Databases and Distributed Deadlocks: A FAQ

Re:Databases and Distributed Deadlocks: A FAQ by marknadal on Thursday August 31, 2017 @08:17PM (#15145164)

This is definitely a challenge and something you have to address when building on top of Postgres because of its architecture. But I want to highly encourage people that you can avoid having to deal with this stuff if you approach it with CRDTs (go google them, they are fun!), like what with Cassandra does - (edit: disclaimer, I work for a competing system) I did this tech talk ( http://gun.js.org/distributed/matters.html ) a few years ago in Berlin on the things to avoid and the possible solutions (and their tradeoffs / dangers).

[ Parent | on HN ]

Re:Databases and Distributed Deadlocks: A FAQ by Twisol on Thursday August 31, 2017 @08:49PM (#15145297)

I'm actually very interested in CRDTs and related approaches, and I've been exploring active research directions in the area, but I'm not sure they share many usage patterns with data stores like Postgres, which support very general access and update semantics.

The primary restriction for anything involving CRDTs is that no operation can "take back" something added by a previous operation, so the data in question can only evolve monotonically. It's great when you can get away with it, but it's unclear how often you can actually get away with it.

[ Parent | on HN ]

Re:Databases and Distributed Deadlocks: A FAQ by marknadal on Thursday August 31, 2017 @09:29PM (#15145466)

Yes, very true. Kyle Kingsbury was at the conference and I talked to him about this, and his conclusion from his research is that Postgres is still all around the best database out there. Especially if you need to do things that CRDTs aren't yet equipped to handle.

Have you made a compilation of your thoughts from the various research you've done? Would love to see them!

[ Parent | on HN ]

Re:Databases and Distributed Deadlocks: A FAQ by greenleafjacob on Thursday August 31, 2017 @10:09PM (#15145606)

Warning: Cassandra counters are not backed by CRDTs [1].

[1] https://aphyr.com/posts/294-jepsen-cassandra

[ Parent | on HN ]

Re:Databases and Distributed Deadlocks: A FAQ by iamaleksey on Friday September 01, 2017 @04:28AM (#15146981)

Except they are. Catalogued as δ-CRDT, under Lexicographic Counter in https://arxiv.org/pdf/1603.01529.pdf

edit: Although to be fair that blog post is ancient and covers Cassandra 2.0, now no longer supported. Counters were rewritten in Cassandra 2.1.

[ Parent | on HN ]

Re:Databases and Distributed Deadlocks: A FAQ by mayank on Thursday August 31, 2017 @09:06PM (#15145363)

It's generally good form here to put in a disclaimer when you're plugging your own product.

[ Parent | on HN ]

Re:Databases and Distributed Deadlocks: A FAQ by kaghyu1667 on Thursday August 31, 2017 @08:22PM (#15145185)

Please stop spamming this.

[ Parent | on HN ]