Automate health checks feat

cmd/nerdctl/container/container_create.go

@@ -279,10 +279,6 @@ func createOptions(cmd *cobra.Command) (types.ContainerCreateOptions, error) {
 	if err != nil {
 		return opt, err
 	}
-	opt.HealthStartInterval, err = cmd.Flags().GetDuration("health-start-interval")
-	if err != nil {
-		return opt, err
-	}
 	opt.NoHealthcheck, err = cmd.Flags().GetBool("no-healthcheck")
 	if err != nil {
 		return opt, err

cmd/nerdctl/container/container_health_check_test.go → cmd/nerdctl/container/container_health_check_linux_test.go

@@ -32,11 +32,13 @@ import (
 	"github.com/containerd/nerdctl/mod/tigron/tig"
 
 	"github.com/containerd/nerdctl/v2/pkg/healthcheck"
+	"github.com/containerd/nerdctl/v2/pkg/rootlessutil"
 	"github.com/containerd/nerdctl/v2/pkg/testutil"
 	"github.com/containerd/nerdctl/v2/pkg/testutil/nerdtest"
 )
 
 func TestContainerHealthCheckBasic(t *testing.T) {
+
 	testCase := nerdtest.Setup()
 
 	// Docker CLI does not provide a standalone healthcheck command.
@@ -602,3 +604,320 @@ func TestContainerHealthCheckAdvance(t *testing.T) {
 
