Hacker News Reader

File system snapshots of databases are not necessarily consistent, and can not always be restored like that.

Atomic snapshots like ZFS's are always consistent for Postgres. I guess other databases with a similar write-ahead log can be snapshotted as well?

> Atomic snapshots like ZFS's are always consistent for Postgres.

As long as you make sure to only use one filesystem, i.e. you don't place pg_xlog or some tablespaces on a different filesystem. You can get very weird corruption in such cases :)

With ZFS you can also do atomic snapshots of multiple filesystems. https://serverfault.com/questions/608223/is-zfs-snapshot-r-o...

Interesting. I played with ZFS years ago on opensolaris, I don't remember there were atomic snapshots. I definitely need to dive into ZFS again.

But if we "stop" this database/schema, make ZFS snapshot (I suggest that the "partition" keeps only this database related files) it may work. It can be useful for testing purposes I guess.

Reply on news.ycombinator.com