Re: HA - High Availability
From: Dan Norris <dannorris_at_dannorris.com>
Date: Wed, 25 Jun 2008 08:11:26 -0500
Message-ID: <486243FE.60802@dannorris.com>
2. I measure failover time from the application perspective. That is, how long does it take for a client to resume the ability to run SQL statements on a surviving node.
3. Node failure (power off), instance failure (shutdown abort), network failure (pull a cable), storage failure (pull a storage cable), process failure (kill a dedicated server shadow). I generally perform each of these tests and record the results, then inform the application development team what exceptions they should prepare to handle.
4. Yes, the T in TAF is for "transparent". Therefore, you should at least be able to verify that you have the ability to connect to a surviving node transparently. However, even TAF can't fail over a transaction in process (uncommitted), so you'll have to handle restarting a transaction and handling the associated ORA- error that comes when a session failover happens while an uncommitted transaction is in process. Your failure testing scenarios should be performed with and without a transaction in process.
5. If you have TYPE=SELECT, you should be able to have that select stmt continue on a surviving member.
6. In 10.2 and higher, the server-side configuration overrides the client configuration. In fact, very little client-side configuration is needed except to support preconnect. See DBMS_SERVICE for more information.
7. I haven't seen any specific HA recommendations. All RAC clusters I've ever seen have had expectation that HA is part of the configuration, so I think most of those recommendations are baked in to the initial configuration.
Good luck!
Dan
Waldirio Manhães Pinheiro wrote:
Date: Wed, 25 Jun 2008 08:11:26 -0500
Message-ID: <486243FE.60802@dannorris.com>
1. Varies. VIP failover takes up to a minute sometimes. However, using FCF and TAF, client failovers can happen on the order of a few seconds instead of a minute or more.
2. I measure failover time from the application perspective. That is, how long does it take for a client to resume the ability to run SQL statements on a surviving node.
3. Node failure (power off), instance failure (shutdown abort), network failure (pull a cable), storage failure (pull a storage cable), process failure (kill a dedicated server shadow). I generally perform each of these tests and record the results, then inform the application development team what exceptions they should prepare to handle.
4. Yes, the T in TAF is for "transparent". Therefore, you should at least be able to verify that you have the ability to connect to a surviving node transparently. However, even TAF can't fail over a transaction in process (uncommitted), so you'll have to handle restarting a transaction and handling the associated ORA- error that comes when a session failover happens while an uncommitted transaction is in process. Your failure testing scenarios should be performed with and without a transaction in process.
5. If you have TYPE=SELECT, you should be able to have that select stmt continue on a surviving member.
6. In 10.2 and higher, the server-side configuration overrides the client configuration. In fact, very little client-side configuration is needed except to support preconnect. See DBMS_SERVICE for more information.
7. I haven't seen any specific HA recommendations. All RAC clusters I've ever seen have had expectation that HA is part of the configuration, so I think most of those recommendations are baked in to the initial configuration.
Good luck!
Dan
Waldirio Manhães Pinheiro wrote:
-- http://www.freelists.org/webpage/oracle-l Received on Wed Jun 25 2008 - 08:11:26 CDTGood morning friendsI'd like to ask about some points in oracle RAC.1. About failover, how long time to complete failover ?(power off first node and power on the vip address on second node)2. In general, how do you measure the "failover time" ?!3. In general, how do you test the Failover in RAC environment ?!4. I have been configured the environment RAC 10.2.0.3 with TAF (preconnect). Can I think in HA if my application was developed without code to HA ?!5. With RAC and TAF environment, the "select * from dba_tab_columns;" command for example will work fine or will stop when the node is power off !? (the node which I'm connected from sqlplus)6. About TAF configuration, It's necessary configure only on client, only on server or in both (Server and Client) ?!7. There are any recommended configuration to HA !? (change about time of service or RAC cluster)Thank you so much in advanced.
Cheers
--
______________
Atenciosamente
Waldirio
msn: wmp@sinope.com.br
Site: www.waldirio.com.br
Blog: blog.waldirio.com.br
PGP: www.waldirio.com.br/public.html