[bind10-dev] MySQL vs PosgreSQL data source flamewar - FIGHT!

[JINMEI Tatuya / 神明達哉]

At Thu, 10 Jan 2013 14:24:29 +0100,
Shane Kerr <shane at isc.org> wrote:

> We knew SQLite had serious limitations from the beginning, and we are
> now hitting up against them.
> 
> Mostly these are to do with poor handling of concurrent operations.
> This is not a problem for smaller zones with minimal updates, but for
> zones with a lot of updates - especially large ones - this is an issue.
> 
> Rather than spend a lot of efforts making hacks with more-or-less work
> around the problem, we should spend this time implementing support for
> an SQL data source that does not have these limitations. We intended
> this from the very beginning.
> 
> The obvious choices are MySQL and PostgreSQL.
> 
> For the record, I am happy with either, and we will ultimately support
> both.

If the idea is to support both largely at the same (e.g., in the same
quarter), I don't have a strong opinion.  Except for this point

> * We are missing a bit of functionality:
>   https://lists.isc.org/pipermail/bind10-dev/2012-October/003868.html
>   We can work around this, but it is additional work.

it shouldn't be difficult to support either of them (I needed 1-2 days
to write my experimental backend for each of them).

To me, a more important point is to use this opportunity to fix
various schema related issues, including

- revisiting the representation of names (currently it only works with
  trivial alpha-numeric characters).
- revisit schema for the zone table.  there's an idea of revising it
  for higher performance.
- maybe using numeric for some parameters
- or maybe using binary blob for some

and also to introduce other performance related enhancements such as
internal caching.

And, at a higher level, it's better to have more confidence about
common use cases with the DB backend.  I've repeatedly requested that
information, but apparently we are failing in it (except for one
pointer you previously provided).  If it was because we were not so
serious about it, it's the time to be so.  If it's really because such
information isn't available, maybe we need to try whatever we come up
with and see if it's accepted by real users.

---
JINMEI, Tatuya