Campbell Ritchie wrote:
A.J. Côté wrote: . . . I had to do a few crash recoveries.
. . . really tested and robust.
Please explain more. Were the crashes unrelated to blinlog? Otherwise, those two statements seem to contradict each other.
Absolutely, completely unrelated to blinlog itself.
When the master crashes at the storage level (for some hardware or other reason), you have to manually replay the binlogs on the master before starting to replicate again.
I have seen cases where we needed to delete duplicate rows on slaves in order for replication to continue.
But overall, we never lost any real time data using it yet.