Fix NullPointerException in MongoSequenceIncrementer

[blake-bauman]

When using a MongoDB repository for the first time, the collections used by MongoSequenceIncrementer don't yet exist, which results in the call to findOneAndUpdate().getLong() to reutrn a null, which then results in an NPE unboxing the Long into a long.

Adding upsert(true) creates the initial document with a count of 0 and immediately increments it to 1.

[blake-bauman]

Is this a valid PR? We aren't able to use MongoDB as a repository without manually creating these documents and it took some reverse engineering to figure out what to create.

[fmbenhassine]

Is this a valid PR? We aren't able to use MongoDB as a repository without manually creating these documents and it took some reverse engineering to figure out what to create.

Similar to the JDBC job repository configuration, it is expected to execute this script in MongoDB before running any job. There is a note about that in the what's new section, but I believe this is not clear enough and we need to update the reference docs with this prerequisite in the job repository configuration section as well. I created #4859 for that.

With that script, there is no need for the upsert flag added in this PR (this test, which initializes the database with that script, passes without the upset flag), so I guess this PR is indeed not valid. Do you agree?

[blake-bauman]

I can see that you'd want to ensure that the indexes get created, so perhaps upsert isn't the right option after all. I appreciate the documentation change. I didn't think of looking in "what's new" since I just went straight to the reference guide.

The big problem, I think is that if the collection is not available, there is an NPE being thrown with no explanation, which could be an issue for those that didn't realize they missed a step. A couple of suggestions I would have is:

• Detecting on startup if a required collection isn't available and failing with an informative message.
• Or.....One of the benefits of MongoDB over JDBC is the ability to create these collections with the right attributes on the fly. So, a nice enhancement would be to detect that the collection is not created and automatically create it with the expected attributes before starting the job.

[fmbenhassine]

I fully agree. We tend to prefer the fail-fast approach, so I would argue that option 1 is better (explicit is better than implicit). This applies to a JDBC setup as well. In addition to the documentation's clarification issue, I created #4865 to improve the code with a configuration check.

I am closing this PR in favor of #4865. Thank you for reporting this issue and for your time on this PR!