This project materializes few dreams in connection to Java database development:
- Eventual Atomicity for Cassandra
- Scenario: transfer money between accounts, need two updates. If second update fails then first one must be
compensated. - This was a primary goal of the entire project and gave project a name
- We beleive that in most cirsumstances it is possible to avoid transactions and use compensations instead
- Pros: No database deadlocks anymore
- So not only Cassandra, but any JDBC source, too
- Scenario: transfer money between accounts, need two updates. If second update fails then first one must be
- History. All operations are tracked automatically
- Binary States/Operations
- No tables anymore to keep states and operations
- Pure Java objects, Cassandra/JDBC as a data storage
- Only two tables are needed - one for operations, one for states, so easy to create an maintain
- Simplicity
- Everything above must be very simple in usage and testing
- Tests do not work with database directly
How to use this
How it works: interactions of core classes/modules
- Don't use In-Memory Databases (like H2) for Tests https://blog.philipphauer.de/dont-use-in-memory-databases-tests-h2/
- https://stackoverflow.com/questions/47655837/cassandra-filter-latest-rows-from-an-append-only-table
- having billions of partitions per node will cause slow down repairs and compactions significantly.
- but: https://stackoverflow.com/questions/30648479/choosing-a-partition-key-for-a-cassandra-table-how-many-is-too-many-partition
- and http://grokbase.com/t/cassandra/user/14c555q5t3/pros-and-cons-of-lots-of-very-small-partitions-versus-fewer-larger-partitions
- https://stackoverflow.com/questions/47631686/what-is-cassandra-static-fields-sense
- https://stackoverflow.com/questions/47354241/cassandra-and-aggregated-data
- https://books.google.ru/books?id=bh8_0FD3qgQC&pg=PA60
Miroservices
- https://stackoverflow.com/questions/39364466/how-to-dockerized-java-microservices-efficiently
- Say you give each JVM a 2 GB heap and add 1 GB for docker+JVM, you are looking needing a 64 GB server to run 20 JVMs/dockers.
- https://developers.redhat.com/blog/2017/03/14/java-inside-docker/
- https://dzone.com/articles/how-to-decrease-jvm-memory-consumption-in-docker-u
- http://trustmeiamadeveloper.com/2016/03/18/where-is-my-memory-java/
- What I can say as a conclusion? Well… never put words “java” and “micro” in the same sentence :) I'm kidding - just remember that dealing with memory in case of java, linux and docker is a bit more tricky thing than it seems at first.
- https://buoyant.io/2016/06/17/small-memory-jvm-techniques-for-microservice-sidecars/
- In this post, we’ll describe how we reduced the memory footprint of linkerd, our JVM-based service mesh for cloud-native applications, by almost 80%—from 500mb to 105mb—by tuning the JVM’s runtime parameters.
- https://stackoverflow.com/questions/38747525/spring-cloud-microservices-memory-usage
- And all of them consume over 350mb-500mb and sometimes crash, although they will not be used. I thought microservices should be lightweight. But with this memory usage
- atomyj, atomagic.org, atomey (atomey.com), atomie, atomeye