[Toolserver-l] Please respect the rev_deleted-field

[DaB.]

Hello,
At Tuesday 28 September 2010 14:00:46 DaB. wrote:
> On Mon, Sep 27, 2010 at 4:17 PM, DaB. <WP at daniel.baur4.info> wrote:
> > I tried it today. It morphs a 1.36s query (simple count-query on my
> > username on dewiki_p) into something I canceled after 2 minutes (I tried
> > both: if- selecting and where-clause).
> 
> What exactly did you try?  You'd need to change indexes on the
> underlying table for this to work acceptably.

I run a non-cached simple count-query on my user-name on dewiki (like I told 
in my email yesterday) and got a result in a sec.
Then I took our spare db-server… - oh wait, we have no spare db-server.
So I took 1 of the sql-s5-db-servers out of rotation… – oh wait we have only 1 
server for sql-s5.
So I just changed the view of revision (like I told in my email yesterday) in 
a way it hides rev_user_text if rev_deleted>0, to test the runtime of a non-
cached simple count. I canceled that after 1 minute or 2. Then I changed the 
view in the way that all rows if rev_deleted>1 are hidden (with a where-
clause) and run the non-cached  count again and canceled it again after a 
minute or two. Then I changed the view back to its original config and run my 
non-cached query again and it returned the count in a sec.

If you think that is done wrong or not trust me or my results, just try it 
yourself, use another db-server or anything. If you find a fast solution, I 
will be the last to not implement it.

> > And I don't see the point to allow users to see the archives-table, but
> > not the deleted revisions (the critical stuff is in the hidden-table
> > anyway).
> 
> I'm not clear on why toolserver users are allowed to see deleted
> revisions to begin with.  rev_deleted is supposed to supplant
> oversight, and that includes things that nobody is really supposed to
> see. 

We hide already the only really critical field: rev_comment, if the revision is 
deleted in any way. Every table on the toolserver can contain hidden stuff 
(even the user-table!), so the only real solution to let nobody see hidden 
stuff is to delete the database (and shut down the toolserver).

> I particularly don't understand why we hide stuff from the
> logging table but not the revision table.

The logging table contains more critical stuff then the revision-table, is not 
queried that often and is nearly never joined with other tables.


To let me be very clear: I (and I know River too) try to hide as much hidden 
stuff from the toolserver-users as possible (we try even to hide some data from 
the roots), but it is impossible to hide everything from them, without making 
the toolserver useless or really, really slow. So it is always a weighting. 
And as long as the toolserver-users follow our rules and think a bit, no 
critical/hidden stuff is visible to the world: And that's what matters.

Sincerly,
DaB.


-- 
Userpage: [[:w:de:User:DaB.]] — PGP: 2B255885
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: This is a digitally signed message part.
Url : http://lists.wikimedia.org/pipermail/toolserver-l/attachments/20100928/8c2c147c/attachment.pgp