 	testCase.Run(t)
 }
+
+func TestHealthCheck_SystemdIntegration_Basic(t *testing.T) {
+	testCase := nerdtest.Setup()
+	testCase.Require = require.Not(nerdtest.Docker)
+	// Skip systemd tests in rootless environment to bypass dbus permission issues
+	if rootlessutil.IsRootless() {
+		t.Skip("systemd healthcheck tests are skipped in rootless environment")
+	}
+
+	testCase.SubTests = []*test.Case{
+		{
+			Description: "Basic healthy container with systemd-triggered healthcheck",
+			Setup: func(data test.Data, helpers test.Helpers) {
+				helpers.Ensure("run", "-d", "--name", data.Identifier(),
+					"--health-cmd", "echo healthy",
+					"--health-interval", "2s",
+					testutil.CommonImage, "sleep", "30")
+				nerdtest.EnsureContainerStarted(helpers, data.Identifier())
+			},
+			Cleanup: func(data test.Data, helpers test.Helpers) {
+				// Ensure proper cleanup of systemd units
+				helpers.Anyhow("stop", data.Identifier())
+				helpers.Anyhow("rm", "-f", data.Identifier())
+			},
+			Expected: func(data test.Data, helpers test.Helpers) *test.Expected {
+				return &test.Expected{
+					ExitCode: 0,
+					Output: expect.All(func(stdout string, t tig.T) {
+						var h *healthcheck.Health
+
+						// Poll up to 5 times for health status
+						maxAttempts := 5
+						var finalStatus string
+
+						for i := 0; i < maxAttempts; i++ {
+							inspect := nerdtest.InspectContainer(helpers, data.Identifier())
+							h = inspect.State.Health
+
+							assert.Assert(t, h != nil, "expected health state to be present")
+							finalStatus = h.Status
+
+							// If healthy, break and pass the test
+							if finalStatus == "healthy" {
+								t.Log(fmt.Sprintf("Container became healthy on attempt %d/%d", i+1, maxAttempts))
+								break
+							}
+
+							// If unhealthy, fail immediately
+							if finalStatus == "unhealthy" {
+								assert.Assert(t, false, fmt.Sprintf("Container became unhealthy on attempt %d/%d, status: %s", i+1, maxAttempts, finalStatus))
+								return
+							}
+
+							// If not the last attempt, wait before retrying
+							if i < maxAttempts-1 {
+								t.Log(fmt.Sprintf("Attempt %d/%d: status is '%s', waiting 1 second before retry", i+1, maxAttempts, finalStatus))
+								time.Sleep(1 * time.Second)
+							}
+						}
+
+						if finalStatus != "healthy" {
+							assert.Assert(t, false, fmt.Sprintf("Container did not become healthy after %d attempts, final status: %s", maxAttempts, finalStatus))
+							return
+						}
+
+						assert.Assert(t, len(h.Log) > 0, "expected at least one health check log entry")
+					}),
+				}
+			},
+		},
+		{
+			Description: "Kill stops healthcheck execution and cleans up systemd timer",
+			Setup: func(data test.Data, helpers test.Helpers) {
+				helpers.Ensure("run", "-d", "--name", data.Identifier(),
+					"--health-cmd", "echo healthy",
+					"--health-interval", "1s",
+					testutil.CommonImage, "sleep", "30")
+				nerdtest.EnsureContainerStarted(helpers, data.Identifier())
+				helpers.Ensure("kill", data.Identifier())
+			},
+			Cleanup: func(data test.Data, helpers test.Helpers) {
+				// Container is already killed, just remove it
+				helpers.Anyhow("rm", "-f", data.Identifier())
+			},
+			Expected: func(data test.Data, helpers test.Helpers) *test.Expected {
+				return &test.Expected{
+					ExitCode: expect.ExitCodeNoCheck,
+					Output: func(stdout string, t tig.T) {
+						// Get container info for verification
+						inspect := nerdtest.InspectContainer(helpers, data.Identifier())
+						containerID := inspect.ID
+						h := inspect.State.Health
+
+						// Verify health state and logs exist
+						assert.Assert(t, h != nil, "expected health state to be present")
+						assert.Assert(t, len(h.Log) > 0, "expected at least one health check log entry")
+
+						// Get container FinishedAt timestamp
+						containerEnd, err := time.Parse(time.RFC3339Nano, inspect.State.FinishedAt)
+						assert.NilError(t, err, "parsing container FinishedAt")
+
+						// Assert all healthcheck log start times are before container finished
+						for _, entry := range h.Log {
+							assert.Assert(t, entry.Start.Before(containerEnd), "healthcheck ran after container was killed")
+						}
+
+						// Ensure systemd timers are removed
+						result := helpers.Custom("systemctl", "list-timers", "--all", "--no-pager")
+						result.Run(&test.Expected{
+							ExitCode: expect.ExitCodeNoCheck,
+							Output: func(stdout string, _ tig.T) {
+								assert.Assert(t, !strings.Contains(stdout, containerID),
+									"expected nerdctl healthcheck timer for container ID %s to be removed after container stop", containerID)
+							},
+						})
+					},
+				}
+			},
+		},
+		{
+			Description: "Remove cleans up systemd timer",
+			Setup: func(data test.Data, helpers test.Helpers) {
+				helpers.Ensure("run", "-d", "--name", data.Identifier(),
+					"--health-cmd", "echo healthy",
+					"--health-interval", "1s",
+					testutil.CommonImage, "sleep", "30")
+				nerdtest.EnsureContainerStarted(helpers, data.Identifier())
+				helpers.Ensure("rm", "-f", data.Identifier())
+			},
+			Cleanup: func(data test.Data, helpers test.Helpers) {
+				// Container is already removed, no cleanup needed
+				helpers.Anyhow("rm", "-f", data.Identifier())
+			},
+			Expected: func(data test.Data, helpers test.Helpers) *test.Expected {
+				return &test.Expected{
+					ExitCode: expect.ExitCodeNoCheck,
+					Output: func(stdout string, t tig.T) {
+						inspect := nerdtest.InspectContainer(helpers, data.Identifier())
+						containerID := inspect.ID
+
+						// Check systemd timers to ensure cleanup
+						result := helpers.Custom("systemctl", "list-timers", "--all", "--no-pager")
+						result.Run(&test.Expected{
+							ExitCode: expect.ExitCodeNoCheck,
+							Output: func(stdout string, _ tig.T) {
+								// Verify systemd timer has been cleaned up by checking systemctl output
+								// We check that no timer contains our test identifier
+								assert.Assert(t, !strings.Contains(stdout, containerID),
+									"expected nerdctl healthcheck timer for container ID %s to be removed after container removal", containerID)
+							},
+						})
+					},
+				}
+			},
+		},
+		{
+			Description: "Stop cleans up systemd timer",
+			Setup: func(data test.Data, helpers test.Helpers) {
