Re: [Tracker] New command line tool tracker-files

[Jamie McCracken <jamiemcc blueyonder co uk>]

Gergan Penkov wrote:
> On Wed, 02 Aug 2006 16:22:15 +0100
> Jamie McCracken <jamiemcc blueyonder co uk> wrote:
>
> > I have added another command line tool "tracker-files" to cvs
> >
> > This can be used to retrieve all files of a service type or a mime
> > type.
> >
> > EG to return all files in service music:
> > tracker-files -s music
> >
> > to return all jpg files:
> > tracker-files -m image/jpeg
> Hi,
> it is working just great here, thank you for your hard work on
> this.
> I have a small problem again :) could the dbus-interface be started a
> little bit earlier, from I could gather from fast skimming the source,
> probably this is the culprit:
>         /* periodically poll directories for changes */ 
>         g_timeout_add_full (G_PRIORITY_LOW, 
>                             FILE_POLL_PERIOD,
>                             (GSourceFunc) start_poll,    
>                             NULL, NULL
>                            );
>
>
>
>         loop = g_main_loop_new (NULL, TRUE);
>
>         main_connection = tracker_dbus_init ();


not really. The timeout function is added to the mainloop and gets 
called periodically as a low priority item (every 15 minutes it polls) 
so there is no actual delay in starting the dbus interfaces.

You can freely search or whatever while it polls/indexes or watches files.


> The problem I have is that I've added not only my home, but also some
> external dirs, although my home directory alone would give also a hard
> time to tracker - it depletes all the inotify watches and begins
> polling, with the current setup with more than 7800 polls over the
> watches (but it will be not that different with only my home), so the
> real problem is that it needs quite a lot of time to begin
> responding to the dbus requests.

I have hardcoded the poll frequency at 15 minute intervals. This is 
probably too short for 7000+ directories so it would probably be 
continuously polling and slowing down the main thread (and consequently 
delaying requests to the  dbus interface)

I could make the poll interval hourly if that would help?

If I were you I would only watch the most frequently changing 
directories as you can call trackerd directory to index indiviudal branches

Also set NoWatchDirectories in ~/.Tracker/tracker.cfg to a list of roots 
that you dont want watched.

Hopefully longer term we will get kernel support for recursively 
watching log files for file changes to the entire FS and do away with 
polling and watches all together


--
Mr Jamie McCracken
http://jamiemcc.livejournal.com/