I don't think this is the right thing to do. The documentation recommends checking against null to see if the job still exists.

What's the motivation for this change?

@pongad My motivation was dry run queries but it might be more general. Running a dry-run query job doesn't actually create a job. Without this change, a dry run query job with an invalid query just returned null instead of anything helpful. With this change, it throws a proper syntax error exception.

In a larger scope, this is in the error path for createing a job. Specifically this is after the job fails creation with an exception. It seemed reasonable to me that if the create fails and the job isn't found (not like a 5XX scenario) we should rethrow the creation exception. See the comment above the code I changed.

Ah you're right. And because this is in create, it shouldn't change behavior of getting jobs elsewhere. Thank you for the PR and explanation.

This LGTM, but @tswast should sign off.