Superfluous use of partprobe

[codefiles]

Pertinent partprobe manual information: [1]

partprobe is a program that informs the operating system kernel of partition table changes.

From pyparted, Disk.commit() also "informs the operating system [kernel] of the changes" to a partition table [2] therefore calling partprobe after DeviceHandler._perform_partitioning() is unnecessary.

archinstall/archinstall/lib/disk/device_handler.py

Lines 327 to 328 in f610632

	disk.deletePartition(part_info.partition)
	disk.commit()

archinstall/archinstall/lib/disk/device_handler.py

Lines 366 to 367 in f610632

	disk.addPartition(partition=partition, constraint=disk.device.optimalAlignedConstraint)
	disk.commit()

archinstall/archinstall/lib/disk/device_handler.py

Lines 523 to 529 in f610632

	for part_mod in filtered_part:
	# if the entire disk got nuked then we don't have to delete
	# any existing partitions anymore because they're all gone already
	requires_delete = modification.wipe is False
	self._perform_partitioning(part_mod, modification.device, disk, requires_delete=requires_delete)
	self.partprobe(modification.device.device_info.path)

A partprobe that is performed after formatting the partitions is unnecessary because formatting the partitions does not change the partition table.

archinstall/archinstall/lib/disk/device_handler.py

Lines 295 to 305 in f610632

	if enc_conf is not None and part_mod in enc_conf.partitions:
	self._perform_enc_formatting(
	part_mod.safe_dev_path,
	part_mod.mapper_name,
	part_mod.safe_fs_type,
	enc_conf
	)
	else:
	self._perform_formatting(part_mod.safe_fs_type, part_mod.safe_dev_path)
	lsblk_info = self._fetch_part_info(part_mod.safe_dev_path)

archinstall/archinstall/lib/disk/device_handler.py

Lines 377 to 389 in f610632

	def _fetch_part_info(self, path: Path) -> LsblkInfo:
	attempts = 3
	lsblk_info: Optional[LsblkInfo] = None
	self.partprobe(path)
	for attempt_nr in range(attempts):
	time.sleep(attempt_nr + 1)
	lsblk_info = get_lsblk_info(path)
	if lsblk_info.partn and lsblk_info.partuuid and lsblk_info.uuid:
	break
	self.partprobe(path)

Encrypting a partition does not change the partition table.

archinstall/archinstall/lib/luks.py

Line 165 in f610632

disk.device_handler.partprobe(self.luks_dev_path)

[1] https://man.archlinux.org/man/partprobe.8
[2] https://github.com/dcantrell/pyparted/blob/d5870bb7a5d512c17c92a4e4fe7e9a8e7c8c3528/src/parted/disk.py#L202-L205

[svartkanin]

Hmm I remember there was a reason we put the partprobe in place after partitioning, I can't fully remember now but I think there was an issue without it not loading the lsblk info correctly

[codefiles]

Regardless of whatever the intent was in the past, no evidence has been provided to justify any of the uses of partprobe outlined in the issue.

[qlyoung]

The patch for this issue reintroduced #1759. See #1759 (comment)

[codefiles]

I doubt that but I could be wrong, logs would help. This type of issue appears to occur when partitions get created with invalid disk geometry (see #2210). I discovered another issue that will cause problems with partitioning and will raise an issue in a moment.

[qlyoung]

I doubt that but I could be wrong

I readded a single call to partprobe() as part of the retry logic after calling lsblk fails and it worked - precisely the same as the original patch that resolved #1759.

logs would help

Getting logs for you would be rather inconvenient.

This type of issue appears to occur when partitions get created with invalid disk geometry (see #2210).

My geometry seems valid enough since the system installed with archinstall after my workaround subsequently booted and works perfectly. I checked manually and it seems fine. It's nothing complicated, just an ESP and a primary.

If I submit my patch, will you nack it? Just trying to gauge up front how much effort it's going to take for me to fix this.