[csw-users] Berkeley DB 4.7 vs. 4.4 issue

[Dagobert Michelsen]

Hi Marsha,

Am 28.08.2009 um 01:28 schrieb PETRY, MARSHA:
> I will go to topic-specific lists after this but since all my  
> software is from opencsw I want to see if anyone has an answer here  
> first...

The problem you describe sounds ugly, but there were no reports about
this problem during upgrade now. I don't know if it really connects
to your update or some other strange bug, but also asking on the
subversion list is definitely a good idea.

> My issue:
> I upgraded svn and trac a couple days ago via opencsw(mostly  
> "testing") and when I used svn I got a FATAL error (below this  
> email, sent to list a couple days ago)
>
> To fix that problem, using a link to old copy from someone on this  
> list, I did the solution:
> 1.  Built an old BDB4.4 version
> 2.  db_archive -s -h old_repo          /*BDB4.4*/
> 3.  db_dump (each name from #2)        /*BDB4.4*/
> 4.  svnadmin create new_repo --fs_type = bdb  /*newest CSWsvn ver.  
> 1.6.4*/

Is there a specific reason why you are still using bdb as fsfs
was preferred by some time now because there are several limitation
with bdb? Please don't get this as fingerpointing, but if you
have to reload the repo I would continue with fsfs from the start.

> 5.  db_load (each name from #2) to new_repo   /*BDB 4.7*/
> 6.  db_checkpoint -1 -v -h new_repo           /*BDB4.7*/
> 7.  db_verify    /*BDB4.7 ... verify each db... no error msgs*/
> 7.  svnserve ... /*svn 1.6.4*/
>
> Works fine checking out code... until I check something in.  The  
> checkin works but when I try to checkout again or update another  
> checked-out copy I get:  "Decompression of .svn diff data failed"
>
> /opt/csw/bin/svnadmin verify gives:
> * Verified revision 0.
> svnadmin: Checksum mismatch on representation '0':
>   expected:  effd2c5b6f40c05496ad82988093ec13
>     actual:  19b07c36abc7abbec99bf2bc432f8f64
>
> db_recover, db_verify give no errors.  Representation '0' is  
> scary... I found a suggestion that I should change the md5 sequences  
> on the representations.dump file and re-load but there are over  
> 30,000 of them.... ugh.

Well, in fact that is vi'ing one file and a few lines on the shell
with a few mugs of coffee in between. Not desirable, but definitely
doable.

> Is there anything obvious related to opencsw that I have missed?  Do  
> my packages look mis-matched?  Are there any opencsw files for svn/ 
> bdb/trac that are known to have an issue?
>
> CSWbdb                    4.7.25,REV=2009.07.01     not in catalog
> CSWbdb4                   4.2.52,REV=2009.07.28     not in catalog
> CSWbdb44                  4.4.20,REV=2009.07.28     not in catalog
> CSWbdb47                  4.7.25,REV=2009.03.09     not in catalog
> CSWsvn                    1.6.4,REV=2009.08.23      SAME
> CSWtrac                   0.11.5,REV=2009.08.03     SAME
> CSWzlib                   1.2.3,REV=2009.04.05      SAME

You are working against the testing catalog from here:
   <http://mirror.opencsw.org/testing.html>
Please be advised that this is highly disregarded for production  
environments
unless explicitly needed for specific tests.

> Thanks for any help. This is a showstopper problem for me and the  
> only way I can think to recover is get rid of all these opencsw  
> packages and use a different source for svn/trac.

I do understand your anger about the project, but I don't see how  
changing
package sources will bring your repository back. It may be best to fix  
the
checksum after making a copy of the current repo e. g. descibed here
   http://blackcompanystudios.co.uk/blog/user/mrcranky/237
and then dump and reload it to fsfs.


Best regards

   -- Dago