tests/ingition/kdump: add a remote NFS kdump test

This way we have good coverage of most-used kdump features.
Some context on the NFS kdump configuration:
coreos/fedora-coreos-tracker#1729

This was previously merged in [1] then reverted in [2] because the nfs
server container was not multi-arch, causing the pipeline to trip on it.

It's also not functionning on systemd256 (so anything f41 and above),
see [3]

This requires #3917 for
the multi-arch container, and
#3921

[1] b10d8dc
[2] af1468c
[3] rhkdump/kdump-utils#52

Author: jbtrystram

mantle/kola/tests/ignition/kdump.go

@@ -28,6 +28,63 @@ func init() {
 		Tags:        []string{"kdump", kola.SkipBaseChecksTag, kola.NeedsInternetTag},
 		Platforms:   []string{"qemu"},
 	})
+	register.RegisterTest(&register.Test{
+		Run:         kdumpNFSTest,
+		ClusterSize: 0,
+		Name:        `kdump.crash.nfs`,
+		Description: "Verifies kdump logs are exported to NFS destination",
+		Tags:        []string{"kdump", kola.SkipBaseChecksTag, kola.NeedsInternetTag},
+		Platforms:   []string{"qemu"},
+	})
+}
+
+// This function test the remote kdump feature by:
+// - making sure kdump is ready
+// - crashing machine
+// - monitoring the expected vmcore path
+func testRemoteKdump(c cluster.TestCluster, kdump_machine platform.Machine, remote_machine platform.Machine, crash_path string) {
+
+	// Wait for kdump to become active
+	// 3 minutes should be enough to generate the kdump initramfs
+	err := util.Retry(12, 15*time.Second, func() error {
+
+		kdump_status, err := c.SSH(kdump_machine, "systemctl is-active kdump.service")
+
+		if err != nil {
+			return err
+		} else if string(kdump_status) == "inactive" {
+			return fmt.Errorf("Kdump.service is not ready: %s.", string(kdump_status))
+		}
+		return nil
+	})
+	if err != nil {
+		c.Fatalf("Timed out while waiting for kdump.service to be ready: %v", err)
+	}
+
+	// crash the kernel
+	// use systemd-run because  direclty calling `echo c > ...` will always
+	// throw an error as  the kernel immediately hangs.
+	_, err = c.SSH(kdump_machine, "sudo systemd-run sh -c 'sleep 5 && echo c > /proc/sysrq-trigger'")
+	if err != nil {
+		c.Fatalf("failed to queue kernel crash: %v", err)
+	}
+
+	// Wait for kdump to create vmcore dump on the remote host
+	err = util.Retry(8, 10*time.Second, func() error {
+
+		// Look for the crash files created on the SSH machine
+		logs, err := c.SSH(remote_machine, fmt.Sprintf("find %s -type f -name vmcore*", crash_path))
+
+		if err != nil {
+			return fmt.Errorf("failed to search for vmcore: %w", err)
+		} else if logs == nil {
+			return fmt.Errorf("No vmcore created on remote host")
+		}
+		return nil
+	})
+	if err != nil {
+		c.Fatalf("Timed out while waiting for kdump to create vmcore files: %v", err)
+	}
 }
 
 // The destination VM for kdump logs
@@ -180,45 +237,122 @@ kernel_arguments:
 		c.Fatalf("Unable to create test machine: %v", err)
 	}
 
-	// Wait for kdump to become active
-	// 3 minutes should be enough to generate the kdump initramfs
-	err = util.Retry(12, 15*time.Second, func() error {
+	testRemoteKdump(c, kdump_machine, ssh_host.Machine, "/home/core/crash")
+}
 
-		kdump_status, err := c.SSH(kdump_machine, "systemctl is-active kdump.service")
+// The destination VM for kdump logs over NFS
+type NfsServer struct {
+	Machine        platform.Machine
+	MachineAddress string
+}
 
