Hacker News Reader

Comment by eikenberry | Hacker News Reader

IMO the main advantage for RethinkDB is its HA story. Last time I had to manage a PostgreSQL cluster (2012-2013) its HA story was pretty bad. It was limited to a master-slave(s) setup with manual failover and manual cluster rebuilding all dependent on incomplete 3rd party tools. Has PostgreSQL improved on this? A quick googling leads me to believe it hasn't and I'd only even consider it again if it were managed by a 3rd party (eg. aws rds).

View on HN

Excellent point! My counter-argument to that would be: How many startups/SMEs embracing Rethink really need full four or five nines availability?

I'd expect the costs of running that HA stuff is an order of magnitude higher than the costs of even a few days of downtime p.a.. And migrating from Postgres to something with HA once needed is probably easier than migrating to Postgres if costs are killing you.

Yes PG 9+ have greatly improved replication.

Yes AWS does it for you.

This was with 9.x using the replication system. It was certainly better than the old wal forwarrding systems, but it still was a giant PITA.

It surely has improved (replication and failover are much easier now, than they used to be), but still not anywhere close to "just give other node's address and we're good". And that BDR thingy for master-master stuff is still an unofficial fork (although, as I get it, it's from one of main Postgres contributors, so it's really close)

It has improved quite a bit since then (in 9.2 or 9.3 when you last used it), but it is still not all the way there. Especially rebuilding the cluster is much easier now.

Amazon recently announced PostgreSQL support for Aurora. That should give you what you are looking for.

If you want a managed solution RedShift supports PGSQL syntax and connectors now I think. That's a fully managed turn-key DB, but of course it locks you into AWS.

Redshift isn't highly available (if a node goes down you have to wait for it to restore a new one, which takes hours) and only has a small subset of PostgreSQL features. It is really only intended for OLAP (analytics) processing.

Not meaning to be an ass, but this is really bad advice.

Redshift is not supposed to be used like Postgres. Redshift is a data warehousing solution, with completely different tradeoffs, and accommodating completely different work loads than your average Postgres database. For example, you can't create indices on Redshift tables or relationships between them, the consistency story is completely different from Postgres and Redshift is optimized for bulk loads, not millions of discrete inserts/updates a second.

Amazon RDS does provide hosted Postgres, and that is what you want to use if you want managed Postgres. Or you can use Herokus's hosted Postgres (it does get expensive with size, though.)

Postgres and Redshift should not be considered equivalent. It's true that Redshift originated as a fork of Postgres 8, but Postgres has come a _long_ way since then and is quite different.

On top of that, Redshift is columnar database, which is an entirely different animal than vanilla Postgres (or any other DBMS).

Ref: http://docs.aws.amazon.com/redshift/latest/dg/c_unsupported-...

https://en.wikipedia.org/wiki/Amazon_Redshift

Reply on news.ycombinator.com