Hi,
The bug isn't reproducible when using AFTER_COMMIT over AFTER_SYNC.
So the race happens somewhere between the after_flush hook and the after_sync
hook, likely because rotate is called after after_flush.
Unfortunately I don't have enough knowledge of the binlog code to do much more
debugg
Note that in the AFTER_COMMIT mode waiting for semi-sync ack happens
later than in AFTER_SYNC mode. So either this is not a race, or the
race is made possible by some semi-sync related code... Sounds pretty
strange...
On Sat, Aug 13, 2016 at 4:30 PM, Joseph Glanville wrote:
> Hi,
>
>
> The bug is
2 matches
Mail list logo