["network"]
@@ -41,7 +51,8 @@ func ConvertToLibpodEvent(e Event) *libpodEvents.Event {
 	delete(details, "podId")
 	delete(details, "error")
 	delete(details, "containerExitCode")
-	return &libpodEvents.Event{
+	delete(details, "oomKilled")
+	newEvent := &libpodEvents.Event{
 		ContainerExitCode: &exitCode,
 		ID:                e.Actor.ID,
 		Image:             image,
@@ -57,6 +68,10 @@ func ConvertToLibpodEvent(e Event) *libpodEvents.Event {
 			Attributes: details,
 		},
 	}
+	if hasOOM {
+		newEvent.OOMKilled = &oomKilled
+	}
+	return newEvent
 }
 
 // ConvertToEntitiesEvent converts a libpod event to an entities one.
@@ -70,6 +85,9 @@ func ConvertToEntitiesEvent(e libpodEvents.Event) *types.Event {
 	if e.ContainerExitCode != nil {
 		attributes["containerExitCode"] = strconv.Itoa(*e.ContainerExitCode)
 	}
+	if e.OOMKilled != nil {
+		attributes["oomKilled"] = strconv.FormatBool(*e.OOMKilled)
+	}
 	attributes["podId"] = e.PodID
 	if e.Network != "" {
 		attributes["network"] = e.Network

test/system/090-events.bats

@@ -338,6 +338,24 @@ EOF
     assert "$output" = "$lvalue" "podman-events output includes container label"
 }
 
+@test "events - died event contains OOMKilled attribute" {
+    local cname=c-$(safename)
+
+    run_podman run -d --rm \
+                --name $cname \
+                --memory 20m \
+                --cgroup-conf=memory.oom.group=1 \
+                $IMAGE sh -c "x=a; while :; do x=\$x\$x\$x\$x; done"
+    run_podman wait $cname
+    run_podman events \
+               --filter container=$cname \
+               --filter event=died \
+               --stream=false \
+               --format "{{.OOMKilled}}"
+
+    assert "$output" = "true" "OOMKilled attribute should be present and set to true"
+}
+
 # bats test_tags=ci:parallel
 @test "events - backend none should error" {
     skip_if_remote "remote does not support --events-backend"