-		if err != nil {
-			return err
-		} else if string(kdump_status) == "inactive" {
-			return fmt.Errorf(fmt.Sprintf("Kdump.service is not ready: %s.", string(kdump_status)))
-		}
-		return nil
-	})
+func setupNFSMachine(c cluster.TestCluster) NfsServer {
+	var m platform.Machine
+	var err error
+
+	options := platform.QemuMachineOptions{
+		HostForwardPorts: []platform.HostForwardPort{
+			{Service: "ssh", HostPort: 0, GuestPort: 22},
+			// Kdump NFS option does not allow a custom port
+			{Service: "nfs", HostPort: 2049, GuestPort: 2049},
+		},
+	}
+
+	nfs_server_butane := conf.Butane(`variant: fcos
+version: 1.5.0
+storage:
+  files:
+    - path: /etc/containers/systemd/nfs.container
+      overwrite: true
+      contents:
+        inline: |
+          [Container]
+          Image=quay.io/coreos-assembler/nfs
+          Volume=/var/nfs:/export
+          Network=host
+          PodmanArgs=--privileged
+          [Install]
+          WantedBy=default.target
+  directories:
+    - path: /var/nfs/crash`)
+
+	// start the machine
+	switch c := c.Cluster.(type) {
+	// These cases have to be separated because when put together to the same case statement
+	// the golang compiler no longer checks that the individual types in the case have the
+	// NewMachineWithQemuOptions function, but rather whether platform.Cluster
+	// does which fails
+	case *qemu.Cluster:
+		m, err = c.NewMachineWithQemuOptions(nfs_server_butane, options)
+	default:
+		panic("unreachable")
+	}
 	if err != nil {
-		c.Fatalf("Timed out while waiting for kdump.service to be ready: %v", err)
+		c.Fatal(err)
 	}
 
-	// crash the kernel
-	// use systemd-run because  direclty calling `echo c...` will alaways
-	// throw an error as  the kernel immediately hangs.
-	_, err = c.SSH(kdump_machine, "sudo systemd-run sh -c 'sleep 5 && echo c > /proc/sysrq-trigger'")
+	return NfsServer{
+		Machine:        m,
+		MachineAddress: "10.0.2.2",
+	}
+}
+
+func kdumpNFSTest(c cluster.TestCluster) {
+	nfs_host := setupNFSMachine(c)
+
+	butane := conf.Butane(fmt.Sprintf(`variant: fcos
+version: 1.5.0
+storage:
+  files:
+    - path: /etc/kdump.conf
+      overwrite: true
+      contents:
+        inline: |
+          nfs %s:/
+          path /crash
+          core_collector makedumpfile -l --message-level 1 -d 31
+          extra_bins /sbin/mount.nfs 
+          extra_modules nfs nfsv3 nfs_layout_nfsv41_files blocklayoutdriver nfs_layout_flexfiles nfs_layout_nfsv41_files
+systemd:
+  units:
+    - name: kdump.service
+      enabled: true
+      dropins:
+        - name: debug.conf
+          contents: |
+            [Service]
+            Environment="debug=1"
+kernel_arguments:
+    should_exist:
+      - crashkernel=512M`,
+		nfs_host.MachineAddress))
+
+	opts := platform.MachineOptions{
+		MinMemory: 2048,
+	}
+
+	kdump_machine, err := c.NewMachineWithOptions(butane, opts)
 	if err != nil {
-		c.Fatalf("failed to queue kernel crash: %v", err)
+		c.Fatalf("Unable to create test machine: %v", err)
 	}
 
-	// Wait for kdump to create vmcore dump on the remote host
-	err = util.Retry(5, 10*time.Second, func() error {
+	// XXX Refactor this
+	// Wait for nfs server to become active
+	// 1 minutes should be enough to pull the container image
+	err = util.Retry(4, 15*time.Second, func() error {
 
-		// Look for the crash files created on the SSH machine
-		logs, err := c.SSH(ssh_host.Machine, "find /home/core/crash -type f -name vmcore*")
+		nfs_status, err := c.SSH(nfs_host.Machine, "systemctl is-active nfs.service")
 
 		if err != nil {
-			return fmt.Errorf("failed to search for vmcore: %w", err)
-		} else if logs == nil {
-			return fmt.Errorf("No vmcore created on remote SSH host")
+			return err
+		} else if string(nfs_status) == "inactive" {
+			return fmt.Errorf("nfs.service is not ready: %s.", string(nfs_status))
 		}
 		return nil
 	})
 	if err != nil {
-		c.Fatalf("Timed out while waiting for kdump to create vmcore files: %v", err)
+		c.Fatalf("Timed out while waiting for nfs.service to be ready: %v", err)
 	}
+
+	testRemoteKdump(c, kdump_machine, nfs_host.Machine, "/var/nfs/crash")
 }