Exploring the structure of tibbles

If you try to print a tibble that describes data stored in Spark, some magic has to happen, since the tibble doesn't keep a copy of the data itself. The magic is that the print method uses your Spark connection, copies some of the contents back to R, and displays those values as though the data had been stored locally. As you saw earlier in the chapter, copying data is a slow operation, so by default, only 10 rows and as many columns will fit onscreen, are printed.

You can change the number of rows that are printed using the n argument to print(). You can also change the width of content to display using the width argument, which is specified as the number of characters (not the number of columns). A nice trick is to use width = Inf to print all the columns.

The str() function is typically used to display the structure of a variable. For data.frames, it gives a nice summary with the type and first few values of each column. For tibbles that have a remote data source however, str() doesn't know how to retrieve the data. That means that if you call str() on a tibble that contains data stored in Spark, you see a list containing a Spark connection object, and a few other bits and pieces.

If you want to see a summary of what each column contains in the dataset that the tibble refers to, you need to call glimpse() instead. Note that for remote data such as those stored in a Spark cluster datasets, the number of rows is a lie! In this case, glimpse() fails to properly report the number of rows.