wmfiv

That's great. Thanks for the response and I'll keep an eye out for the blogs.

Are there published benchmarks for multi-key operations and more complex SELECT statements? I apologize if I missed them.

I'm trying to determine whether there's a place for Cockroach within what I think are the constraints in the database space.

* Traditional SQL Databases

  - Go to solution for every project until proven otherwise.

  - Battle tested and unmatched features.

  - Hugely optimized with incredible single node performance.

  - Good replication and failover solutions.

* Cassandra

  - Solved massive data insert and retention.

  - Battle tested linear scalability to thousands of nodes.

  - Good per node performance.

  - Limited features.

It seems like many new databases tend to suffer from providing scale out but relatively poor per node performance so that a mid-size cluster still performs worse than a single node solution based on a traditional SQL database.

And if you genuinely need huge insert volumes, because of the per node performance you'd need an enormous cluster whereas Cassandra would deal with it quite comfortably.

I'm not sure. How much does Mozilla actually invest in Rust? Same with Google and Go. I would guess over the past 15 years that Sun/Oracle (Java) and Microsoft (C#/.NET) each annually spend more on language / platform development than everyone else put together does over 5 or 10 years.

Something like http://www.easybatch.org/ may provide quite a bit of the infrastructure you're looking for. I'm sure there are equivalents for other languages.

In addition to the options identified in the article definitely take a look at ScalaJS and ClojureScript. I've used ScalaJS quite a bit and it's a more or less flawless experience.

For people interested in functional programming, I think Scala still has the following things going for it,

- Built in types (immutable persistent collections, Try, etc). You can get these in Java with http://www.javaslang.io/ but it's a relatively new library.

- A concise syntax for creating value types without using something like Lombok.

- Mature libraries (ScalaZ, Monocle, shapeless, etc) for people interested in more sophisticated functional patterns.

Axon Framework is a very mature solution for ES on the JVM.

http://www.axonframework.org/

Here's the section of their manual related to this. http://www.axonframework.org/docs/2.4/repositories-and-event...

Akka Persistence is the other JVM ES framework that I'm familiar with. it supports upcasting as well.

http://doc.akka.io/docs/akka/2.4/java/persistence-schema-evo...

I believe the default for SQL Server is actually read committed.

https://msdn.microsoft.com/en-us/library/ms175909.aspx

If you're a SQL fan and using the JVM, JOOQ is pretty much the best thing since sliced bread. It has excellent cross database window support.

http://www.jooq.org/doc/3.8/manual/sql-building/sql-statemen...

Personally I value mature, well understood solutions. Angular delivers that and the ecosystem around it is finally about there. I can use ui-bootstrap (Angular material not as much) on a relatively large application without running into defects at every step. And there is an enormously high probability that the answer to any issue my team runs into is one Google away.

I'd much rather stay on the well beaten path with a mature, less sexy, less efficient solution than hack my way through the jungle with something more contemporary and powerful but less mature and less well understood.

There's likely a significant difference in cost and capacity between copying a PB of data from a corporate datacenter over that corporation's connectivity to S3 and copying from S3 to Google over Amazon's and Google's connectivity.

I should have been more clear. I was referring to storage auto-scaling. Aurora gives you access to up to 64TB of disk billed in 10GB increments. You don't need to preallocate and manage the scaling of your disk as you do with RDS.

Storage cost, performance and auto-scaling are also big advantages.

Absolutely it is. As long as you're transparent with your employees about the opportunity and the risks.

Postgres doesn't provide a store engine API. You can achieve some of those goals by using the foreign data wrapper (fdw) api.

https://www.postgresql.org/docs/9.5/static/postgres-fdw.html

For example, Citus Data provides a column store for Postgres via the fdw api.

https://github.com/citusdata/cstore_fdw

Yeah. I'm not suggesting that it's an A+ job well done. But at the same time, relative to the target they are, seems like they're doing something right.

I'm genuinely interested in why not? Maybe the way I wrote it was a bit flippant, but surely bank computers that control the flow of trillions of dollars are a huge target. The fact there has never (?) been a massive breach that resulted in billions beings stolen must be a sign that someone is doing something right. No?

Isn't that the point? If you can't monetize a breach of their network then it seems like they're doing their job.

Seems like the banking industry has at least a degree of competency here as well. Who would bother with identity theft if you could just hack the banks and steal the money directly?

Isn't linear scalability a hugely desirable feature? Especially for a SQL RDMS like Postgres?

I would think it's unrelated. The issue described by Uber and the Postgres team is specifically related to UPDATEs against highly indexed tables. In an event store the data is typically completely immutable.

Akka HTTP is very nice.

I don't believe it's a technical problem. It just hasn't completed the internal AWS process for HIPAA certification to be added to their BAA. Postgres is in the same boat.

Not using the out of the box solutions. But while I haven't done this personally my understanding is Azul Zing will allow you to efficiently use multi TB heaps in Java.

What is too big? Too many concurrent jobs? We found the integration with AWS to autolaunch workers to be quite easy to setup.

Fair enough about the bespoke products.

But does anyone really do this? Are there published examples of people doing this or talking about doing it? I can find plenty of examples of hedge funds betting on single mid or late stage startups but not doing what you're suggesting here.

What you're suggesting is incredibly interesting so I don't intend to be rude but the concept strikes me as ludicrous. The assumptions that would go into a financial model for something like this seem ridiculous to me. You'd be betting/guessing/hedging on countless company, industry and macro-economic variables over a long, long time frame.

I believe transactions must process against a single node.

In a completely theoretical way what you say is possible. But in practice does anyone actually do this?

What's the time frame you'd be willing to stake your career on for a startup (event a fairly mature one) to be meaningfully impacting the overall profitability of it's astronomically larger competitors? And you'd be simultaneously betting against overall bull markets and that the incumbents won't adapt, whether in the market directly against the startup or in any of the other markets they're in.

To my knowledge you can't even trade in stock futures with sufficiently distant expiration dates to do this.

So what changed?

RLS doesn't require the use of a db user. For example you could use a session variable to identify your logical user and use that in the RLS query.