CSE Roadmap

Subject

Databases

Modelling, indexing, transactions and scale.

Start at Level 0Interview questions

Foundations

What a database is, the vocabulary of tables and keys, and the running example used throughout.

  1. Level 0

    How databases work

    What a database management system actually is, the vocabulary of tables, rows and keys, where SQL fits, and the small bookstore database used as a running example for the rest of this roadmap.

    10 modules · about 2 weeks

SQL

Querying and changing data, from a single SELECT to views, window functions and recursive queries.

  1. Level 1

    SQL fundamentals

    The core of SQL: filtering and sorting rows, the relational algebra and calculus behind a query, every kind of join, grouping and aggregation, subqueries, set operations, three-valued logic with NULL, and changing data safely.

    9 modules · about 2 weeks

  2. Level 2

    Advanced SQL

    Views and materialized views, window functions, common table expressions and recursive queries, stored procedures and triggers, integrity constraints, a first look at indexes from the SQL side, and authorization.

    7 modules · about 2 weeks

Data modeling and design

Designing a schema before any table exists, and normalizing it once it does.

  1. Level 3

    Data modeling and design

    Designing a schema before any table exists: entity-relationship modeling, mapping diagrams to tables, functional dependencies, normal forms through 5NF, decomposition, when to denormalize on purpose, and modeling data whose history matters.

    9 modules · about 3 weeks

Storage and indexing

What is actually on disk, and the structures that let a query avoid scanning all of it.

  1. Level 4

    Storage, the physical layer

    What is actually on disk underneath a table: storage media, blocks, file organization, the buffer pool that caches pages in memory, row-oriented versus column-oriented layout, and the real path a write takes to become durable.

    6 modules · about 2 weeks

  2. Level 5

    Indexing

    Why an unindexed query must scan everything, B+-trees and hash indexes, LSM-trees for write-heavy workloads, composite and covering indexes, bloom filters and bitmap indices, when an index actually hurts, and reading an execution plan to check any of it.

    8 modules · about 2 weeks

Query processing and optimization

What happens between a SQL statement and a result set, and how to help the optimizer.

  1. Level 6

    Query processing and optimization

    The stages between a SQL statement and a result set: turning SQL into a query plan, the different join algorithms an optimizer can pick, sorting costs, cost-based optimization and why its estimates can be wrong, writing index-friendly queries, pagination that doesn't slow down on later pages, and materialized views.

    7 modules · about 2 weeks

Transactions and concurrency control

Making concurrent changes safe, precisely, not just by hoping nothing collides.

  1. Level 7

    Transactions and concurrency control

    A precise definition of ACID, when an interleaving of concurrent transactions is actually safe, what each isolation level really prevents, lock-based and timestamp-based protocols, deadlocks, multiversion concurrency control, the anomalies snapshot isolation still allows, serializable snapshot isolation, and picking the right isolation level for a real application.

    10 modules · about 3 weeks

Recovery and durability

Surviving a crash, a bad disk and a human mistake.

  1. Level 8

    Recovery and durability

    The kinds of failure a database has to survive, the write-ahead log that makes crash recovery possible, checkpoints that bound how far back recovery looks, the ARIES algorithm in outline, backups and point-in-time recovery, and keeping a remote standby ready to take over.

    6 modules · about 2 weeks

Distributed data

Replication, partitioning and the real difficulty of machines that talk over a network.

  1. Level 9

    Replication

    The leader-follower replication shape, the concrete glitches asynchronous replication lag causes and how each is fixed, accepting writes in more than one place with multi-leader replication, removing the leader entirely with leaderless replication and quorums, and choosing a topology for a real requirement.

    5 modules · about 1 week

  2. Level 10

    Partitioning and distributed transactions

    Splitting one dataset across many machines, indexing partitioned data, rebalancing as a cluster changes shape, why unreliable networks and clocks make distributed systems genuinely hard, two-phase commit, what consensus actually solves, and the CAP theorem used carefully rather than as a slogan.

    7 modules · about 2 weeks

Alternative data models

Document, key-value, graph and column-family stores, and when each actually fits.

  1. Level 11

    NoSQL and alternative data models

    Document databases, key-value stores, graph databases and column-family stores as four different ways to model the same data, each suited to a different access pattern, and how to pick a model for a real workload rather than a trend.

    5 modules · about 1 week

Big data and analytics

A second access pattern, built for asking questions across history instead of serving one request.

  1. Level 12

    Big data and analytics

    OLTP versus OLAP as two different access patterns that both call themselves queries, data warehousing and the star schema, MapReduce and distributed batch processing, stream processing for events as they arrive, and column-oriented storage built for analytical queries.

    5 modules · about 1 week

Practice and interviews

Tuning, choosing, designing and talking through it out loud.

  1. Level 13

    Practical performance, operations and interviews

    Measuring load and reasoning about scalability, tuning a genuinely slow query end to end, a decision framework for choosing a database for a workload, a full worked database system-design interview case study, and a capstone designing this very platform's own database.

    5 modules · about 2 weeks