TM mode only with no connection pooling

[rrobetti]

We want to integrate atomikos in a Spring boot app with Oracle DB (using Oracle UCP) and IBM queues, Oracle UCP is good for us because it offers extra capabilities when interacting with the database. We have tried integrating both Oracle UCP with Atomikos without success, and that is due to the fact that both try to manage connections and that becomes messy under load.

Furthermore, we would like Atomikos to work in a Transaction Manager mode only, meaning it would not take care of connection pooling, leaving that job to Oracle UCP for example.

We are currently considering moving to Narayana as an alternative solution. Our implementation at the moment uses Atomikos but does not use Oracle UCP.

[GuyPardon]

We support JTA/XA enlist/delist if you want to do that. Can you explain why that is not sufficient? What does Narayana do more than that?

[rrobetti]

We support JTA/XA enlist/delist if you want to do that. Can you explain why that is not sufficient? What does Narayana do more than that?

We are still to investigate Narayana further, but apparently it can handle TM-only (no pooling) leaving the pooling to Oracle UCP, so Narayana would take care of: global transaction lifecycle, XA enlist/delist, 2PC, recovery.

[GuyPardon]

Well, Atomikos was designed to do all this. If you have concrete information to share please do, happy to make things better where we can. From what we hear, Atomikos has been doing better than the rest of the JTA/XA market but we sure want to keep it that way.

[rrobetti]

@GuyPardon I think the challenge lies on the fact that it is not possible to turn off the connection pooling in Atomikos and that would be desirable when working with Oracle UCP as per the latter gives great disaster recovery and fault resistance capabilities when working with Oracle database. I suppose we simply need a flag in Atomikos that turns off the pooling because we want Oracle UCP to take care of that part. Would that be possible? So we would wrap the native Oracle datasource in Oracle UCP datasource (with pooling configured) and would wrap the Oracle UCP datasource in Atomikos datasource (with pooling disable) so from atomikos perspective every time it needs a connection it creates a "new" one (pooled from Oracle UCP) it creates it from the datasource-oracle-ucp-wrapped and when finished with it "close" it (which in fact will just return it to Oracle UCP pool).

[rrobetti]

@GuyPardon I have opened a PR, please verify if that is something you can merge into the source code: #248

[rrobetti]

@GuyPardon is there an update on this? My understanding was that a separated PR would be created to implement this in a different way then my proposed PR here and it would be relseased now in March, is that still correct?