Re: [Firebird-devel] Possible improvement to gbak restore?

[Ann W. H. <aha...@ib...>]

At 05:06 PM 7/31/2001 +1000, Danny Mavromatis wrote:
>
>Could I request that we modify IDX_create_index in jrd/idx.c to not do the
>the referential integrity check when creating a foreign key index on a
>database restore operation.

>+  else if (attachment && (attachment->att_flags & ATT_gbak_attachment))
>+      /* Don't bother checking Referential Integrity on a gbak
>+         restore (it must have been okay at backup time) */
>+     result = idx_e_ok;
>    else
>       result = check_partner_index (tdbb, relation, record,
>                       transaction, idx, partner_relation, partner_index_id);

I don't like this change. Nor, for that matter, do I like the
ATT_gbak_attachment option - it opens the door for too many hacks
that will tend to make gbak (and other code that uses that attachment
option) behave differently from normal database connections.  It's
one thing to have secret attachment options in closed code.  In open
code, it's a catastrophe waiting to occur.

Further, this check is done once for each index, so it's relatively
cheap.

>The restore time from backup on HP-UX systems is very, very slow

That may well be, and the problem may well be in building indexes.
However, this change will not fix that problem.   A more fruitful
place to look may be increasing sort speed by adjusting the temp
file handling.



Regards,

Ann
www.ibphoenix.com
We have answers.