You could setup a replication to another PG server that is setup on a COW file system like zfs. Then you can take snapshots on the replica.
On Mon, May 4, 2026 at 5:59 AM masheed ullah <[email protected]> wrote: > Thank you every one for your output. > > Although pgBeckRest has no more support. > Should I suggest starting testing backup and restore with pgBectRest? > I did not find any 3rd Party tool. > > On Mon, May 4, 2026 at 12:29 PM Ilya Anfimov <[email protected]> wrote: > >> On Sun, May 03, 2026 at 12:42:21PM +0300, masheed ullah wrote: >> > Hi, >> > We have a database of 20TB, and it's taking almost 24 hours to >> complete. >> > While full restore takes 28 hours. Although we have HA enabled with >> > primary and multiple replica's. >> > So we need a solution/tool to reduce the RTO, incase of disk failure/ >> > Human error for data/tables deletion. >> > >> > We are using Barman for backup. >> >> First, you'd need engineers to measure -- what's going on. >> Who is the slowest part? The network (most probable)? The post- >> gres replication process? The disk that postgres reads? The bar- >> man itself? Receiver disks/S3 daemon/etc? TLS of one of the >> above? >> Measures is the key. Then the solutions may be searched. >> >> btw, a modern postgres can definitely supply 1GB/s to barman on a >> modern system. This is not a much, and probably could be im- >> proved by special tricks like rsync or manual prefetch or incre- >> mental copies -- however, it's less than 6 hours estimate and >> therefore is fine for you. >> >> > My question is "Are there any tools like ZDLRA appliances for >> Oracle" to >> > reduce the RTO to less than 10 hours". >> > Best Regards, >> > Khattak >> >> >> > > > >
pg-zfs-snapshot.sh
Description: application/shellscript
