On 3/3/2005 12:19 PM, Darcy Buskermolen wrote:

> On Thursday 03 March 2005 08:57, Josh Berkus wrote:
>> Here's a hypothetical question:
>>
>> Say that we have a master and 2 slaves, and the master has Fsync = False.  
>> If there is a power-out in the data center, and there is data
>> loss/corruption on the master, is there any way in which that data loss
>> could cascade to the slaves?
> 
> No, because Slony is a trigger based replication engine, it never really truly 
> cares if the transaction really made it to disk, only that the trigger was 
> "fired".  Now if the event was never actually confirmed by a subscriber, (ie 
> never replicated)  it's 10000000% possible that it's lost in that disk 
> corruption someware in the event_log tables.

A real on-disk DB corruption due to fsync-off can have all sorts of 
funny side effects. Who knows if the log you see afterwards consists of 
full, consistent and contiguous transactions? If that crash removed one 
out of 3 log rows for a transaction (because the other 2 were written 
due to some cache eviction and unfortunately made it to disk), you end 
up replicating a partial transaction.


Jan

-- 
#======================================================================#
# It's easier to get forgiveness for being wrong than for being right. #
# Let's break this rule - forgive me.                                  #
#================================================== JanWieck at Yahoo.com #