[matheusd]

One issue we had: pgdump performance was abysmal for dumping a single schema out of thousands.

We had to write a custom backup script to handle individual backups, taking advantage of the knowledge of our own database architecture (basically: we don't have to read the schema list and figure out relationships between them because we already know that).

If you read pgdump's source code, when doing the actual backup it uses postgres COPY command, so it was easy enough to write our custom exporter.