Rewrite JdbcOrphanLockAwareIdempotentRepository::insert to not rely on exception #9286
Conversation
- the initial exception will mark the transaction as dirty, blocking subsequent changes - not all databases support rollback of DDL (e.g MySQL, older Oracle DBs)
|
🌟 Thank you for your contribution to the Apache Camel project! 🌟 If necessary Apache Camel Committers may access logs and test results in the job summaries! |
Components tested:
|
|
@turing85 I can't understand
What is the stack trace for the next creation sql? |
Reproducer:
|
|
An alternative solution would be to:
This would look something like this ( |
|
Thanks @turing85 for sharing and I will check it. |
|
@zhfeng this got a lot more complicated. As soon as we set the route This is due to the fact that we try to commit a local transaction (through I haven't checked yet if this also affects other components, but I suspect it does. This seems to be a general issue. |
|
Hi @turing85 I can re-producer the issue with It only needs to change
I think this might be a limitation on |
|
Just get something from https://stackoverflow.com/questions/10399727/psqlexception-current-transaction-is-aborted-commands-ignored-until-end-of-tra It confirms that this is an expected behavior on |
|
OK, I tried the following configurations and it can work with "%dev":
quarkus:
datasource:
db-kind: postgresql
devservices:
properties:
autosave: always
flyway:
migrate-at-start: false |
|
@turing85 I also investigate
public Route(@SuppressWarnings("CdiInjectionPointsInspection") DataSource dataSource,
@SuppressWarnings("CdiInjectionPointsInspection") TransactionManager transactionManager) {
TransactionTemplate template = new TransactionTemplate(new JtaTransactionManager(transactionManager));
idempotentRepository =
constructRepository(dataSource, template);
}
private static JdbcMessageIdRepository constructRepository(DataSource dataSource, TransactionTemplate template) {
final JdbcMessageIdRepository repository =
new JdbcMessageIdRepository(dataSource, template, ROUTE_ID);
repository.setCreateTableIfNotExists(true);
return repository;
}I test it locally and it should work with |
|
I would say this is a great example for using Thanks a lot! |
@zhfeng I can confirm that the issue does not occur with h2. |
|
@turing85 is it OK to close this PR and maybe we can add a NOTE with |
|
@zhfeng I think it is not only postgres; db2 might also be affected. |
|
Please check https://jdbc.postgresql.org/documentation/use/ for |
It could be good but I wonder if |
This query is vendor-specific. But this part does not seem to be coupled to the global transaction manager, hence we might be in the clear here. The critical part is the |
|
If I uderstand correctly, you were wanting to seperate the transaction in |
|
Either this or rewrite |
turing85
changed the title
|
@zhfeng I revised tie PR:
With those changes and your comment about the |
Components tested:
|
|
@turing85 Well done! - I will have a look today. |
Components tested:
|
|
Can we also have the catch duplicate exception still just in case it still happens in some of the other databases ? |
I belive that there will be no |
Ok perfect |
|
@turing85 One more question: If I understand correctly, this |
@zhfeng Hmm... we could use |
|
@zhfeng An I am not sure if this is a good user experience; those queries are complex... |
|
I think this PR is good as-is. And agree its better to support existing databases as-is. We had SQL for many many decades and UPSERT is not common practice and every database is a bit special. |
|
@davsclaus should I then squash and push again, or will you guys squash on merge? |
|
github has a squash and merge button that we use |
|
Thanks @turing85 for your contributions! |
|
@zhfeng you're welcome 🙂 I'll add the example when I have some spare time. |
4 participants
Rewrote
JdbcOrphanLockAwareIdempotentRepository::insertto not rely on an exception. The exception caused problems with postgresql databases ifautosave=alwayswas not set.