[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: (ITS#5798) MMR/mirror mode circumvents schema check



h.b.furuseth@usit.uio.no wrote:
> hyc@symas.com writes:
>> IMO defaulting schemacheck off for replication is fine because
>> replicated entries will have been checked on the master.
> 
> What if two simulaneous, valid changes to the same entry on two masters
> produce a schema error when combined?  I haven't been tracking MMR -
> does the entry end up with a schema error, or does MMR consider two
> changes to the same entry a conflict and fail to replicate, or...?

AFAIK, the last entryCSN should win.  Something like:

          MM1             MM2

time 0   csn0            csn0           (same)

time 1   csn0+1    ->    -

time 2   -         <-    csn0+2

time 3   csn0+1,csn0+2   csn0+2,csn0+1  (one wins, I assume the latest)

At all times, each entry is sane.  The one that wins is sane, the one 
that loses will be replaced by the one that wins.

The reason for this ITS and other stuff that will come shortly is that 
I've been asked to audit MMR, by designing a suite of tests that check 
functionality and reliability of MMR under possibly reproducible 
conflict, concurrency and other critical conditions.

I'm highlighting a series of issues.  I won't post ITSes or other 
notices until I find out whether it's slapd's or my tests' fault.

p.


Ing. Pierangelo Masarati
OpenLDAP Core Team

SysNet s.r.l.
via Dossi, 8 - 27100 Pavia - ITALIA
http://www.sys-net.it
-----------------------------------
Office:  +39 02 23998309
Mobile:  +39 333 4963172
Fax:     +39 0382 476497
Email:   ando@sys-net.it
-----------------------------------