This video explains how to compact the datastore, but how does Discovery actually compact it?

It depends whether the datastore is multi generational or not. See How to check if multi generational datastore is enabled? 

When multi generational datastore is disabled: For each file in /usr/tideway/var/tideway.db/data/datadir (for example: p0002_nHost_hist):

1. 1. a (compacted) copy of the file in is created in /usr/tideway/var/tideway.db/data/datadir/n00000
   2. the file is removed from datadir

When no files left in datadir, the (compacted) files are copied from n00000 back to datadir.

When multi generational datastore is enabled, tw_ds_compact --online executes the steps below:

1. The existing generations are compacted/merged into a new directory named "n00000<number>"
2. The files in "n00000<number>" are moved to another folder named "c00000<number>"
3. The previous read-only generation named "g00000<number>" are deleted
4. A new current generation named "g00000<number>" is created

*Para ver una versión en español de este articulo, consulte: KA 000357478