Bug 138046

Summary:	Don´t run the "suse-updatedb" scripts when system is at high usage
Product:	[openSUSE] SUSE LINUX 10.0	Reporter:	Michael Stather <kontakt>
Component:	Basesystem	Assignee:	Andreas Schwab <schwab>
Status:	RESOLVED WORKSFORME	QA Contact:	E-mail List <qa-bugs>
Severity:	Enhancement
Priority:	P5 - None
Version:	Final
Target Milestone:	---
Hardware:	Other
OS:	Other
Whiteboard:
Found By:	Other	Services Priority:
Business Priority:		Blocker:	---
Marketing QA Status:	---	IT Deployment:	---

Description Michael Stather 2005-12-12 14:56:00 UTC

When using SuSE with a system which has a slow harddisk, like a notebook, those "suse-updatedb" scripts which are run by a cronjob slow down the system for about half an hour which can be very annoying. I don´t know what they do but since I assume they update a database perhaps it could be changed so that the crobjob is started much more often, e.g. every 15 minutes. But noit the actual process is started but a helper script which checks whether the rebuilding has to be done (because the last successful update has been x days ago). If yes it´ll start the update but the update will only be done in idle state, so if the system load goes above a treshold the update is cancelled (the update should be done with a copy so the actual database isn´t corrupted in such a case). So it is tried again after 15 minutes. I think this (or a similar system) would be much better and most of all would guarantee a system ready to work at any time.

Comment 1 Andreas Schwab 2005-12-12 15:11:47 UTC

This needs to be fixed in the kernel, if ionice doesn't work as designed.

Comment 3 Jens Axboe 2005-12-13 14:52:50 UTC

The update isn't cancelled, the io is just slowed down (very temporarily suspended, actually) until the disk is idle again. This does mean that the update itself can take a long time, but it should not be a resource intensive process. The atime updates are not done at idle priority though, they will go out as normal writes. So that can skew the idleness a little. Also note that idle io only works for the CFQ io scheduler, so if you changed that then the updatedb will run at the normal priority.

We can't really improve the idle io class a lot, there are a lot of potential for livelocking the system because of priority inversion problems.

You don't want to run the updatedb script more often, it will pollute the page cache a lot and the atime updates will hurt you a lot.

Assigning this back, don't know where it should go. If the updatedb script needs changing, it's not a kernel issue of course.

Comment 4 Lars Marowsky-Bree 2005-12-13 14:57:36 UTC

OK. Not a kernel bug, then. Please either modify the cronjob or close the bug, if this is not deemed necessary.

Comment 5 Andreas Schwab 2005-12-13 15:29:30 UTC

I don't see any of these effects on my notebook.

Comment 6 Michael Stather 2005-12-13 17:38:31 UTC

Why is this set to resolved?
Even if you don´t like my proposal it´s really annoying on older Computers, even if it doesn´t happen "on your notebooks".
I´ve a Compaq EVO N 600C with a Pentium III 1000 Mhz. This should be and is enough for running SuSE, but if you do a "find" what is what the scripts do, the system becomes unusable for almost half an hour. This is not very good when being at work!