Java ORM test

Initial assumptions

• JPA is f***ing slow, due too many layers of abstraction
• NO xml and other shit-styled non-code-like sources must be needed
• no additional tools to work (only boxed with technology set or none is applicable)

Initila list of techs

• Apache Cayenne 📝 🧰 ✨
• Apache OpenJPA 🧬 ✨
• Ebean 🧱 🧰 ✨
• jOOQ 💳 📝 🧰
• speedment 💳 📝
• ORMLite Core
• MyBatis 📝
• Querydsl
• Hibernate 🧬 🩹 ✨
• EclipseLink 🧬 ✨
• DataNucleus 🧬 🦠 ✨

Initial criteria

Must:

• encrypted columns
• no raw SQL for simple CRUD operations
• support for PostgreSQL

Good to see if:

• support for MariaDB
• support extended PostgreSQL features (including json and array)
• "N+1" problem automatic solutions (as much as possible)
• ability to generate raw SQL (to be executed elsewhere)
• support for binary (BLOB) storage and extraction (10Mb per record max)

Test plan

• postgres version 14 & 14-alpine (test database performance baseline)
• two databases (one with partitioning, one without)
• tables per database
  • xxs [50 records]
  • xs [500 records]
  • s [5000 records]
  • m [50000 records]
  • l [500000 records]
  • xl [5000000 records]
  • xxl [50000000 records]
• each test module starts a fresh copy of database
• database content is procedure-generated
• random sequences must be pregenerated to be exactly same for all tests (to eliminate test-path diviations)
• random sequence for each test is unique and not intersect (to eliminate postgresql optimizations)

Requests plan

• insert
  • single
    • correct data
    • constrain violation
  • batch
    • correct
    • invalids (by hand to check if full transaction will roll back)
• find
  • by equals (1)
    • integer
    • string
    • date
  • by more X & less Y (N)
    • integer
    • float
    • date
• update
  • random by id (single)
  • by criteria (batch)
• delete
  • random by id (single)
  • by criteria (batch)

Documentation analysis

• MyBatis - out of competitions, due to ass-whipping-xml notation.

Initial project setup analysis

• speedment - cool-looking streaming API, but it's way of from common SQL-like syntax & failes to work with UUID columns corectly, also they wanna' our money, dude if we need something not straightly mapped. Out of competition.
• jOOQ - enterprise style peace of ..., but still in.

Article cross analysis

Sources:

• Performance Evaluation of JPA Based ORM Techniques - Hibernate and EclipseJPA are best of JPA realizations (2016)
• Optimizing JPA Performance: An EclipseLink, Hibernate, and OpenJPA Comparison - Hibernate looks better in comparison (2010)
• JPA Performance Benchmark - good starting point for testing JPA-based ORMs. Best parter for PostgreSQL denoted here is EclipseLink and Hibernate (2012)
• Why JDO ? - feature comparison table for JDBC, EJB, JPA, JDO, etc.

Tools:

• HibernateTools - Reverse Engineering and Code Generation - how to generate classes for Hibernate from plain DB.

Honorable mentioning

• EJB 4.0 (now stands for Jakarta Enterprise Beans) is supported only in GlassFish 6 which is a part of Eclipse Foundation family. It's not included in research due to high binding to one particular implimentation with low chance of change in future.
• EOF (Enterprise Objects Framework) is a part of Apple now (originaly designed in NeXT). It supports java from back in a days, but now it's obsolete as of April 2005 (rise of Objective-C and Mac OS Tiger Intel-epoch)
• Oracle Kodo - JPA-based framework from Oracle, does not compete due to Oracle involvment and price tag.
• Oracle TopLink - part of Oracles' Fusion Middleware, was donated to Eclipse Foundation in 2007, and renamed to EclipseLink, so TopLink is obsolete now.