No SQL error shown when using the JDBC operator

[jyotsa09]

Apache Airflow version

2.2.0b1 (beta snapshot)

Operating System

Debian buster

Versions of Apache Airflow Providers

No response

Deployment

Astronomer

Deployment details

astro dev start
Dockerfile - https://gist.github.com/jyotsa09/267940333ffae4d9f3a51ac19762c094#file-dockerfile

What happened

Both connection pointed to the same database where table "Person" doesn't exist.

When there is a SQL failure in PostgressOperator then logs are saying -

psycopg2.errors.UndefinedTable: relation "person" does not exist
LINE 1: Select * from Person

When there is a SQL failure in JDBCOperator then logs are just saying -

[2021-09-23, 19:06:22 UTC] {local_task_job.py:154} INFO - Task exited with return code

What you expected to happen

JDBCOperator should show "Task failed with exception" or something similar to Postgress.

How to reproduce

Use this dag - https://gist.github.com/jyotsa09/267940333ffae4d9f3a51ac19762c094
(Connection extras is in the gist.)

Anything else

Related: #16564

Are you willing to submit PR?

• Yes I am willing to submit a PR!

Code of Conduct

• I agree to follow this project's Code of Conduct

[subkanthi]

@potiuk I can try this out.

[kaxil]

@subkanthi assigned it to you, go for it

[uranusjr]

I remember chatting with another developer on this. Note that this seems to be problem specific to JDBC; after #16805 (the PR fixing #16564), most task errors can correctly be logged—but not JDBC. In fact, the JDBC error does not even go through the code path implemented in the PR; when using JDBCOperator, you get no errors at all, not the Task failed with exception message an exception would normally trigger (for example, the Postgres example above would emit this log). My current hypothesis is JDBC (or the operator/hook) is swallowing the exception somehow and the scheduler never actually received it, let alone logging.

[subkanthi]

Im having trouble getting the jar working for the JDBC connection, unassigning.