Re: [Tracker] Ready for merge

[Jamie McCracken <jamie mccrack googlemail com>]

On Thu, 2008-08-07 at 19:40 +0100, Martyn Russell wrote:
> Martyn Russell wrote:
> > Laurent Aguerreche wrote:
> > > Hello,
>
> Hi :)
>
> > There might be some tweaking that is needed. I personally don't see this
> > issue and my machine is not state of the art. I notice it is running,
> > when I am playing games or really loading my system, but it is no worse
> > than the old tracker was AFAICR.
> >
> > > Do you intend to fix these issues before the merge (notably the second
> > > one...)?
>
> I have to say, after replying to your mail I decided to test this again.
> It appears it is the crawling itself which is destroying the machine.
> The indexing is fine. The problem with the crawling is. It is cached the
> more times you do it. Also, sometimes I have found the machine is
> responsive, others, not.
>
> Tomorrow I will look into adding some sort of throttle to the crawler so
> it doesn't sit there and make your machine useless for 3 minutes.
> Phillip has a patch we can try for this. There are other means we use in
> the indexer too which I will play with.

if you are storing the inotify fd of the directory in the DB then make
sure its being written in a transaction spanning 1000's of updates. I
have not checked the code here but the original trackerd did this but i
image you changed this for GIO

Slowing down of other apps is always caused by disk writes and not disk
reads (the latter is only effectively throttled by ioprio) which leaves
apps needing I/O stalled

Ergo its essential that all updates to any of the dbs are done in bulk
transactions

jamie