We intentionally close Oracle connections during Offload transport to get around problems with idle sessions being sniped. We recently introduced a bug by moving code that checked the oracle version to later in the process of building a transport source query without including a disconnect. Due to this a second partition batch failed with ORA-3113.
If Offload transport used OracleFrontendApi then this would not have happened, I've created an issue for that, https://github.com/gluent/goe/issues/161, but for now we should introduce the missing disconnect. That is what this PR does.
We intentionally close Oracle connections during Offload transport to get around problems with idle sessions being sniped. We recently introduced a bug by moving code that checked the oracle version to later in the process of building a transport source query without including a disconnect. Due to this a second partition batch failed with ORA-3113.
If Offload transport used
OracleFrontendApi
then this would not have happened, I've created an issue for that, https://github.com/gluent/goe/issues/161, but for now we should introduce the missing disconnect. That is what this PR does.