Parallel fsck is very slow with multiple partitions on one drive

Bug #619012 reported by Ivan Kozik
16
This bug affects 3 people
Affects Status Importance Assigned to Milestone
mountall (Ubuntu)
Confirmed
Undecided
Unassigned

Bug Description

Binary package hint: mountall

Before 10.04, Ubuntu fsck'ed every partition serially. In 10.04, multiple partitions on one drive are fscked in parallel. The change was made in:

https://bugs.launchpad.net/ubuntu/+source/mountall/+bug/491389

I have three ext4 partitions on a 640GB drive, which I fsck on every boot. This used to take ~1 minute, because ext4 fsck is very fast. In 10.04, it takes about 10 minutes. I don't think the "set their priorities so that thrashing is avoided" solution mentioned in the above bug is adequate (if it's working at all). Please disable the parallelism for multiple partitions, or make it obvious how to do so.

Revision history for this message
Ivan Kozik (ludios) wrote :

fwiw, I'm using Ubuntu 10.04 64-bit in VMWare Workstation 7.1 on a Server 2008 R2 host. Maybe that's why reads aren't being prioritized. Also, I'm not using LVM; my partitions are /dev/sda[1,6-8]

Changed in mountall (Ubuntu):
status: New → Confirmed
Revision history for this message
Ivan Kozik (ludios) wrote :

This bug no longer affects me because I moved to one partition, and then later moved to an SSD. It might affect others, of course, and I hope they comment.

Revision history for this message
Steve Langasek (vorlon) wrote :

Johan, I see that you were the author of the mountall code to use ioprio to prioritize the fsck runs. Can you comment on why this wouldn't have been working for Ivan?

Revision history for this message
gregrwm (gregrwm) wrote :

in the quest to make this better, perhaps we could be a tad circumspect, and have a peek at a description of doing this well:

http://www.freebsd.org/cgi/man.cgi?query=fstab&sektion=5

start reading at "The sixth field".

Revision history for this message
gregrwm (gregrwm) wrote :

on my box, currently, trusty is running fsck's simultaneously on the same disc. and even when this bug gets fixed, i've heard of hardware where there's no reliable way to automatically determine whether certain partitions share a spindle.

so, as per my previous comment, it would be kind and generous to allow the admin to override mountall's presumptions of what would be efficient.

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

Other bug subscribers

Remote bug watches

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