Initial install, RAID; use --assume-clean with mdadm, and prompt for chunk size

Bug #77470 reported by Michael Milligan
12
Affects Status Importance Assigned to Milestone
mdcfg (Ubuntu)
Triaged
Wishlist
Unassigned
Declined for Feisty by Henrik Nilsen Omma
Declined for Gutsy by Henrik Nilsen Omma

Bug Description

Binary package hint: base-installer

Using Edgy to build a couple servers, amd64 and i386... found that I had to fool around in a shell to adjust the chunksize of the RAID 5 arrays I built on these servers, and also found that mdadm is being called without the --assume-clean flag when telling the installer to create new file systems on the array, which does not bypass array reconstruction (which is very annoying). The last bit is problematic because when it tells you to reboot and you do it, the arrays will start reconstruction all over again after reboot, slowing down the whole install process.

It would be extremely helpful to be prompted for chunksize, with a blurb of course that says if you don't know what this is for, just use the default (64). I have big arrays that work much better (30-40% faster) with 256 chunksize and would have preferred not having to shell out and umount/mount stuff and continue.

Revision history for this message
Michael Milligan (milli) wrote :

I also don't know if the "-R stride=X" parameter is being passed to mke2fs either, but that also needs to be adjusted based on the chunksize.

Revision history for this message
Christoph Lechleitner (lech) wrote :

I just had the same expirience with gutsy.
I absolutely support that the chunksize should be configurable in base-installer's partitioning section, and that the mke2fs call should be checked for optimization.

Revision history for this message
Gaetan Nadon (memsize) wrote :

Thank you for your suggestion. However, the changes you are requesting aren't really a bug and require more discussion, which should be done on an appropriate mailing list or forum. http://www.ubuntu.com/support/community/mailinglists might be a good start for determining which mailing list to use.
BugSquad

Changed in mdcfg:
status: New → Invalid
Revision history for this message
Colin Watson (cjwatson) wrote :

Gaetan: Why isn't this a bug? I don't think it's a good idea to select exactly those bugs for closing where the submitter clearly put effort into producing a well-thought-out proposal. Please stop doing this.

I notice that the mdadm(8) manual page says:

       --assume-clean
              Tell mdadm that the array pre-existed and is known to be
              clean. It can be useful when trying to recover from a
              major failure as you can be sure that no data will be
              affected unless you actually write to the array. It can
              also be used when creating a RAID1 or RAID10 if you want
              to avoid the initial resync, however this practice --
              while normally safe -- is not recommended. Use this
              only if you really know what you are doing.

Thus I rather feel we shouldn't do this by default.

A chunk size prompt seems reasonable, perhaps only in expert mode.

Changed in mdcfg:
importance: Undecided → Wishlist
status: Invalid → Triaged
Revision history for this message
Michael Milligan (milli) wrote :

This doesn't much matter to me anymore, I use FreeBSD now for servers when I have a choice.

Revision history for this message
Stephen Warren (srwarren) wrote :

At least for RAID-1, and I assume probably for other RAID levels, using --assume-clean when creating the arrays would be an extremely bad idea. The md feature to check whether the two mirrors of the array are still identical would fail horribly if they never started out being identical.

To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Duplicates of this bug

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.