RE: DataGuard

Laimutis,

For a standby to become unsynchronized, primary
doesn't have to crash. It might simply be unable to
deal with redo rates high enough. In our case we
experienced different kinds of ORA-16166 when this
happens like "ORA-16166: LGWR timed out on Network
Server 2 due to buffer full condition" or "ORA-16166:
LGWR network server failed to send remote message" 
But your scenario is easier to simulate :-)

Now, are you concerned with potential data loss or
possibility of a FSFO to an unsynchronized standby and
users connected to two primary databases? If it's the
former than that can certainly happen unless you
elevate your protection mode to max protection. If
it's the latter (which I suspect was your question)
than I don't see ? at least in theory (i.e. no bugs) ?
how can this occur.

Let's go step by step. When target standby goes down,
primary knows about it right away. You see something
like the following in the primary alert logs...

LGWR: Attempting destination LOG_ARCHIVE_DEST_x
network reconnect (3113)
LGWR: Destination LOG_ARCHIVE_DEST_x network reconnect
abandoned
Errors in file /.../<sid>_lgwr_2621.trc:
ORA-03113: end-of-file on communication channel
LGWR: Network asynch I/O wait error 3113 log 5 service
'(DESCRIPTION=

Destination LOG_ARCHIVE_DEST_x is UNSYNCHRONIZED

LGWR: Failed to archive log 5 thread 1 sequence 350
(3113)
LGWR: Closing remote archive destination
LOG_ARCHIVE_DEST_x: '(DESCRIPTION=
Errors in file /.../<sid>_lgwr_2621.trc:
ORA-01041: internal error. hostdef extension doesn't
exist
LGWR: Error 1041 closing archivelog file
'(DESCRIPTION=
LGWR: Error 1041 disconnecting from destination
LOG_ARCHIVE_DEST_x standby host '(DESCRIPTION=


... or may be like this (I think depending on the way
target standby went down and which instance?s
alert.log you monitor [if your primary is RAC]):

LGWR: Attempting destination LOG_ARCHIVE_DEST_x
network reconnect (1089)
LGWR: Error 1041 disconnecting from destination
LOG_ARCHIVE_DEST_x standby host '(DESCRIPTION=?
*******************************************************WARNING:
Detected remote instance standby database network
disconnect
WARNING: Current instance standby database disconnect
may be required
*******************************************************LNSd
started with pid=46, OS id=29335
Error 12528 received logging on to the standby
LGWR: Error 12528 attaching to RFS for reconnect
LNSd started with pid=49, OS id=30161
Error 12528 received logging on to the standby
LGWR: Error 12528 attaching to RFS for reconnect
ARC1: Attempting destination LOG_ARCHIVE_DEST_x
network reconnect (3113)
ARC1: Destination LOG_ARCHIVE_DEST_x network reconnect
abandoned
PING[ARC1]: Error 3113 when pinging standby
(DESCRIPTION=?
LNSd started with pid=39, OS id=31001
Error 12514 received logging on to the standby
LGWR: Error 12514 attaching to RFS for reconnect
LNSd started with pid=49, OS id=31589
Error 12514 received logging on to the standby
LGWR: Error 12514 attaching to RFS for reconnect


... and v$database.fs_failover_status obviously
becomes "UNSYNCHRONIZED". If all instances of primary
go down afterwards -- but before target standby came
back up to resynchronize  -- than naturally no FSFO
can occur. There's no target standby to fail over to.

Now, if target standby than comes back ?- and all
primary instances are still down ?- than still no FSFO
takes place as Oracle knows that this standby is not
synchronized. In addition to fs_failover_status you
should also see v$database.protection_level switching
to "RESYNCHRONIZATION", while protection_mode remains
"MAXIMUM AVAILABILITY". Finally restarting primary
resynchs all databases.

Thanks,
Boris Dali.


--- Laimutis Nedzinskas
<Laimutis.Nedzinskas@xxxxxxxxxxxxx> wrote:

> I already described the scenario of primary crashing
> while standby is unavailable (in other words - 
> primary crashing in downgraded performance
> mode.) Standby is unaware then of primary's
> progress
...


__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 
--
http://www.freelists.org/webpage/oracle-l


Other related posts: