fix: ensure containerd-related directories removed on failed bootstrap/join-cluster.

src/k8s/pkg/k8sd/app/hooks_remove.go

@@ -145,5 +145,9 @@ func (a *App) onPreRemove(ctx context.Context, s state.State, force bool) (rerr
 		}
 	}
 
+	if err := setup.TryCleanupContainerdPaths(snap); err != nil {
+		log.Error(err, "Failed to perform containerd cleanup")
+	}
+
 	return nil
 }

src/k8s/pkg/k8sd/setup/containerd.go

@@ -8,6 +8,7 @@ import (
 
 	"dario.cat/mergo"
 	"github.com/canonical/k8s/pkg/k8sd/images"
+	"github.com/canonical/k8s/pkg/log"
 	"github.com/canonical/k8s/pkg/snap"
 	snaputil "github.com/canonical/k8s/pkg/snap/util"
 	"github.com/canonical/k8s/pkg/utils"
@@ -166,23 +167,79 @@ func Containerd(snap snap.Snap, extraContainerdConfig map[string]any, extraArgs
 	return nil
 }
 
-func saveSnapContainerdPaths(s snap.Snap) error {
-	// Write the containerd-related paths to files to properly clean-up on removal.
+// Returns a mapping between the absolute paths of the lockfiles within the
+// k8s snap and the absolute paths of the containerd directory they lock.
+func containerdLockPathsForSnap(s snap.Snap) map[string]string {
 	m := map[string]string{
 		"containerd-socket-path": s.ContainerdSocketDir(),
 		"containerd-config-dir":  s.ContainerdConfigDir(),
 		"containerd-root-dir":    s.ContainerdRootDir(),
 		"containerd-cni-bin-dir": s.CNIBinDir(),
 	}
 
-	for filename, content := range m {
-		if err := utils.WriteFile(filepath.Join(s.LockFilesDir(), filename), []byte(content), 0o600); err != nil {
-			return fmt.Errorf("failed to write %s: %w", filename, err)
+	prefixed := map[string]string{}
+	for k, v := range m {
+		prefixed[filepath.Join(s.LockFilesDir(), k)] = v
+	}
+
+	return prefixed
+}
+
+// Creates the lock files for the containerd directory paths to be used for later cleanup.
+func saveSnapContainerdPaths(s snap.Snap) error {
+	for lockpath, dirpath := range containerdLockPathsForSnap(s) {
+		if err := utils.WriteFile(lockpath, []byte(dirpath), 0o600); err != nil {
+			return fmt.Errorf("failed to write %s: %w", lockpath, err)
 		}
 	}
 	return nil
 }
 
+// Attempts to clean up all containerd directories which were created
+// by the k8s-snap based on the existence of their respective lockfiles
+// located in the directory returned by `s.LockFilesDir()`.
+func TryCleanupContainerdPaths(s snap.Snap) error {
+	log := log.L()
+
+	for lockpath, dirpath := range containerdLockPathsForSnap(s) {
+		// Ensure lockfile exists:
+		if _, err := os.Stat(lockpath); os.IsNotExist(err) {
+			log.Info("WARN: failed to find containerd lockfile, no cleanup will be perfomed", "lockfile", lockpath, "directory", dirpath)
+			continue
+		}
+
+		// Ensure lockfile's contents is the one we expect:
+		lockfile_contents := ""
+		if contents, err := os.ReadFile(lockpath); err != nil {
+			log.Info("WARN: failed to read contents of lockfile", "lockfile", lockpath, "error", err)
+			continue
+		} else {
+			lockfile_contents = string(contents)
+		}
+
+		if lockfile_contents != dirpath {
+			log.Info("WARN: lockfile points to different path than expected", "lockfile", lockpath, "expected", dirpath, "actual", lockfile_contents)
+			continue
+		}
+
+		// Check directory exists before attempting to remove:
+		if _, err := os.Stat(dirpath); os.IsNotExist(err) {
+			log.Info("Containerd directory doesn't exist; skipping cleanup", "directory", dirpath)
+		} else {
+			if err := os.RemoveAll(dirpath); err != nil {
+				log.Info("WARN: failed to remove containerd data directory", "directory", dirpath, "error", err)
+				continue // Avoid removing the lockfile path.
+			}
+		}
+
+		if err := os.Remove(lockpath); err != nil {
+			log.Info("WARN: Failed to remove containerd lockfile", "lockfile", lockpath)
+		}
+	}
+
+	return nil
+}
+
 func init() {
 	images.Register(defaultPauseImage)
 }

src/k8s/pkg/k8sd/setup/containerd_test.go

@@ -31,6 +31,7 @@ func TestContainerd(t *testing.T) {
 			ContainerdRegistryConfigDir: filepath.Join(dir, "containerd-hosts"),
 			ContainerdStateDir:          filepath.Join(dir, "containerd-state"),
 			ContainerdExtraConfigDir:    filepath.Join(dir, "containerd-confd"),
+			LockFilesDir:                filepath.Join(dir, "lockfiles"),
 			ServiceArgumentsDir:         filepath.Join(dir, "args"),
 			CNIBinDir:                   filepath.Join(dir, "opt-cni-bin"),
 			CNIConfDir:                  filepath.Join(dir, "cni-netd"),
@@ -129,9 +130,16 @@ func TestContainerd(t *testing.T) {
 			"containerd-root-dir":    s.ContainerdRootDir(),
 			"containerd-cni-bin-dir": s.CNIBinDir(),
 		}
-		for filename, content := range m {
 
