Re: When does LGWR flush the ebuffer?

There is no lock while the log buffer is being flushed. If there were, then transactions would come to a halt until the log buffer is flushed. Transactions are allowed to write to the log buffer while LGWR is writing to the online redo logs.

HTH,
Brian

Peter wrote:
>
> On Tue, 22 Apr 2003 23:30:44 +1000, "Richard Foote"
> <richard.foote_at_bigpond.com> wrote:
>
> >"Peter" <peter_at_nomorenewsspammin.ca> wrote in message
> >news:kur9avoueoqsaaepunact06qcraj7vla7o_at_4ax.com...
> >>
> >> The LGWR flushes the buffer every 3 seconds, or when there is a
> >> commit.
> >> Does it ever flush it when the buffer is 1/3 or totally full?
> >> If it flushes it when buffer 1/3 full, then it will never get filled
> >> up?
> >>
> >Hi Peter,
> >
> >The following wakes up LGWR:
> >
> >- a commit
> >- redo log buffer reaching 1/3 full or 1M (whichever comes first)
> >- when DBWR wakes up (such as every 3 seconds, checkpoint, etc.)
> >
> >If you have a bucket with a hole in it which you cover with your finger
> >until the bucket is 1/3 full, does that mean the bucket will never get full
> >?
> >
> >Will that kinda depends on how "turned on" the tap is and how quickly water
> >enters the bucket. If water enters at a faster rate than the water leaves
> >the bucket via the hole, then yes, eventually the bucket can and will fill
> >up.
> >
> >Same for the redo log buffer. In fact, it's potentially a little worse in
> >that until the 1/3 is written out, Oracle can't overwrite that region of the
> >buffer. So if your buffer is sized too small, this event can eventuate.
> >
>
> Now I see why. But is there any type of lock that Oracle places on the
> buffer while redo log is being flush.
> It looks like filling continues while flushing is being done?
>
> Thanks.
>

> Received on Tue Apr 22 2003 - 15:07:09 CDT