Morning

@mccraigmccraig do you use much tinkerpop? I was looking at the various gremlin dsls last night.

Writing a datalog -> gremlin bytecode compiler is appealing, mostly because I like datalog.

we’re not using any tinkerpop atm - i’ve been trying to avoid being committed to DSE, so whatever graph stuff we do is currently done with set ops

@mccraigmccraig what are set ops? I don't think you would be committed to DSE, as you can use janusgraph, or is that not true?

Morning, and hello @gingerwizard

@dominicm set ops as in joins on edge-sets... i hadn’t come across janusgraph though - that might be useful

@mccraigmccraig fun history lesson. JanusGraph is forked from TitanGraph. TitanGraph were bought by DSE.

@mccraigmccraig another question, what are oltp queries like on dse?

yeah, i knew about titan->dse... didn’t know there was still a live oss version tho

dse is cassandra plus some integrated bits (spark aka analytics, graph aka titan and search aka solr)...

we’re only using cassandra + spark

so our oltp is just cassandra - it’s fast and scales well, but modelling is very different from an rdb

Morning

Wasn't sure about cassandra with oltp queries at all. Hard to get a feel when most people use it for olap/big-data stuff. I'm wondering about if mixing oltp & olap data in cassandra is fine, especially if you have graph queries to do lovely work on them both.

C* people generally want to heavy reads on a replicated dB rather than on the one taking heavy writes

C*?

morning (near enough)

Oh C* is a Cassandra abbreviation.

@otfrom what do you mean by replicated db here? A materialised view?

moin moin

A separate analytical database cluster

I see. That explains the Spark+C* of DSE better

@dominicm generally you have a separate “datacenter” (c* terminology) for spark, still synced with all the other “datacenters” in your cluster so no need for any ETL

DSE has a metadata store impl for spark-sql too, do you can point BI tools like tableau st your c*+spark cluster

@mccraigmccraig I'm wondering if that plays nice with the graph stuff. Why metadata and not something else?

the metadata server is part of the DSE special sauce - spark+c* can be made to work with oss versions