Grouper Users - Open Discussion List

[Colin Hudler <>]

We also use Oracle with Data Guard and Grouper. We are using the 
full/native Oracle client, rather than the "thin" adapter. Our 
connection string looks like this:

hibernate.connection.url = jdbc:oracle:oci:@GRPR.WORLD

tnsnames something like this:
GRPR.WORLD =  (DESCRIPTION =  (ADDRESS_LIST =  (ADDRESS = (PROTOCOL = 
TCP)(HOST = groupsdb-proda)(PORT = 1522))(ADDRESS = (PROTOCOL = 
TCP)(HOST = groupsdb-prodb)(PORT = 1522)))(CONNECT_DATA = (SERVICE_NAME= 
GRPDB.uchicago.edu)))

We are using the OCI source because it supports transparent failover 
(TAF), even with dataguard. In practice, we found that outstanding 
transactions (read or write), and existing connections are not always 
moved after failover[1]. Using OCI data source may not be necessary, 
YMMV. Our tests showed it not necessary to do anything to grouper after 
the database has been failed over. It recovered very well. There was 
about a 10 second delay when the UI/WS/API would simply not connect, but 
it recovers without intervention. I think this would also depend on the 
volume and nature of requests that your grouper gets, especially during 
the failover.

On 06/09/2011 01:33 PM, Peter DiCamillo wrote:
> Thanks for the information. Sorry for the premature request-- it turns 
> out that our DBAs don't have RAC installed, but use Oracle Data Guard. 
> So I'll be investigating that, and would be interested in any 
> information regarding use of it with Grouper.
>
> Peter
>
> Chris Hyzer wrote:
> > Let me say another thing... if you get the latest oracle driver jar, 
> > and configure the rac nodes in the connect string of 
> > grouper.hibernate.properties, e.g.
> >
> >
> > hibernate.connection.url = 
> > jdbc:oracle:thin:@(DESCRIPTION=(FAILOVER=on)(ADDRESS_LIST=(ADDRESS=(PROTOCOL=TCP)(HOST=server1.school.edu)(PORT=1521))(ADDRESS=(PROTOCOL=TCP)(HOST=server2.school.edu)(PORT=1521)))(CONNECT_DATA=(SID=myid)))
> >
> > <note, you will need to adjust the failover settings etc>
> >
> > Maybe you need to adjust the c3p0 settings so the connections timeout 
> > after some short period of time (1 minute?)
> >
> > Then I would think when a rac node went down, then the driver would 
> > use the other node.  However, on our non-grouper systems, we tried 
> > this (maybe it was our c3p0 settings...), but it didn't really work 
> > when a rac node failed.  The disadvantage of this is there is not the 
> > ONS to proactively tell the driver to take a node out of the pool, 
> > however, the hardcoded lists is better then nothing if ONS is not 
> > available in Grouper yet.  Bottom line, it might be worth a shot to 
> > try something like this and see if it works until ONS :)
> >
> > Thanks,
> > Chris
> >
> > -----Original Message-----
> > From: Chris Hyzer Sent: Thursday, June 09, 2011 9:24 AM
> > To: 'William G. Thompson, Jr.'; Peter DiCamillo
> > Cc: 
> >
> > Subject: RE: [grouper-users] Grouper and database failover
> >
> > > Curious if you (or anyone else) has also consider a NoSQL option for
> > > this?  OrientDB looks particularly interesting...
> > > http://code.google.com/p/orient/
> >
> > The Grouper API heavily depends on a relational database...
> >
> > > We're looking at the possibility of configuring Grouper so we can use
> > > database failover. We'd like to be able to keep Grouper up during 
> > > database
> > > maintenance, and also in disaster recover situations. We use Grouper 
> > > with an
> > > Oracle database and with tomcat. We also have some java programs 
> > > which use
> > > the Grouper API and do not run under tomcat.
> > >
> > > I don't have extensive knowledge of Oracle or jdbc, but some quick 
> > > Google
> > > research suggests that this may be possible, and the recommended method
> > > involves using Oracle's Fast Connection Failover mechanism.
> > >
> > > Does anyone have experience with Grouper and database failover? Are 
> > > there
> > > known reasons why it could or could not be made to work?
> >
> > Here is my understanding... FCF requires either oracle's db pool 
> > where you configure the ONS and that tells it what servers are in the 
> > pool, and which ones to take out when they go down.  You can use FCF 
> > with c3p0 (our db pool) if you use the "simple fan" where you write 
> > code to listen for ons events and adjust the pool accordingly (sounds 
> > less appealing to me).  Right now Grouper uses c3p0 and a DB driver, 
> > so if that is all that is needed, configure it as you like.  If you 
> > are using ONS, I think we will need to make the pool type 
> > configurable, and do some work to make sure it is ONS compatible.  
> > Penn is probably going to need this at some point, and we use ONS for 
> > other things, so we will add this into Grouper at some point if 
> > someone else doesn't do it sooner :)
> >
> > Btw, we are just using hibernate with c3p0 in hibernate, maybe there 
> > is a way to adjust the grouper.hibernate.properties to use fcf 
> > without changing code (I wouldn't bet on it, but maybe) :)
> >
> > Thanks,
> > Chris