On Mon, Apr 19, 2010 at 1:07 AM, Johan De Meersman <vegivamp@stripped> wrote:
> On Mon, Apr 19, 2010 at 6:48 AM, Rob Wultsch <wultsch@stripped> wrote:
>> And if your slave's IO lags badly enough this will hose you. Further
> True, but if you remove logs that haven't been transferred, yet, you lose
> your slave.
> Transfer of logs shouldn't be lagging that much, really, unless you're
> replicating over some POTS line. Don't forget that log transfer doesn't have
> to wait for processing on the slave.
Agreed. 7 days is an absurd length of time for the io thread to lag.
However, if someone has setup replication but not monitoring it, 7
days isn't that long. The behavior of mysql after a crash (that is,
breaking the io thread) makes me weary of suggesting to a
inexperienced user that they should turn on this features.
In an ideal world the user should determined how many days of backups
they need and further how much pitr. It might be very sane to say 30
days of daily backups with 7 days of pitr. One way or another this
should be a conscious decision, not a copy and paste from a mailing