sreadahead uses 100 % of the CPU in Karmic

[François Blondel]

Binary package hint: sreadahead

Hi,
Sreadahead uses 100 % of the CPU in Karmic. I haven't any SSD disk, and I use ext4.
Is this package installed by default ? Or did I installed it myselft by error ?
When I try to kill it with "Moniteur système 2.27.4" (french), this monitor crash.
What should I join to help ? dmesg ? ls /dev ?
Thanks

[Virtex (c-launchpad-virtex-org)]

I can confirm this bug as I've seen the same behavior. After a fresh boot, when I log in my CPU is running at 100% and the culprit is sreadahead. sreadahead then exits after a minute or so. If you want, you can disable sreadahead by editing the file at /etc/init/sreadahead.conf and commenting out the line that reads:

exec /sbin/sreadahead -t 0

I did this on my Thinkpad R61 (with a spinning hard drive) and my boot time dropped from 61 seconds to 49 seconds.

[Scott James Remnant (Canonical) (canonical-scott)]

This is quite normal, it takes a few minutes after the first boot to generate the lists - it's operating as a "very low priority" so shouldn't affect system performance.

If you don't let it finish, it will be slow every time you boot ;-)

[max (mikhmv)]

I have never interrupt it but after 27 reboots. it wasn't improved.......

[Simon Steinbeiß (ochosi)]

i can confirm this annoying behaviour in karmic64bit.
for some reason sreadahead keeps popping up after every reboot with 100%cpu for 10+ seconds. quite a nuisance.

(and yes, this wasn't the case before readahead was replaced by sreadahead)

[MichaelRobbert (mrobbert-gmail)]

I am seeing this on every reboot, not just the first. I just rebooted twice in a row without doing any updates. The system had been running smoothly overnight before the first reboot, then I let sreadahead finish before rebooting again and it started up again. If this is considered normal behavior then let us know and/or close the ticket, otherwise this looks like a bug that should be fixed.

[Steve Dodier-Lazaro (sidi)]

Scott, is the intented behaviour of sreadahead to run only once, or after each reboot ? If it is meant to run only once as you said, then there definately is a problem, because it keeps running after the first one.

[Scott James Remnant (Canonical) (canonical-scott)]

On Thu, 2009-09-17 at 15:25 +0000, Steve Dodier wrote:

> Scott, is the intented behaviour of sreadahead to run only once, or
> after each reboot ? If it is meant to run only once as you said, then
> there definately is a problem, because it keeps running after the first
> one.
>
It will usually profile on any reboot where you've done package installs
or upgrades that touch /etc/init or /etc/init.d

Scott
--
Scott James Remnant
<email address hidden>

[Anders Kaseorg (andersk)]

As it turns out, the 100% CPU time is spent bubble-sorting a list of readahead records. That’s idiotic. The C library comes with a standard qsort() function for a reason. This is easy to fix and we should do so.

[Anders Kaseorg (andersk)]

Patch: http://bazaar.launchpad.net/~anders-kaseorg/sreadahead/qsort/revision/44
Fixed packages: https://launchpad.net/~anders-kaseorg/+archive/ppa

[Scott James Remnant (Canonical) (canonical-scott)]

On Sun, 2009-09-20 at 17:52 +0000, Anders Kaseorg wrote:

> As it turns out, the 100% CPU time is spent bubble-sorting a list of
> readahead records. That’s idiotic. The C library comes with a standard
> qsort() function for a reason. This is easy to fix and we should do so.
>
Right, I'd looked into this before and switched to qsort() but then
broke other things.

Hadn't had a change to follow-up further.

Thanks for investigating and doing the patch, I'll certainly apply this
one! Good work!

Scott
--
Scott James Remnant
<email address hidden>

[Scott James Remnant (Canonical) (canonical-scott)]

I have uploaded the fix for this bug, however due to the freeze it may take a little while to show up, so I've also uploaded it to the ubuntu-boot PPA. Please test.

[Launchpad Janitor (janitor)]

This bug was fixed in the package sreadahead - 1.0-4

---------------
sreadahead (1.0-4) karmic; urgency=low

  * Replace custom bubble sort with qsort(), to avoid using 100% CPU
    for a long time after boot. LP: #421116, #434116.

 -- Anders Kaseorg <email address hidden> Sun, 20 Sep 2009 13:31:45 -0400

[Virtex (c-launchpad-virtex-org)]

This fix appears to have corrected the problem, at least for me. I no longer see the CPU at 100% when I log in after a major software upgrade.