Running PostgreSQL, Kong and Konga on a single Docker network, then registering a Service and a Route and checking that requests get routed. For each command I explain why the step is there and what the ports and environment variables mean.

Ahead of an ISMS audit, passwords had to be hashed one way. Here's how I migrated them on a live service without a batch job, using only MySQL's SHA2() and INSERT … SELECT, laid out as a plan that splits the test accounts in two: a pre-run test, the main query, and a post-run check.

With more than one server, synchronized can't prevent concurrency problems. We already had Redis, so I picked a Redisson distributed lock and wrapped it in a custom @RedissonLock annotation and an AOP aspect that handle the SpEL key, waitTime and leaseTime. This covers that build, what to watch for around failed lock acquisition and transaction ordering, and a concurrency test with three threads.

Ahead of an ISMS audit, we upgraded end-of-life Spring Boot 2 to 3.3 and Java 11 to 21. I first cleaned up dependencies on 2.7.18, the last 2.x release, then moved to 3.3. This covers the Gradle wrapper, the Querydsl jakarta setup, javax to jakarta, rewriting SecurityConfig in the Spring Security 6 lambda DSL, and the traps that are easy to miss.

Looking back on 2023, the year I looked for my first job as a backend developer: thinking algorithms were all I needed before graduating, the frustration of getting no feedback after, the three criteria I used to pick target companies, and how I changed my resume, up to joining Tripbtoz.

I thought I'd cleared out every N+1 with IN clauses, but a performance test showed one contact query per user again. Why switching @OneToOne to LAZY didn't stop it, how default_batch_fetch_size took the response from 124ms to 54ms, and why I didn't switch to @ManyToOne instead.

Writing one method with the position filter and one without means the number of queries grows with every combination of conditions. Using the fact that Querydsl ignores null conditions in where(), I merged a cursor-paginated query into one, split out the count query, and explain why the cursor key and the sort key have to match.

When a team leader opened the list of users looking for a team, the API looked up offers for each user in a loop and sent 22 queries. Here's how I folded the offer lookup into a single Querydsl IN clause, stitched the results back on with a Map, and cut it to 3 queries and the response time from 0.337s to 0.039s.

I'd been writing email duplicate checks with findBy out of habit. On 50,000 rows I logged the SQL that Spring Data JPA's findBy, countBy and existsBy actually generate, got 214ms, 196ms and 166ms, and explain why existsBy is the right fit for a duplicate check.