Re: [Tracker] 0.12 & tracker-store tuning

[Martyn Russell <martyn lanedo com>]

On 15/01/12 19:27, Michael Steiner wrote:
> Hi,

Hello Michael,

> I'm using tracker 0.10.x for quite a while quite happily on Ubuntu
> Natty/Gnome 2 and recently upgraded to 0.12 (now running
> 0.12.9). While it function generally fine, the performance of
> tracker-store gives me a bit grief. Even after small changes in
> monitored files/dirs, tracker-store causes minutes of heavy I/O
> activity which causes quite some negative interference with other
> processes.  I do index a lot (~/.cache/tracker is 5.8GB) but i didn't
> change the index scope from what i did with 0.10 and the index size is
> basically the same as before. Yet with 0.10 I have not seen even close
> as much I/O.

Yes, we moved from an INI type file implementation in 
~/.config/tracker/*.cfg to GSettings, which is more of a database type 
approach. You can get some idea of how to configure with GSettings on 
our wiki page here:

  https://live.gnome.org/Tracker/Documentation/Debugging

The tracker-preferences application should also allow customisation.

> Is there any way to tune the performance of tracker-store (or maybe
> tracker-miner)?  0.12 seems to have removed a lot of the configuration
> parameters which exited for 0.10 in ~/.config/tracker/*cfg, e.g., for
> caching and throttling. Did they move elsewhere, are they gone or
> replaced with some other tuning parameters?

This has always been configurable with /usr/bin/tracker-preferences.

> In any case, any tips on how to improve the performance is highly
> appreciated.

I would also suggest upgrading if possible. We're currently doing stable 
releases of 0.16. and there have been many improvements since 0.12.

> -michael-
>
> PS: two additional questions:
>
> - Is there a way to index also bzip-compressed files? It seems right
>    now compressed files are only supported in a limited form? Is there
>    any thoughts/work to make a ``stackable'' decompression filter which
>    passes the output to the standard miners?

Sadly, this is an area we never got around to working on. You could 
store the data in Tracker with the ontology, but we don't extract the 
data unfortunately.

> - Is there a way to ignore files or directories based on full (or
>    partial) path instead of just the directory local name?

Yes there is. Again, using tracker-preferences, you can provide either 
Globs (e.g. *.txt) or exact paths (e.g. /etc/hosts). When I say path 
here, I also don't mean just directories, but files too.

The UI has a way to specify this for files and directories.

Happy to help, :)

--
Regards,
Martyn

Founder and CEO of Lanedo GmbH.