Performance of PG_DUMP vs Replication with master/slave?

Question

So I need to backup my database, and have 2 options I am considering:

1) Using pg_dump to do a daily dump 2) Setting up a slave database that will be a replication of the master.

I'm not asking which one is the best option, but rather asking:

How does the performance of these 2 compare? Intuitively, both are doing similar things, which is streaming all the data in 1 single process from 1 server to another.

Which one of the 2 has less of an impact, CPU/load wise on the master database server? Which one will have less of an impact on the current queries being processed?

Answer 1

That's an apples to oranges type of question. pg_dump gives you a logical backup of your database, which you can manipulate or move to other systems to reload as an internally-consistent database. The replicated slave is an exact copy of the live master, which changes as the master changes. Now, if you wanted to go for a hybrid approach, you could set up replication between master and slave, and configure the hot standby so that you could take a pg_dump from the slave. That's the method I usually employ in production environments because of the flexibility and reduced impact on the master.