19700: Enable site restore in Checkmk Docker containers

CMK-23809
CMK-28699
SUP-26665

Change-Id: If0bc6269a36cdbfe6ed76325199a3cc26d240038

Author: SoloJacobs

.werks/19700.md

@@ -0,0 +1,25 @@
+[//]: # (werk v3)
+# Enable site restore in Checkmk Docker containers
+
+key        | value
+---------- | ---
+date       | 2026-04-09T11:38:14.120155+00:00
+version    | 2.6.0b1
+class      | fix
+edition    | community
+component  | omd
+level      | 1
+compatible | yes
+
+Previously, restoring a site backup in a Checkmk Docker container could result in the following error:
+
+```sh
+File "/omd/versions/2.3.0p41.cce/lib/python3.12/shutil.py", line 662, in _rmtree_safe_fd
+    os.rmdir(name, dir_fd=dirfd)
+OSError: [Errno 16] Device or resource busy: '/omd/sites/disaster/tmp'
+```
+
+The restore would abort partway through, leaving the site home directory nearly empty and the site inaccessible.
+
+Users who cannot update can work around this issue by not using a `--tmpfs`
+mount at `tmp/` and retrying the restore.

omd/packages/omd/omdlib/restore.py

@@ -12,7 +12,7 @@
 from omdlib.contexts import SiteContext
 from omdlib.init_scripts import call_init_scripts
 from omdlib.site_paths import SitePaths
-from omdlib.tmpfs import unmount_tmpfs
+from omdlib.tmpfs import unmount_tmpfs_without_save
 from omdlib.user_processes import kill_site_user_processes
 
 
@@ -30,7 +30,8 @@ def prepare_restore_as_site_user(site: SiteContext, kill: bool, verbose: bool) -
     kill_site_user_processes(site.name, verbose)
     ok()
 
-    unmount_tmpfs(site.name, site_home, site.tmp_dir)
+    # We don't need to save the `tmp/` folder, since `clear_site_home` will remove it anyway.
+    unmount_tmpfs_without_save(site.name, site.tmp_dir, output=True, kill=kill)
 
     sys.stdout.write("Deleting existing site data...")
     clear_site_home(Path(site_home))
@@ -77,6 +78,11 @@ def clear_site_home(site_home: Path) -> None:
         for entry in scaniter:
             if entry.name == restore_working_dir.name:
                 continue
+            elif entry.name == "tmp":
+                # tmp is excluded from backups. The content is cleared by `unmount_tmpfs_without_save`,
+                # but the directory must remain, since it may be a Docker-managed tmpfs mount point
+                # that cannot be removed (EBUSY).
+                continue
             elif entry.is_dir(follow_symlinks=False):
                 shutil.rmtree(entry.path)
             else:

tests/unit/omdlib/test_restore.py

@@ -209,6 +209,38 @@ def by_name(dir_: Directory) -> str:
             frozenset({Symlink(name="a link", path=Path("../up"))}),
             frozenset({Directory(name=".restore_working_dir")}),
         ),
+        (
+            frozenset(
+                {
+                    Directory(
+                        name="etc",
+                        files=frozenset(
+                            {File("environment", content=b"# Custom environment variables\n")}
+                        ),
+                    ),
+                }
+            ),
+            frozenset(),
+            frozenset(
+                {
+                    # tmp/ is skipped entirely by clear_site_home. Its contents are cleared by
+                    # unmount_tmpfs_without_save (called before clear_site_home), and the directory
+                    # itself must remain as it may be a Docker-managed tmpfs mount point (EBUSY).
+                    Directory(
+                        name="tmp",
+                        directories=frozenset(
+                            {
+                                Directory(
+                                    name="run",
+                                    files=frozenset({File("rrdcached.sock", content=b"")}),
+                                )
+                            }
+                        ),
+                    ),
+                    Directory(name=".restore_working_dir"),
+                }
+            ),
+        ),
     ],
 )
 def test_clear_site_home(directories: Dirs, files: Files, untouched: Dirs, tmp_path: Path) -> None: