When should log events be captured in a database?

This is slightly off-topic, but the PostgreSQL cognoscenti is likely to be the best audience for the question.

I am writing an application -- an application-specific (scaled down) Web server in Node.js.  A question I keep asking myself is whether it it better to simply log incoming requests, or write them to a database.  Ultimately, I would like to analyze the data, so moving into a database makes sense.  However, I also can see the point of logging as a simpler, less CPU & I/O intensive activity.  When life goes wrong, capturing data in a log file may be the easier.  Plus, resources are freed to handle requests which is the fundamental goal of a Web server anyways.

Yet if the database resides on another machine, this dismisses some of the CPU - I/O load argument.

I can't believe that the frequency data is acquired is the determinant, but I may be wrong.

I am also aware that there are various tools available for parsing log file data into databases, & writing such tools is not altogether complicated.  Nevertheless, this seems to be a redundant exercise.  

So, I come back full circle.  Even PostgreSQL itself has its log files.  Not everything is written to database tables proper.  Yet at what point does data take on a new status such that it should be collected in a database over simply being written to logs?

Thanks for all candor shared.