Hi,
Why was the improve-relatime.patch dropped from F10's kernel? F9 had relatime on by default using this patch. I first thought the upstream kernel had the patch now, but /proc/mounts does not show relatime in F10.
I found this bug, looks related:
mkinitrd can't deal with the "relatime" mount option https://bugzilla.redhat.com/show_bug.cgi?id=430280
More info: http://kerneltrap.org/node/14148 http://people.redhat.com/mingo/relatime-patches/improve-relatime.patch
2008/12/28 Ricardo Argüello ricardo@fedoraproject.org:
Hi,
Why was the improve-relatime.patch dropped from F10's kernel? F9 had relatime on by default using this patch. I first thought the upstream kernel had the patch now, but /proc/mounts does not show relatime in F10.
I found this bug, looks related:
mkinitrd can't deal with the "relatime" mount option https://bugzilla.redhat.com/show_bug.cgi?id=430280
More info: http://kerneltrap.org/node/14148 http://people.redhat.com/mingo/relatime-patches/improve-relatime.patch
You should send this to the fedora-kernel list. Its non-subscription and might get a better review.
Regards
Ricardo Argüello wrote:
Why was the improve-relatime.patch dropped from F10's kernel? F9 had relatime on by default using this patch. I first thought the upstream kernel had the patch now, but /proc/mounts does not show relatime in F10.
I found this bug, looks related:
mkinitrd can't deal with the "relatime" mount option https://bugzilla.redhat.com/show_bug.cgi?id=430280
Speaking of, this looks to be... how old? Any chance of getting it fixed? (Especially with SSD's becoming more popular, for which it makes particular sense to request relatime?)
I mean... if you don't know about this bug and add relatime to fstab, you've just made it impossible to upgrade the kernel. And the work-around (edit fstab to remove relatime, 'yum update kernel', edit fstab to put back relatime) isn't very friendly.
Should I report the dropped improve-relatime.patch as a bug? The mkinitrd is a whole different thing.
What I wanted was the improve-relatime.patch to be included in F10, as it is included in F9 by default: http://people.redhat.com/mingo/relatime-patches/improve-relatime.patch
I reported it as a bug:
Add improve-relatime.patch to F10 and rawhide https://bugzilla.redhat.com/show_bug.cgi?id=479052
On Tue, Jan 06, 2009 at 11:33:57AM -0600, Matthew Woehlke wrote:
I mean... if you don't know about this bug and add relatime to fstab, you've just made it impossible to upgrade the kernel. And the work-around (edit fstab to remove relatime, 'yum update kernel', edit fstab to put back relatime) isn't very friendly.
relatime is supported in the stock kernel, but without the heuristic that avoids tmpreaper deleting your files at inappropriate times. It should cause failure.
On Tue, Jan 06, 2009 at 02:16:40PM -0500, Ricardo Argüello wrote:
On Tue, Jan 06, 2009 at 11:33:57AM -0600, Matthew Woehlke wrote:
relatime is supported in the stock kernel, but without the heuristic that avoids tmpreaper deleting your files at inappropriate times. It should cause failure.
Would the "improved relatime" patch solve that problem?
Huh. I actually meant "shouldn't cause failure" there, in terms of it still being a valid mount option. The improved relatime code adds that heuristic, which avoids the tmpreaper failure. I'm trying to get it upstream, at which point I'll backport it - I'm not enthusiastic about carrying it around as a custom modification forever.
On Wed, Jan 07, 2009 at 12:54:52AM +0000, Matthew Garrett wrote:
Huh. I actually meant "shouldn't cause failure" there, in terms of it still being a valid mount option. The improved relatime code adds that heuristic, which avoids the tmpreaper failure. I'm trying to get it upstream, at which point I'll backport it - I'm not enthusiastic about carrying it around as a custom modification forever.
This got merged for 2.6.30 and I've backported it to the rawhide tree. It should also turn up in the 2.6.29 kernel update for F10, but possibly won't be enabled by default there while we work out how to handle the userland transition.
Thanks Matthew, this is perhaps one of the most important performance tweaks for Fedora 11 and beyond. We need to add this to the Release Notes or something like that!
Should we close the bug now?
On Thu, Mar 26, 2009 at 3:36 PM, Matthew Garrett mjg@redhat.com wrote:
On Wed, Jan 07, 2009 at 12:54:52AM +0000, Matthew Garrett wrote:
Huh. I actually meant "shouldn't cause failure" there, in terms of it still being a valid mount option. The improved relatime code adds that heuristic, which avoids the tmpreaper failure. I'm trying to get it upstream, at which point I'll backport it - I'm not enthusiastic about carrying it around as a custom modification forever.
This got merged for 2.6.30 and I've backported it to the rawhide tree. It should also turn up in the 2.6.29 kernel update for F10, but possibly won't be enabled by default there while we work out how to handle the userland transition.
-- Matthew Garrett | mjg59@srcf.ucam.org
-- fedora-devel-list mailing list fedora-devel-list@redhat.com https://www.redhat.com/mailman/listinfo/fedora-devel-list
Ricardo Argüello wrote:
Thanks Matthew, this is perhaps one of the most important performance tweaks for Fedora 11 and beyond. We need to add this to the Release Notes or something like that!
Sure. Added to
https://fedoraproject.org/wiki/Documentation_Kernel_Beat
Let me know if you need more info added.
Rahul
Matthew Garrett wrote:
On Tue, Jan 06, 2009 at 11:33:57AM -0600, Matthew Woehlke wrote:
I mean... if you don't know about this bug and add relatime to fstab, you've just made it impossible to upgrade the kernel. And the work-around (edit fstab to remove relatime, 'yum update kernel', edit fstab to put back relatime) isn't very friendly.
relatime is supported in the stock kernel, but without the heuristic that avoids tmpreaper deleting your files at inappropriate times. It shouldn't cause failure.
As per the bug mentioned in the OP, it may be coming from around initrd and not the kernel per-se. I intend to check next time a kernel update is available.
What does tmpreaper normally reap? If it is only /tmp, then I don't care about the missing heuristic ;-).
-
Christopher Brown -
Matthew Garrett -
Matthew Woehlke -
Rahul Sundaram -
Ricardo Argüello