platform: determine hostname / fqdn / hostid lazily, refs #9470

Determining the fqdn issues a DNS query. That happened at import time, so
EVERY borg process paid for it - which can take very long on hosts whose
own hostname does not resolve (~35s per process on github's macOS runners,
which made the macOS test jobs take hours: each remote test spawns ~10
borg serve / client processes).

Replace the module-level hostname / fqdn / hostid attributes with
get_hostname() / get_fqdn() / get_hostid() functions. A cached
_gethostname() keeps all derived values consistent for the process
lifetime; get_fqdn() is additionally cached because it issues the DNS
query. Processes that never use them no longer resolve anything at all,
on any platform. get_hostname() never needs DNS, and a set BORG_HOST_ID
avoids the fqdn lookup entirely.

Co-Authored-By: Claude Opus 5 <[email protected]>

src/borg/archive.py

@@ -50,7 +50,7 @@ from .helpers import utcnow
 from .lrucache import LRUCache
 from .patterns import PathPrefixPattern, FnmatchPattern, IECommand
 from .item import Item, ArchiveItem, ItemDiff
-from .platform import acl_get, acl_set, set_flags, get_flags, swidth, hostname
+from .platform import acl_get, acl_set, set_flags, get_flags, swidth, get_hostname
 from .remote import cache_if_remote
 from .repository import Repository, LIST_SCAN_LIMIT
 
@@ -636,7 +636,7 @@ Utilization of max. archive size: {csize_max:.0%}
             'comment': comment or '',
             'items': self.items_buffer.chunks,
             'cmdline': sys.argv,
-            'hostname': os.environ.get('BORG_HOSTNAME') or hostname,
+            'hostname': os.environ.get('BORG_HOSTNAME') or get_hostname(),
             'username': os.environ.get('BORG_USERNAME') or getuser(),
             'time': start.strftime(ISO_FORMAT),
             'time_end': end.strftime(ISO_FORMAT),

src/borg/helpers/parseformat.py

@@ -205,13 +205,14 @@ def format_line(format, data):
 
 def replace_placeholders(text, overrides={}):
     """Replace placeholders in text with their values."""
-    from ..platform import fqdn, hostname, getosusername
+    from ..platform import get_fqdn, get_hostname, getosusername
+    fqdn = get_fqdn()
     current_time = datetime.now(timezone.utc)
     data = {
         'pid': os.getpid(),
         'fqdn': fqdn,
         'reverse-fqdn': '.'.join(reversed(fqdn.split('.'))),
-        'hostname': os.environ.get('BORG_HOSTNAME') or hostname,
+        'hostname': os.environ.get('BORG_HOSTNAME') or get_hostname(),
         'now': DatetimeWrapper(current_time.astimezone(None)),
         'utcnow': DatetimeWrapper(current_time),
         'unixtime': int(current_time.timestamp()),

src/borg/platform/__init__.py

@@ -11,7 +11,7 @@ from .base import acl_get, acl_set
 from .base import set_flags, get_flags
 from .base import SaveFile, SyncFile, sync_dir, fdatasync, safe_fadvise
 from .base import swidth, API_VERSION
-from .base import process_alive, get_process_id, local_pid_alive, fqdn, hostname, hostid
+from .base import process_alive, get_process_id, local_pid_alive, get_hostname, get_fqdn, get_hostid
 
 OS_API_VERSION = API_VERSION
 

src/borg/platform/base.py

@@ -1,4 +1,5 @@
 import errno
+import functools
 import os
 import socket
 import uuid
@@ -276,20 +277,33 @@ def getfqdn(name=''):
     return name
 
 
-# For performance reasons, only determine hostname / FQDN / host ID once.
-# XXX This sometimes requires live internet access for issuing a DNS query in the background.
-hostname = socket.gethostname()
-fqdn = getfqdn(hostname)
-# Some people put the FQDN into /etc/hostname (which is wrong; it should be the short hostname).
-# Fix this (do the same as "hostname --short" CLI command does internally):
-hostname = hostname.split('.')[0]
+@functools.cache
+def _gethostname():
+    """socket.gethostname(), cached so all derived values are consistent for the process lifetime."""
+    return socket.gethostname()
 
-# uuid.getnode() is problematic in some environments (e.g., OpenVZ, see #3968) where the virtual MAC address
-# is all-zero. uuid.getnode falls back to returning a random value in that case, which is not what we want.
-# Thus, we offer BORG_HOST_ID where a user can set an own, unique ID for each of his hosts.
-hostid = os.environ.get('BORG_HOST_ID')
-if not hostid:
-    hostid = f'{fqdn}@{uuid.getnode()}'
+
+def get_hostname():
+    """Return the (short) hostname of this machine."""
+    # Some people put the FQDN into /etc/hostname (which is wrong; it should be the short hostname).
+    # Fix this (do the same as "hostname --short" CLI command does internally):
+    return _gethostname().split('.')[0]
+
+
+@functools.cache
+def get_fqdn():
+    """Return the fully qualified domain name of this machine."""
+    # Cached: this issues a DNS query, which can take very long on hosts whose own
+    # hostname does not resolve (e.g., ~35s per query on github's macOS CI runners, see #9470).
+    return getfqdn(_gethostname())
+
+
+def get_hostid():
+    """Return an identifier that is unique for this machine (host)."""
+    # uuid.getnode() is problematic in some environments (e.g., OpenVZ, see #3968) where the virtual MAC address
+    # is all-zero. uuid.getnode falls back to returning a random value in that case, which is not what we want.
+    # Thus, we offer BORG_HOST_ID where a user can set an own, unique ID for each of his hosts.
+    return os.environ.get('BORG_HOST_ID') or f'{get_fqdn()}@{uuid.getnode()}'
 
 
 def get_process_id():
@@ -301,7 +315,7 @@ def get_process_id():
     """
     thread_id = 0
     pid = os.getpid()
-    return hostid, pid, thread_id
+    return get_hostid(), pid, thread_id
 
 
 def process_alive(host, pid, thread):

src/borg/platform/posix.pyx

@@ -42,8 +42,9 @@ def process_alive(host, pid, thread):
     always returns True, since there is no real way to check.
     """
     from . import local_pid_alive
-    from . import hostid
+    from . import get_hostid
 
+    hostid = get_hostid()
     assert isinstance(host, str)
     assert isinstance(hostid, str)
     assert isinstance(pid, int)

src/borg/testsuite/helpers.py

@@ -243,7 +243,8 @@ class TestLocationWithoutEnv:
             Location('ssh://user@host:/path')
 
     def test_omit_archive(self):
-        from borg.platform import hostname
+        from borg.platform import get_hostname
+        hostname = get_hostname()
         loc = Location('ssh://user@host:1234/repos/{hostname}::archive')
         loc_without_archive = loc.omit_archive()
         assert loc_without_archive.archive is None
@@ -262,7 +263,8 @@ class TestLocationWithEnv:
                "Location(proto='ssh', user='user', host='host', port=1234, path='/some/path', archive=None)"
 
     def test_ssh_placeholder(self, monkeypatch):
-        from borg.platform import hostname
+        from borg.platform import get_hostname
+        hostname = get_hostname()
         monkeypatch.setenv('BORG_REPO', 'ssh://user@host:1234/{hostname}')
         assert repr(Location('::archive')) == \
             f"Location(proto='ssh', user='user', host='host', port=1234, path='/{hostname}', archive='archive')"