Weblogic 7.0 SP1 JDBC Multipooling does not appear to fail over consistantly

Question

We are running Weblogic Platform 7.0 SP1 on a Sun with Solaris 8 O/S
We have a connection multipool set up associated with a Data Source.
The multipool has two connection pools, each pointing to a seperate but
Identical database. The multipool is created with the high availability algorithm
set.
both connection pools have the following settings:
(Connections Tab) (Testing Tab)
Initial Capacity:0 Test Table Name: dual
Maximum Capacity: 20 Test Reserved Connections: Checked
Capacity Increment: 2
Login Delay Seconds: 2
Refresh Period: 15
Allow Shrinking: Checked
Shrink Period: 15
When the first database goes down for a backup, We can see the exceptions thrown
from the
broken acitve connections, which would be expected. But we also notice a significant
response time increase (3-7 seconds) and we notice continuing exceptions from
what appears to be the application EJB still attempting to connect to the first
instance. It is only when a DISABLE_POOL is issued using the command line interface
to disable the first pool in the multipool that the response time drops to normal
and the exceptions stop. The exceptions don't happen for every query, but we cannot
understand why the test-on-reserve doesn't seem to always work.
Any help would be appreciated

Answer 1

This is known behavior of the "high availability" algorithm for multipools.
Whenever a multipool is asked for a connection, it will always try the first pool first
(unless it has been diabled). The algorithm doesn't remember that it had trouble with
the first pool and start with the second pool (e.g., put the first pool at the end of the circular list
of pools to be tried). This is the way it was designed (in my opinion, this defies the
principal of least astonishment, but not everyone agrees with me).
We are planning to add a new algorithm in the next release that does keep state.
The way it currently works, it depends on how long it takes to figure out that
the first pool is not available. This depends on the database client driver and
the network connection (it could take from 100ms to multiple seconds to minutes
to figure it out).
"Joe Doyle" <[email protected]> wrote in message news:[email protected]...
>
We are running Weblogic Platform 7.0 SP1 on a Sun with Solaris 8 O/S
We have a connection multipool set up associated with a Data Source.
The multipool has two connection pools, each pointing to a seperate but
Identical database. The multipool is created with the high availability algorithm
set.
both connection pools have the following settings:
(Connections Tab) (Testing Tab)
Initial Capacity:0 Test Table Name: dual
Maximum Capacity: 20 Test Reserved Connections: Checked
Capacity Increment: 2
Login Delay Seconds: 2
Refresh Period: 15
Allow Shrinking: Checked
Shrink Period: 15
When the first database goes down for a backup, We can see the exceptions thrown
from the
broken acitve connections, which would be expected. But we also notice a significant
response time increase (3-7 seconds) and we notice continuing exceptions from
what appears to be the application EJB still attempting to connect to the first
instance. It is only when a DISABLE_POOL is issued using the command line interface
to disable the first pool in the multipool that the response time drops to normal
and the exceptions stop. The exceptions don't happen for every query, but we cannot
understand why the test-on-reserve doesn't seem to always work.
Any help would be appreciated