But I still not understand why ?
In safety-mode, principal do not applied change on datafile until it's write in transaction log of the mirror.
So, if principal crashed, the transaction not written to the mirror will be rollback on the old principal when it will come back.
So the data will not be lost because they will not be applied on old principal.
missing something ?
regards,
There are a couple of reasons why transactions would not have come over I can think of. One is if the mirror server was in a suspended state when the primary failed and then forced to become the primary. In that scenario there are commands that could have applied on the primary that did not apply on the mirror.
Another reason could be because db mirroring works in near real-time using queues. Depending on the amount of traffic and the latency between the primary and mirror servers I can see a scenario where commands are not yet received at the mirror before failover. I believe Paul Randal did an article on this for TechNet magazine.
Some time later...
Yep, he did.
http://technet.microsoft.com/en-us/magazine/hh334997.aspx
Joie Andrew
"Since 1982"