Retry the partition commit when udev holds the disk

Full-disk installs flaked with "Partition(s) 1, 2 on /dev/vda have been
written, but we have been unable to inform the kernel of the change":
archinstall's disk wipe and parted's own partition registrations fire
udev probes, and when one is still holding the disk mid-commit, the
BLKPG sync fails and the install halts. The table is written correctly
when this happens — only the kernel's view is stale — so settle udev
before partitioning and redo the filesystem operations (wipe, partition,
format) when parted throws exactly this error. Recoveries are logged so
they show up in the support log.

Reproduced headlessly by booting the ISO on a serial console and driving
the real installer: stock failed 1 in 5 full-disk cycles at this exact
phase, with an audit watch confirming nothing else writes the disk. With
the fix, 13 of 13 cycles got cleanly through partitioning.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
David Heinemeier HanssonandClaude Fable 5 committed 2026-07-19 20:58:39 -07:00
1 parent c3f9743847
commit 04aafcb781
1 file changed
+26 -4
@@ -3,7 +3,7 @@
ONLY this module imports from archinstall. Everything else uses these helpers.
If archinstall's API churns, the blast radius is contained here.
Tested against archinstall 4.3 (Python 3.14).
Tested against archinstall 4.4 (Python 3.14).
The canonical call sequence (mirrored from archinstall.scripts.guided.py) is:
@@ -44,7 +44,7 @@ from typing import Iterator
from archinstall.lib.args import ArchConfig, ArchConfigHandler
from archinstall.lib.authentication.authentication_handler import AuthenticationHandler
from archinstall.lib.disk.filesystem import FilesystemHandler
from archinstall.lib.disk.utils import get_parent_device_path, get_unique_path_for_device
from archinstall.lib.disk.utils import get_parent_device_path, get_unique_path_for_device, udev_sync
from archinstall.lib.hardware import SysInfo
from archinstall.lib.installer import Installer
from archinstall.lib.mirror.mirror_handler import MirrorListHandler
@@ -52,6 +52,8 @@ from archinstall.lib.models import Bootloader
from archinstall.lib.models.device import DiskLayoutType, EncryptionType
from archinstall.lib.models.users import User
from .ui import info
def load_arch_config(config_path: Path, creds_path: Path) -> ArchConfigHandler:
"""Build an ArchConfigHandler from on-disk JSON.
@@ -83,10 +85,30 @@ def make_mirror_handler(offline: bool = True) -> MirrorListHandler:
def perform_filesystem_operations(arch_config: ArchConfig) -> None:
"""Partition, format, encrypt. archinstall's FilesystemHandler is its own
object (separate from Installer) so we run it before opening the
Installer context manager."""
Installer context manager.
parted's partition-table commit races udev: the disk wipe and the first
BLKPG partition registration trigger probes that briefly hold the disk
open, and the kernel then refuses to register the remaining partitions
(_ped.IOException: "Partition(s) ... have been written, but we have been
unable to inform the kernel"). The table is written correctly when that
happens — only the kernel's view is stale — so settle udev and redo the
operations. The wipe/partition/format sequence is idempotent."""
if not arch_config.disk_config:
raise RuntimeError("disk_config missing from arch config")
FilesystemHandler(arch_config.disk_config).perform_filesystem_operations()
handler = FilesystemHandler(arch_config.disk_config)
attempts = 3
for attempt in range(1, attempts + 1):
udev_sync()
try:
handler.perform_filesystem_operations()
return
except Exception as exc:
if attempt == attempts or "unable to inform the kernel" not in str(exc):
raise
info(f"› partition commit lost a udev race (attempt {attempt}/{attempts}); retrying")
@contextmanager