Fixes for DRBD >= 8.4.

[mbakke]

This is a cleaned-up version of #1296, taking the comments from @apoikos into account. I have tested this with drbd-utils 9.13.1 and kernel module version 8.4.11 (Linux 5.7.6).

More testing very welcome!

Cc @vladimir-ipatov

[rbott]

I have successfully tested this with a full DRBD cluster QA run on Debian Buster (DRBD 8.4.10, drbd-utils 9.5.0) and Ubuntu Focal Fossa (DRBD 8.4.11, drbd-utils 9.11.0). I will also run this through Debian Bullseye tomorrow.

However, the error handling the in the retry code seems wrong, but that is already an open discussion/code comment:

ganeti/lib/storage/drbd.py

Lines 407 to 421 in b5663b4

	def _WaitForMinorSyncParams():
	"""Call _SetMinorSyncParams and raise RetryAgain on errors.
	"""
	if self._SetMinorSyncParams(minor, self.params):
	raise utils.RetryAgain()
	else:
	return []
	if self._NeedsLocalSyncerParams():
	# Retry because disk config for DRBD resource may be still uninitialized.
	try:
	utils.Retry(_WaitForMinorSyncParams, 1.0, 5.0)
	except utils.RetryTimeout:
	base.ThrowError("drbd%d: can't set the synchronization parameters: %s" %
	(minor, utils.CommaJoin(sync_errors)))

We need to sort this out but other than that, LGTM.

[mbakke]

@rbott thanks a lot for reviewing, I think the error handling issue has been resolved with the latest revision.

[apoikos]

Thanks to both, @vladimir-ipatov and @mbakke!

In principle it looks good, however I'd like to link this PR to specific issues. Given that DRBD 8.4 generally works without this PR, what is it exactly that we are fixing here? Neither this PR, nor #1296 point to specific issues, and we need that rationale at least for the changelog. I'm not saying these changes are unnecessary, I just want a documented justification for them :)

[apoikos]

Additionally checking that version["k_major"] == 8 would make the code more robust.

[vladimir-ipatov]

Thanks to both, @vladimir-ipatov and @mbakke!

In principle it looks good, however I'd like to link this PR to specific issues. Given that DRBD 8.4 generally works without this PR, what is it exactly that we are fixing here? Neither this PR, nor #1296 point to specific issues, and we need that rationale at least for the changelog. I'm not saying these changes are unnecessary, I just want a documented justification for them :)

Hello!

Main issue is:
Without this fixes ganeti with drbd 8.4 will not assemble drbd device if one of two nodes don't have lvm volumes for it (diskless).
According to my information, there is no issues on github about it.

It was mentioned in my pull request description (but may be uncertainly :) )"
"fixes for drbd >= 8.4: add drbdsetup new-device and new-mirror for network part assembling of drbd (needed for diskless states of drbd)"

May be there is some other minor issues, I don't remember.

[apoikos]

Merged, thanks to both of you!