Ian
You are partly right. When you issue the alter tablespace begin backup
command the file headers are frozen with the latest checkpoint SCN. In other
words the file is checkpointed. After this oracle is not concerned about
what is being written to the file since every write is fully logged in the
redo log file and full block images are written. And I believe when you do
the split the filesystem headers are synced up too. We are in the process of
implementing such a scheme and will let you know once we are done
Murali
"Ian Nash" <iannash_at_au1.ibm.com> wrote in message
news:3bc3b5f4.0201312114.2161a5ed_at_posting.google.com...
> Murali,
>
> Thanks for the clarification about "alter database suspend" in Oracle
> 8i. I did not know that this command is in 8i.
> With using this command to suspend writes to the DB - what affect does
> this have on applications using the DB ? Do they hang until the
> resume, or are writes cached within the DB Buffer area in memory ?
> With this command used before the split takes place, I believe there
> is 100% assurance that the mirror can be split without any chance of
> loss of data - as long as the mirrors being split only contain Oracle
> data and metadata. In particular, my concern is at the UNIX filesystem
> level, writes are still occurring (to the Oracle redo logs, dbf
> files).
> A split of the mirror while the writes are occurring could result in
> undetected data loss in the mirror copy of the files, as data may not
> be completely written (ie. a half-filled block, for instance). This
> would not be detected unless a dbverify is run on the entire mirror
> copy of the database, as the database may well start and recover
> itself without errors, and the filesystem may pass an fsck correctly -
> but data may be lost. Presumably, the synchronous nature of writing
> the redo log data will ensure recovery of the database only to the
> point of the last committed write - the question is when, during the
> backup process, did this last committed write take place ?
> Thanks for any further clarifications....
> Ian.
>
> "Muralidhar Prabhakaran" <muralip_at_earthlink.net> wrote in message
news:<YEb68.10190$ks5.981785_at_newsread1.prod.itd.earthlink.net>...
> > Ian,
> > Let me answer your question in two folds. First oracle 8i does have
IO
> > suspend features. ALter database suspend does that.
> > Secondly the split mirror backup can be accomplished by issuing the
alter
> > tablespace begin backup command just before the split and the alter
> > tablespace end backup just after. By doing this the file headers are
frozen
> > and splitting at that point gives a backup although not consistent. But
> > while recovering if we have all the archive and online log files then
the
> > recovery can be made consistent.
> >
> > HTH
> > Murali
> > "Ian Nash" <iannash_at_au1.ibm.com> wrote in message
> > news:3bc3b5f4.0201302239.4115eb6f_at_posting.google.com...
> > > Hi,
> > >
> > > I am trying to understand how Oracle provides a guaranteed consistent
> > > database backup when using the "split-mirror" online backup approach.
> > > Because the "split-mirror" relies on the underlying hardware and/or
> > > software to manage, it is outside of Oracle's influence, and in
> > > particular, is relying on the Operating System to manage the
> > > consistency of the data in the filesystems in the mirrors.
> > > In AIX, there is no guarantee about data consistency (ie all blocks
> > > written and metadata consistent) unless the filesystem is unmounted.
> > > This is why it is recommended that a filesystem is unmounted when a
> > > mirror split takes place, as it ensures that all data blocks are
> > > written, and that no changes to data blocks are taking place as the
> > > split occurs.
> > > Hence, my question is how does Oracle (especially Oracle 8i which -
> > > AFAIK - does not have I/O suspend and resume features) guarantee a
> > > consistent and complete DB backup using the split-mirror online backup
> > > approach, when the underlying filesystem may not contain all written
> > > blocks if a split is taken with the filesystem mounted ?
> > > I have read the Oracle Whitepapers on this backup approach, and still
> > > cannot see how data integrity can be guaranteed for the data copy
> > > residing in the mirror.
> > > Ian Nash
> > > Adv(Thanks)ance
> > >
>
Received on Fri Feb 01 2002 - 07:29:47 CST
