Ignacio_Romaguera answered on June 3rd 19 at 21:01
Depends on the requirements bekapa.
If you need point in time recovery, i.e. the ability to restore the state of the cluster to some arbitrary transaction - it is pg_basebackup + WAL files.
To collect designer manually - there is a barman. If bekapit want somewhere in the s3 - it is easier to be wal-e or wal-g
One base separately so bekapit and restore will not work, only the whole instance as a whole. Volume backup = volume of the entire instance + volume of all WAL basebackup from the beginning to the desired point of recovery, the number of wal depends on your writing load.
If enough logical snapshot of the data, for example, every day, then Yes, pg_dump. Restore, respectively, only at the beginning of the lifting of the dump.
Typically done pg_dumpall -g to preserve the global cluster data - user, tablespace, database names and a separate pg_dump for those bases that have bekapit.
The format is often custom made, because he was a staff shrinks and allows using pg_restore to restore specific objects from the dump, not just the whole thing as text using psql. For GB base makes no sense, but for bigger databases it makes sense to do the format of the directory in multiple threads.