Oracle 19c in Practice + New Features in 23ai/26ai — PL/SQL, Zero-Downtime Changes, Incident Diagnosis, and Vectors

I’ll learn SQLP practical tuning by reproducing the execution plans from the course on my own PC using Oracle 26ai Free. I’ll evaluate 30 hands-on practice problems I created myself using measured figures (blocks read and execution counts), then write out the answers as well.

After learning SQL tuning, automate recurring tuning tasks. This hands-on project connects everything from collecting MySQL queries to AI diagnosis, empirical validation, handoff to the person in charge, and tracking recurrence.

Ahead of a major event, what should you prepare in your database, and what should you check first when it goes live? Practice your pre-, during, and post-event response sequence using training Aurora logs of Reader CPU spikes, lock waits, deadlocks, and failovers, along with diagnostic SQL.

We fix product and event APIs that buckle under traffic at sale launch with local L1 and remote L2 caches, stampede prevention, invalidation, and a kill switch—and prove with before-and-after numbers how much bandwidth, latency, and database load drop under the same traffic.

What should you prepare in PostgreSQL ahead of a large-scale event, and what should you check first when it goes live? Practice the response sequence before, during, and after the event with hands-on experiments in educational Aurora PostgreSQL that trigger lock waits, deadlocks, VACUUM delays, failovers, and Reader CPU spikes—along with diagnostic SQL and playbooks for seven event types.

We take a monthly seller settlement batch for a virtual shopping mall from start to finish. Using Pro*C fundamentals, we read the source, fix four hidden defects with minimal patches, reduce database round trips from 265,893 to 359 through array processing, then port it to ECPG and JDBC and prove that all three outputs are identical, line for line. We run every command directly against Oracle Database Free and PostgreSQL in Docker.

From sign-up to emails, meeting minutes, reports, PDFs, spreadsheets, and presentation materials, follow hands-on exercises set in a fictional company to learn how to delegate tasks to Claude, compare the results with the originals, and use them with confidence.

We’ve learned how to use Claude Code, but how do we use it with our legacy system? We add a safety net to a Spring Boot order service with zero tests, catch five hidden bugs, and validate AI-written tests with mutation testing.

Six bugs that pass H2 and mock tests but break only on production MySQL, Redis, and Kafka—reproduced and fixed one by one with Testcontainers integration tests. Plus container singletons and reuse, CI, and flaky test diagnosis.

After Kubernetes Basics: When Your Service Is Sick. Diagnose and fix issues by looking at symptoms like restarts, slowdowns, 502s, and stalled deployments, using only kubectl and metrics. Trigger eight failures yourself in a local Kubernetes environment on your laptop, then measure the before-and-after numbers under the same load.

I’ve learned the principles of database design, but how do I turn orders, refunds, and settlements into tables? We’ll evolve the schema over 41 steps so a single order can flow through payment, partial cancellation, returns, refunds, and settlement, meeting all 38 requirements one by one.

Replace synchronous coupon issuance and product registration APIs that buckle under sale-launch traffic with a queue, outbox, idempotent consumer, DLQ, and Temporal—and reproduce the before-and-after metrics (latency, failure rate, and DB commits) under the same load on my computer using Docker.

Reproduce firsthand the problems that arise in a microservice flow from order API → payment → inventory → Kafka → consumer → batch: traces breaking, growing to thousands of spans and silently being dropped, or disappearing from dashboards. Then fix them in symptom → diagnosis → resolution order, completing traces and dashboards that work end to end.

I build a DBA agent right on my laptop in Docker that measures MySQL and PostgreSQL every morning, creates a diagnostic report with A/B/C/N letter grades, analyzes problematic queries with EXPLAIN and drafts improvement requests, and reviews plans and costs for scaling connections and Readers up or down before a sale.

[Final Goal] Build an operational agent that handles the full cycle: fault injection → alert → initial AI analysis → runbook and remediation plan → human-approved action → postmortem draft. Run the virtual commerce platform 'Loop Market' with Docker and practice using Prometheus, Grafana, Loki, and Claude.

Build a development pipeline with Claude Code’s headless mode, where a single-line issue comment triggers a PR ready for review. Codify team rules in CLAUDE.md, skills, and hooks, and add three human approval gates, a review bot, a false-positive ledger, and a cost cap.

Final goal: Turn one of my repetitive tasks into a Claude skill → have a colleague install and use it → report the time saved in numbers. Without writing code, I’ll use materials from “Loop Market,” a fictional sports and outdoor marketplace, to run and review 12 role-specific skills myself, then complete one based on my own work for the capstone.

If you run your database in Docker, you’ll learn how to reproduce 30 scenarios on your own computer that can wipe your data or make it unreadable—from `down -v` and `prune` to folder names, password environment variables, 10-second shutdowns, memory limits, image tags, and volume-copy backups. You’ll read the database logs to find the cause, then learn how to restore your data using point-in-time recovery.

Before the Monday 9 a.m. meeting, delegate performance tracking for three ad channels, coupon reconciliation, review complaints, and inventory checks to Cowork, while people just review the reconciliation sheet. Using data from a virtual shopping mall, create about 40 results files organized by owner and five skills, and set up a scheduled run for every Monday at 8 a.m.

Turn half a day of month-end closing into 30 minutes of uploading files and checking results. Give Claude Cowork 16 sample channel files in Smart Store, Coupang, and Cafe24 formats, and build an integrated order sheet → settlement reconciliation (orders ↔ settlements ↔ bank statements, with zero unexplained discrepancies) → returns P&L → inventory and purchase orders → pending customer service list → month-end closing skill. Give instructions in Korean—no coding or formulas—and verify everything against the original files. Learn from 28 real-world cases and actual Claude execution logs.