Re: old cluster does not use data checksums but the new one does

> On May 5, 2019, at 2:42 PM, Mariel Cherkassky <mariel(dot)cherkassky(at)gmail(dot)com> wrote:
>
> in aspect of upgrade, what will be faster, pg_upgrade or dump and restore each db in the cluster ?
>
> ‫בתאריך יום א׳, 5 במאי 2019 ב-19:11 מאת ‪Scott Ribe‬‏ <‪scott_ribe(at)elevated-dev(dot)com‬‏>:‬
> > On May 5, 2019, at 9:58 AM, Mariel Cherkassky <mariel(dot)cherkassky(at)gmail(dot)com> wrote:
> >
> > My only option is doing the upgrade with other tools and not with pg_upgrade ?
>
> Dump & restore; you're changing the on-disk file format, so attempts to use current disk files cannot work.

In my opinion, depends on your confidence in your disks & file system etc. For instance, given enterprisey hardware (channel per disk, no half-backed SATA multipliers, decent disks) + ZFS with appropriate redundancy, I wouldn't see the need. On the other hand, a desktop PC with some multi-disk USB box, and hell yeah I want checksums. Systems in between, harder judgment call...