On Mar 24, 2008, at 2:39 PM, Christopher Browne wrote:
> Kenny Gorman <kgorman at hi5.com> writes:
>> I poked through the archives but didn't see any answers to this
>> question.
>>
>>
>> We have many databases that we replicate via slony (200+ GB each).
>> The problem is that the initial COPY is slow, and a single process.
>> We have a faster, lower level mechanism to clone an entire database
>> much faster, thus I would like to disable the initial truncate/COPY  
>> of
>> slony during initialization.  This approach would require a small
>> downtime to perform the last sync/quiesce, but our application can
>> tolerate that.  Essentially I would guarantee that the databases are
>> in sync, exactly, using a different mechanism (in this case disk  
>> level
>> snapshots).  We use this mechanism currently to clone databases, but
>> it would be nice to extend it for slony setup.  I looked through the
>> source, and it seems possible to change things to support this type  
>> of
>> initialization.
>>
>> The process I am thinking of is something like:
>>
>> start disk sync on A to B
>> shutdown A
>> sync last bits from A to B
>> startup A on different port ( no traffic except me )
>> startup B on different port ( no traffic except me )
>> subscribe set in slony w/o TRUNCATE/COPY
>> restart A on prod port
>> restart B on prod port
>> start slon processes
>
> Patches are in place to support this process in CVS HEAD, via a new
> pair of commands:
>
> CLONE PREPARE - which you would run somewhere near the beginning of
>                 the process; this indicates the plan for a new node
>                 to be created;
>
> CLONE FINISH - runs later, in effect, once the DB for B becomes
>                available.  It then does the "identity change"
>                to tell database "B" (which still thinks it's A)
>                that it is now "B".
>
> http://lists.slony.info/pipermail/slony1-commit/2008-January/002145.html
> http://lists.slony.info/pipermail/slony1-commit/2008-January/002146.html
> http://lists.slony.info/pipermail/slony1-commit/2008-January/002147.html
> http://lists.slony.info/pipermail/slony1-commit/2008-January/002148.html
> http://lists.slony.info/pipermail/slony1-commit/2008-January/002149.html
> http://lists.slony.info/pipermail/slony1-commit/2008-January/002150.html
> http://lists.slony.info/pipermail/slony1-commit/2008-January/002151.html
> http://lists.slony.info/pipermail/slony1-commit/2008-January/002152.html
> -- 
>

Christopher,

Thanks for pointing me in the right direction!  This should suffice  
for what we need, I will check it out.  Anyone using this option with  
success?

Thanks,
Kenny Gorman
DBA
www.hi5.com