☕VoxxedDays Zurich 2024

Cool Java Stuff

• CodeScene: Code Analysis / Code Health report, not free

• HTTPClient

• Structured Concurrency

• String Templates

• Simple FileServer (JDK 20+)

JPA & Hibernate

• Digma shows suspected n+1 problems for code & DB performance issues

• DB Connection Pool sizing: keep pool small, 10 connections is enough

  • Flexy Pool: helps to find the correct pool size

  • Spring Boot Datasource Decorator: used to make using FlexyPool easier

• Disable: jpa.open-in-view

• Use TransactionTemplate instead of @Transactional

  • can mark only part of a method to run in a transaction

• use getReferenceById instead of fetching data from DB

• n+1 problem: for every query we execute one more query to fetch data

  • solve with @Query to fetch additional fields or @EntityGraph

• @DynamicUpdate. only updates changed fields -> but costs more memory/cpu

• Hibernate Hypersistence Optimizer runs as a test and shows you bad db configurations -> not free

  • alternative: quickperf, doesn't support newest spring boot yet but has workaround

    • can make test with @ExpectedSelect/Delete etc. to verify the number of db calls

• Fetch projections instead of entities for reading data. Only fetch entites when we want to modify them.

Spring Boot

• Enable virtual threads: spring.threads.virutal.enabled = true (default false)

• Use TestContainers @Testcontainers annotation. Can connect to a docker container directly from the test.

• docker compose support in spring boot -> spring-boot-docker-compose

• store the spring boot fat jar in an exploded fashion in a docker container is better for performance (both boot and resource access)

Cool sites

• https://inside.java/

• https://dev.java/

Cool IntelliJ Stuff

• Distraction free mode for focused work -> shortcut

• set custom project icon -> icon.png in idea folder

• Live template cmd+j

  • probably not really needed that much anymore due to AI

• Debugging

  • Expression Breakpoints in line for streams/lambdas

    • In Debugger: Trace Current Stream Chain -> visualises how the chain looks like

  • Interface Breakpoints

  • shift-click on variable, then side gutter -> have advanced tooling for breakpoint, can evaluate e.g. x = null everytime when the code runs by the breakpoint

  • Analyze dataflow: can see how many branches reach the same code

  • Have breakpoint only active if previous non suspending breakpoint was reached, can help to only get to specific state of method

Cool AI Stuff

ChatGPT

Priming tips

• Tell it: start every response with >, to produce markdown, then you see when it runs out of context

• Improve personal base prompt to reduce length a bit

• Base prompt

  • no ai disclosure

  • tell it how it should respond: asciidoc, prefered libraries etc.

• prime the context with small-talk

Solving problems

• When uploading documents, tell it to summarise before starting work

• Images in pdfs etc. are ignored, upload them separatly and ask it to understand

• Tell it to ask questions about the task

• Ask it to repeat the task you want to solve

• Rephrase the prompt instead of making a new prompt to keep the context shorter and to prevent poisoning the session with bad information.

• Let the system first describe with text the context and then ask it e.g. for a plant uml diagram

  • Ask the model to first explain how it would solve the problem

Java ML

• Classical programming vs. machine learning

  • Classical programming -> we define the rules and the machine follows them to get an output

  • Machine learning -> we let the machine figure out the rules to automate the task

• Java Libs for ML

  • DL4J, DJL, Tribuo (Oracle), JSR381 (Image Recognition Spec)

• recommendation: DJL, it's an interface to the real engines (pytorch etc.)

  • has pretrained model for tagging images, etc. z.B: resNet

  • easy to integrate

Companies present at VoxxedDays

Redhat, ZKB, BJB, 47North, claranet, ELCA, 42Talents, Axa