pg.ddx.io  pgbackrest/pgbackrest GitHub issues and pull requests  
help / color / mirror / Atom feed
From: dwsteele (@dwsteele) <noreply+dwsteele@github.com>
To: pgbackrest/pgbackrest <noreply+pgbackrest-pgbackrest@github.com>
Subject: Re: [pgbackrest/pgbackrest] PR #2883: RFC: Derive the info file key with PBKDF2 in repository format 6.
Date: Tue, 20 Oct 2026 07:48:21 +0000
Message-ID: <gh-pgbackrest-pgbackrest-2883-c6286080749@github.com> (raw)
In-Reply-To: <gh-pgbackrest-pgbackrest-2883@github.com>
References: <gh-pgbackrest-pgbackrest-2883@github.com>

> The cost of 600000 rounds of PBKDF2-HMAC-SHA256 was about 100 ms on my macbook

On my m4 it is 200ms. On server CPUs, which tend to be slower per core I think you can expect 400-500ms. Given the way people tend to provision systems with fractional cores it might be more like a second.

> the foreground archive-get and archive-push do not load archive.info

They do when when `archive-async=n` so that would be a fairly major performance regression. Unfortunately this setting is still the default and therefore the norm.

I feel like this is something that is going to cause performance issues and people are not going to figure out what they need to change to "fix" it.

Given all that how to you feel about the heuristics in this patch to detect strong-looking keys and set rounds to 1 automatically. Should we keep that or skip it?

One thing to note is that before we release this there will be a rekey command that will allow you to change the keys *and* the rounds after stanza-create. So a bad initial decision is fixable. I'm just worried many users won't get that far and will just think pgBackRest is slow.

view thread (10+ messages)  latest in thread

Message-ID: <gh-pgbackrest-pgbackrest-2883-c6286080749@github.com>
Permalink:  ../gh-pgbackrest-pgbackrest-2883-c6286080749@github.com/

 · 


This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox