Calling commit multiple times from an implicit transaction causes the code to hang

[AlexFlat]

Hi, We are using your embedded postgres library for unit tests and it has been working great. However, we have recently noticed that the tests seem to hang during a parallel execution.

I have managed to reproduce this behavior in a separate solution and it seems to be related to the fact we are attempting to call "commit" from an implicit transaction multiple times. Our code is attempting to perform a "set statement_timeout" as per articles such as; https://stackoverflow.com/questions/24092463/psql-set-default-statement-timeout-as-a-user-in-postgres

Here is the project used to reproduce this. The instructions are in the readme of the repro. https://github.com/AlexFlat/NPGSQLConnectionSetTimeout/tree/master

I am logging this here as I am unable to reproduce this using a "real" postgres install and wanted to see if there was some setting we have missed when configuring the PGServer.

Thanks in advance for your time

[mysticmind]

@AlexFlat I shall have a look at this issue and get back to you. Following are the possible quick remedies:

• pass a server param statement_timeout and the value so that you can use it for all SQL queries without doing it for every call?
• Adjust the set statement to not add commit if you are not using a transaction.

Please let me know your thoughts.

[AlexFlat]

@mysticmind thanks for the quick reply.

Responses to your suggestions;

• pass a server param statement_timeout and the value so that you can use it for all SQL queries without doing it for every call? [We need to be able to adjust this timeout on a per-request basis so we cannot hard code this for the server]
• Adjust the set statement to not add commit if you are not using a transaction. [I discussed this with my colleagues, however the concern was that the statement will not take effect for the current connection unless the "commit;" is included. Happy to be proved wrong on this one, but all the references on the web seem to always include the "commit;"]

Thanks again for the quick reply

[mysticmind]

I am not sure what I could fix on this. Closing this for now.