+				helpers.Ensure("run", "-d", "--name", data.Identifier(),
+					"--health-cmd", "echo healthy",
+					"--health-interval", "1s",
+					testutil.CommonImage, "sleep", "30")
+				nerdtest.EnsureContainerStarted(helpers, data.Identifier())
+				helpers.Ensure("stop", data.Identifier())
+			},
+			Cleanup: func(data test.Data, helpers test.Helpers) {
+				// Container is already stopped, just remove it
+				helpers.Anyhow("rm", "-f", data.Identifier())
+			},
+			Expected: func(data test.Data, helpers test.Helpers) *test.Expected {
+				return &test.Expected{
+					ExitCode: expect.ExitCodeNoCheck,
+					Output: func(stdout string, t tig.T) {
+						// Get container info for verification
+						inspect := nerdtest.InspectContainer(helpers, data.Identifier())
+						containerID := inspect.ID
+
+						// Ensure systemd timers are removed
+						result := helpers.Custom("systemctl", "list-timers", "--all", "--no-pager")
+						result.Run(&test.Expected{
+							ExitCode: expect.ExitCodeNoCheck,
+							Output: func(stdout string, _ tig.T) {
+								assert.Assert(t, !strings.Contains(stdout, containerID),
+									"expected nerdctl healthcheck timer for container ID %s to be removed after container stop", containerID)
+							},
+						})
+					},
+				}
+			},
+		},
+	}
+	testCase.Run(t)
+}
+
+func TestHealthCheck_SystemdIntegration_Advanced(t *testing.T) {
+
+	testCase := nerdtest.Setup()
+	testCase.Require = require.Not(nerdtest.Docker)
+	// Skip systemd tests in rootless environment to bypass dbus permission issues
+	if rootlessutil.IsRootless() {
+		t.Skip("systemd healthcheck tests are skipped in rootless environment")
+	}
+
+	testCase.SubTests = []*test.Case{
+		{
+			// Tests that CreateTimer() successfully creates systemd timer units and
+			// RemoveTransientHealthCheckFiles() properly cleans up units when container stops.
+			Description: "Systemd timer unit creation and cleanup",
+			Setup: func(data test.Data, helpers test.Helpers) {
+				helpers.Ensure("run", "-d", "--name", data.Identifier(),
+					"--health-cmd", "echo healthy",
+					"--health-interval", "1s",
+					testutil.CommonImage, "sleep", "30")
+				nerdtest.EnsureContainerStarted(helpers, data.Identifier())
+			},
+			Cleanup: func(data test.Data, helpers test.Helpers) {
+				helpers.Anyhow("rm", "-f", data.Identifier())
+			},
+			Command: func(data test.Data, helpers test.Helpers) test.TestableCommand {
+				return helpers.Command("inspect", data.Identifier())
+			},
+			Expected: func(data test.Data, helpers test.Helpers) *test.Expected {
+				return &test.Expected{
+					ExitCode: 0,
+					Output: expect.All(func(stdout string, t tig.T) {
+						// Get container ID and check systemd timer
+						containerInspect := nerdtest.InspectContainer(helpers, data.Identifier())
+						containerID := containerInspect.ID
+
+						// Check systemd timer
+						result := helpers.Custom("systemctl", "list-timers", "--all", "--no-pager")
+						result.Run(&test.Expected{
+							ExitCode: expect.ExitCodeNoCheck,
+							Output: func(stdout string, _ tig.T) {
+								// Verify that a timer exists for this specific container
+								assert.Assert(t, strings.Contains(stdout, containerID),
+									"expected to find nerdctl healthcheck timer containing container ID: %s", containerID)
+							},
+						})
+						// Stop container and verify cleanup
+						helpers.Ensure("stop", data.Identifier())
+
+						// Check that timer is gone
+						result = helpers.Custom("systemctl", "list-timers", "--all", "--no-pager")
+						result.Run(&test.Expected{
+							ExitCode: expect.ExitCodeNoCheck,
+							Output: func(stdout string, _ tig.T) {
+								assert.Assert(t, !strings.Contains(stdout, containerID),
+									"expected nerdctl healthcheck timer for container ID %s to be removed after container stop", containerID)
+							},
+						})
+					}),
+				}
+			},
+		},
+		{
+			Description: "Container restart recreates systemd timer",
+			Setup: func(data test.Data, helpers test.Helpers) {
+				helpers.Ensure("run", "-d", "--name", data.Identifier(),
+					"--health-cmd", "echo restart-test",
+					"--health-interval", "2s",
+					testutil.CommonImage, "sleep", "60")
+				nerdtest.EnsureContainerStarted(helpers, data.Identifier())
+			},
+			Cleanup: func(data test.Data, helpers test.Helpers) {
+				helpers.Anyhow("rm", "-f", data.Identifier())
+			},
+			Command: func(data test.Data, helpers test.Helpers) test.TestableCommand {
+				// Get container ID for verification
+				containerInspect := nerdtest.InspectContainer(helpers, data.Identifier())
+				containerID := containerInspect.ID
+
+				// Step 1: Verify timer exists initially
+				result := helpers.Custom("systemctl", "list-timers", "--all", "--no-pager")
+				result.Run(&test.Expected{
+					ExitCode: expect.ExitCodeNoCheck,
+					Output: func(stdout string, t tig.T) {
+						assert.Assert(t, strings.Contains(stdout, containerID),
+							"expected timer for container %s to exist initially", containerID)
+					},
+				})
+
+				// Step 2: Stop container
+				helpers.Ensure("stop", data.Identifier())
+
+				// Step 3: Verify timer is removed after stop
+				result = helpers.Custom("systemctl", "list-timers", "--all", "--no-pager")
+				result.Run(&test.Expected{
+					ExitCode: expect.ExitCodeNoCheck,
+					Output: func(stdout string, t tig.T) {
+						assert.Assert(t, !strings.Contains(stdout, containerID),
+							"expected timer for container %s to be removed after stop", containerID)
+					},
+				})
+
+				// Step 4: Restart container
+				helpers.Ensure("start", data.Identifier())
+				nerdtest.EnsureContainerStarted(helpers, data.Identifier())
+
+				// Step 5: Verify timer is recreated after restart - this is our final verification
+				return helpers.Custom("systemctl", "list-timers", "--all", "--no-pager")
+			},
+			Expected: func(data test.Data, helpers test.Helpers) *test.Expected {
+				return &test.Expected{
+					ExitCode: expect.ExitCodeNoCheck,
+					Output: func(stdout string, t tig.T) {
+						containerInspect := nerdtest.InspectContainer(helpers, data.Identifier())
+						containerID := containerInspect.ID
+						assert.Assert(t, strings.Contains(stdout, containerID),
+							"expected timer for container %s to be recreated after restart", containerID)
+					},
+				}
+			},
+		},
+	}
+	testCase.Run(t)
+}