-			b, err := os.ReadFile(filepath.Join(s.LockFilesDir(), filename))
+		// Ensure locks directory exists:
+		files, err := os.ReadDir(s.LockFilesDir())
+		g.Expect(err).To(Not(HaveOccurred()))
+		t.Logf("Lockfiles currently in %q: %v", s.LockFilesDir(), files)
+
+		for filename, content := range m {
+			path := filepath.Join(s.LockFilesDir(), filename)
+			t.Logf("Checking path: %s", path)
+			b, err := os.ReadFile(path)
 			g.Expect(err).To(Not(HaveOccurred()))
 			g.Expect(string(b)).To(Equal(content))
 		}

tests/integration/tests/test_cleanup.py

@@ -5,10 +5,12 @@
 from typing import List
 
 import pytest
-from test_util import harness, tags, util
+from test_util import config, harness, tags, util
 
 LOG = logging.getLogger(__name__)
 
+KUBE_CONTROLLER_MANAGER_SNAP_PORT = 10257
+
 CONTAINERD_PATHS = [
     "/etc/containerd",
     "/opt/cni/bin",
@@ -17,21 +19,71 @@
 ]
 
 
+def _assert_paths_not_exist(instance: harness.Instance, paths: List[str]):
+    process = instance.exec(["ls", *paths], capture_output=True, text=True, check=False)
+    for path in paths:
+        assert f"cannot access '{path}': No such file or directory" in process.stderr
+
+
 @pytest.mark.node_count(1)
 @pytest.mark.tags(tags.NIGHTLY)
 def test_node_cleanup(instances: List[harness.Instance], tmp_path):
+    """Verifies that a `snap remove k8s` will perform proper cleanup."""
     instance = instances[0]
     util.wait_for_dns(instance)
     util.wait_for_network(instance)
 
     util.remove_k8s_snap(instance)
 
     # Check that the containerd-related folders are removed on snap removal.
-    process = instance.exec(
-        ["ls", *CONTAINERD_PATHS], capture_output=True, text=True, check=False
-    )
-    for path in CONTAINERD_PATHS:
-        assert f"cannot access '{path}': No such file or directory" in process.stderr
+    _assert_paths_not_exist(instance, CONTAINERD_PATHS)
 
     util.setup_k8s_snap(instance, tmp_path)
     instance.exec(["k8s", "bootstrap"])
+
+
+@pytest.mark.node_count(1)
+@pytest.mark.no_setup()
+@pytest.mark.tags(tags.NIGHTLY)
+def test_containerd_path_cleanup_on_failed_init(
+    instances: List[harness.Instance], tmp_path
+):
+    """Tests that a failed `bootsrap` properly cleans up any
+    containerd-related paths it may have created as part of the
+    failed `bootstrap`.
+
+    It induces a bootstrap failure by pre-binding a required k8s service
+    port (10257 for the kube-controller-manager) before running `k8s bootstrap`.
+
+    NOTE: a failed `join-cluster` will trigger the exact same cleanup
+    hook, so the test implicitly applies to it as well.
+    """
+    instance = instances[0]
+    expected_code = 1
+    expected_message = (
+        "Encountered error(s) while verifying port availability for Kubernetes "
+        "services: Port 10257 (needed by: kube-controller-manager) is already in use."
+    )
+
+    with util.open_port(KUBE_CONTROLLER_MANAGER_SNAP_PORT) as _:
+        util.setup_k8s_snap(instance, tmp_path, config.SNAP, connect_interfaces=False)
+
+        proc = instance.exec(
+            ["k8s", "bootstrap"], capture_output=True, text=True, check=False
+        )
+
+        if proc.returncode != expected_code:
+            raise AssertionError(
+                f"Expected `k8s bootstrap` to exit with code {expected_code}, "
+                f"but it exited with {proc.returncode}.\n"
+                f"Stdout was: \n{proc.stdout}.\nStderr was: \n{proc.stderr}"
+            )
+
+        if expected_message not in proc.stderr:
+            raise AssertionError(
+                f"Expected to find port-related warning '{expected_message}' in "
+                "stderr of the `k8s bootstrap` command.\n"
+                f"Stdout was: \n{proc.stdout}.\nStderr was: \n{proc.stderr}"
+            )
+
+        _assert_paths_not_exist(instance, CONTAINERD_PATHS)

tests/integration/tests/test_util/util.py

@@ -1,11 +1,13 @@
 #
 # Copyright 2024 Canonical, Ltd.
 #
+import contextlib
 import ipaddress
 import json
 import logging
 import re
 import shlex
+import socket
 import subprocess
 import urllib.request
 from datetime import datetime
@@ -514,3 +516,28 @@ def find_suitable_cidr(parent_cidr: str, excluded_ips: List[str]):
 
         return str(lb_net)
     raise RuntimeError("Could not find a suitable CIDR for LoadBalancer services")
+
+
+@contextlib.contextmanager
+def open_port(
+    port: int,
+    host: str = "",
+    address_family: socket.AddressFamily = socket.AF_INET,
+    socket_kind: socket.SocketKind = socket.SOCK_STREAM,
+):
+    """Context manager which opens a socket with the given properties
+    and binds it to the given port.
+
+    Yields the socket object for use.
+    """
+    sock = socket.socket(family=address_family, type=socket_kind)
+    if not host:
+        host = socket.gethostname()
+    sock.bind((host, port))
+    LOG.info(f"Successfully bound new socket on '{host}:{port}'")
+
+    try:
+        yield sock
+    finally:
+        sock.close()
+        LOG.info(f"Closed socket on '{host}:{port}'")