Well aware of the -b bugs.  We have filed many issues and have them all resolved on our version.

We don't xfer the ckp's because it is 60GB over a WAN.  We also "zero-out" the .jnl files before and after the ones we want to rollforward.  We do NOT monkey with the .cnf file, just transfer it.

-----Original Message-----
From: martin.bowes@ctsu.ox.ac.uk [mailto:martin.bowes@ctsu.ox.ac.uk]
Sent: Thursday, June 09, 2005 4:13 AM
To: Kale, Mark (STP)
Cc: info-ingres@cariboulake.com
Subject: Re: [Info-ingres] Re: Ingres stand-by database


Hi Mark 

	I think I may have missed a point here, but...

	You do a backup of your production database and then only 
shove the journals over to the recovery host - not the checkpoint you 
just took.

	Then, after monkeying with the config file, you use the journals 
to do a rollforwarddb -c -b?

	Why?

	Why don't you use the checkpoint you just made on the 
production host, transfer those checkpoint tar files to the DR host and 
do a normal recovery?

	What have I missed?

	BTW. The -b flag on rollforwarddb has been subject to some 
bugs. Specifically relating to whether or not it represents the transaction 
commit time - as it is supposed to.

Marty

> Chloe,
> 
> Thought I would chime in.  We have a drp box configured almost
> identically to production.  We have a cronjob that copies over journal
> files (not yet copied) every 10 minutes to our drp server.  We set a
> server parameter to make sure the log is cleared at least every 5
> minutes (in reality it is done about each minute).  So worst case, we
> would lose 15 minutes of data (configurable).
> 
> Each night we checkpoint our database on production and copy the
> cnf/dmp files to drp.  We run a cute script to "zero out" the journal
> files that show up in infodb after the checkpoint.  We then get the
> next-to-last checkpoint time and run the following rollforwarddb:
> 
> # copy config file to safe place
> cp -p /usr/ingres/data1/ingres/data/default/${db}/aaaaaaaa.cnf \
>         /ingsav/cnf/${db}/aaaaaaaa.cnf.${tm}.roll-c+j.before
> 
> echo "Before config file copied to /ingsav/cnf/$db"
> 
> #
> # Must use force flag, here is Ingres error without it:
> #
> # Mon May 17 16:22:53 2004 E_DM1351_RFP_BTIME_NOFORCE     Rollfoward
> parameter e
> rror.  'ROLLFORWARDDB -b' has been requested without the force flag
> (#f).  This recovery starts from an unknown database state, and cannot
> be performed without specifying the force flag.  The results of a
> ROLLFORWARDDB -b recovery are unpre dictable unless it immediately
> follows a ROLLFORWARDDB -e -norollback specifying
>  exactly the same time.
> #
> 
> ${II_SYSTEM}/ingres/bin/rollforwarddb "#c${ckpnum}" -c +j "#m4" "#f"
> -b${ckpdate } ${db} -u${dbo} >> $log 2>&1 ret_code=${?}
> 
> 
> We follow this with full checksums against another copy of the
> production checkpoint that we just took.  It has worked on Ingres 2.6
> for a couple of years anyway.
> 
> Hope this helps,
> 
> Mark Kale
> 
> _______________________________________________
> Info-ingres mailing list
> Info-ingres@cariboulake.com
> http://mailman.cariboulake.com/mailman/listinfo.py/info-ingres
>