| Oracle FAQ | Your Portal to the Oracle Knowledge Grid | |
 |
 | |||
| Oracle FAQ | Your Portal to the Oracle Knowledge Grid | |
|
| |||
Home -> Community -> Mailing Lists -> Oracle-L -> Re: OPS and redos
Fawzia, What are you querying to get the SCN
information. I would also be curious where you read
about the recovery issues in OPS environment? Oracle
maintains what is called a snapshot SCN and global
SCN. The snapshot SCN is usually maintained in the SGA
and the global SCN is maintained by the DLM. All
transactions committed in an OPS environment use the
global SCN and that SCN is always incrementing even in
if both instances are idle. In an OPS environment it
is possible for multiple transactions on different
nodes to commit with the same SCN but each transaction
will commit with an SCN that is greater whether these
transactions are on the same node or different nodes.
The bottom line is that Oracle basically treats SCN's
with committed transactions the same whether it is a
single instance or OPS.
The snapshot SCN is used for queries and the snapshot SCN could be older than current or global SCN. This SCN only effects queries not DML. The only issue here is using this snapshot SCN could give read consistency on transactions that where committed on the on the other node.
an example
Time 10:00:00 - Node2 issues DML and commits updates deptno from 10 to 11 committed with current SCN 1000, current SCN is now incremented to 1001
Time 10:00:01 - Node1 issues a select on deptno and sees 10 (should see 11 because the transaction was commmited) because the snapshot SCN is 998 and the transaction was committed with newer SCN we must perform a read consistent read.
PMON is the process that synchs current or global SCN with the snapshot SCN and PMON usually does this every 3 seconds
so if you issue the query
Time 10:00:05 (5 seconds after the commit) - Node1
issues the select on deptno and now sees 11. This is
because PMON sync-ed up the snapshot SCN with the
current or global SCN and now my SCN for my query is
newer than the SCN the transaction you committed with
so there is no need to perform a CR you can just use
the current version.
The above issue is really only a problem is a system if you have a lot of DML on 1 node and a lot of selects on the other node. It this is an issue you can set MAX_COMMIT_PROPAGATION_DELAY=0 and this will force the queries to always use the current or global SCN and now you should see the committed data as soon as it is committed because both the DML and queries are use the current or global SCN.
I find odd that you have SCN of 500 and 2000. The way Oracle maintains SCN's in an OPS environment shouldn't allow for this large of a gap in SCN's. This looks more like log sequence numbers not SCN's. If these are indeed log sequence then this gap is not an issue. Recovery is based more on SCN's not log sequence numbers.
If the gap in the log sequence gets to big Oracle will force the idle node to do log switches and if you are archiving you will see archive logs 1 OS block in size. When Oracle does this you should a KK "Redo log Kick" lock being allocated.
Hope this helps,
Scott
-- Please see the official ORACLE-L FAQ: http://www.orafaq.com -- Author: Scott INET: oraracdba_at_yahoo.com Fat City Network Services -- (858) 538-5051 FAX: (858) 538-5051 San Diego, California -- Public Internet access / Mailing Lists -------------------------------------------------------------------- To REMOVE yourself from this mailing list, send an E-Mail message to: ListGuru_at_fatcity.com (note EXACT spelling of 'ListGuru') and in the message BODY, include a line containing: UNSUB ORACLE-L (or the name of mailing list you want to be removed from). You may also send the HELP command for other information (like subscribing).Received on Tue Jul 02 2002 - 13:43:36 CDT
 |
 |