cmd/nerdctl/container/container_run.go

@@ -37,6 +37,7 @@ import (
 	"github.com/containerd/nerdctl/v2/pkg/containerutil"
 	"github.com/containerd/nerdctl/v2/pkg/defaults"
 	"github.com/containerd/nerdctl/v2/pkg/errutil"
+	"github.com/containerd/nerdctl/v2/pkg/healthcheck"
 	"github.com/containerd/nerdctl/v2/pkg/labels"
 	"github.com/containerd/nerdctl/v2/pkg/logging"
 	"github.com/containerd/nerdctl/v2/pkg/netutil"
@@ -240,7 +241,6 @@ func setCreateFlags(cmd *cobra.Command) {
 	cmd.Flags().Duration("health-timeout", 0, "Maximum time to allow one check to run (default: 30s)")
 	cmd.Flags().Int("health-retries", 0, "Consecutive failures needed to report unhealthy (default: 3)")
 	cmd.Flags().Duration("health-start-period", 0, "Start period for the container to initialize before starting health-retries countdown")
-	cmd.Flags().Duration("health-start-interval", 0, "Time between running the checks during the start period")
 	cmd.Flags().Bool("no-healthcheck", false, "Disable any container-specified HEALTHCHECK")
 
 	// #region env flags
@@ -445,6 +445,14 @@ func runAction(cmd *cobra.Command, args []string) error {
 		return err
 	}
 
+	// Setup container healthchecks.
+	if err := healthcheck.CreateTimer(ctx, c); err != nil {
+		return fmt.Errorf("failed to create healthcheck timer: %w", err)
+	}
+	if err := healthcheck.StartTimer(ctx, c); err != nil {
+		return fmt.Errorf("failed to start healthcheck timer: %w", err)
+	}
+
 	if createOpt.Detach {
 		fmt.Fprintln(createOpt.Stdout, id)
 		return nil

cmd/nerdctl/container/container_run_test.go

@@ -841,6 +841,9 @@ func TestRunDomainname(t *testing.T) {
 }
 
 func TestRunHealthcheckFlags(t *testing.T) {
+	if rootlessutil.IsRootless() {
+		t.Skip("healthcheck tests are skipped in rootless environment")
+	}
 	testCase := nerdtest.Setup()
 
 	testCases := []struct {
@@ -990,6 +993,9 @@ func TestRunHealthcheckFlags(t *testing.T) {
 }
 
 func TestRunHealthcheckFromImage(t *testing.T) {
+	if rootlessutil.IsRootless() {
+		t.Skip("healthcheck tests are skipped in rootless environment")
+	}
 	nerdtest.Setup()
 
 	dockerfile := fmt.Sprintf(`FROM %s