[16.0][FIX] queue_job: fix deadlock

[BT-sschmid]

Fix a deadlock condition when there are dependent stored fields on queue.job:state.

Example:

• Sale-Order has a stored computed field which depends on job-state.
• A job updates a sale-order record.
• A exception is raised and in the final error-handling with a child cursor in

  queue/queue_job/controllers/main.py

  Line 144 in c93a253

  with registry(job.env.cr.dbname).cursor() as new_cr:

  the job's state is updated, which also triggers an update in the sale order. Result is a deadlock.

This commit adds a rollback in case of error during job execution, so that any locks are released.

[OCA-git-bot]

Hi @guewen,
some modules you are maintaining are being modified, check this out!

[amh-mw]

Though I generally agree with the intent of this change, there is already some rollback code in the enclosing runjob call, specifically a rollback in the event of a RetryableJobError. It makes me question why the except (FailedJobError, Exception) stanza doesn't rollback as well, choosing instead to grab a different cursor to commit the job state.

[BT-sschmid]

Though I generally agree with the intent of this change, there is already some rollback code in the enclosing runjob call, specifically a rollback in the event of a RetryableJobError. It makes me question why the except (FailedJobError, Exception) stanza doesn't rollback as well, choosing instead to grab a different cursor to commit the job state.

now I got your comment ^^
Indeed I also don't see the need to use a new cursor if the tx is rolled back. If you agree I change it to just rollback.