Distributed Systems Engineering learning path: everything to study, topic by topic

Understanding the CAP theorem — Consistency, Availability, and Partition Tolerance — and the real-world tradeoffs distributed system designers must make when network partitions occur

The spectrum of consistency guarantees — from strict serializability to read-your-writes, monotonic reads, and session consistency — and when each is appropriate

How eventually consistent systems converge, conflict resolution strategies including last-write-wins and CRDTs, and the application design patterns needed to handle staleness

Linearizability as the strongest single-object consistency guarantee, its cost in latency and availability, and the distinction from serializability for multi-object operations

Single-leader, multi-leader, and leaderless replication architectures — their write paths, read paths, durability guarantees, and operational trade-offs

The problems caused by replication lag — read-after-write violations, monotonic read failures, causality violations — and the techniques to detect and mitigate them

How distributed systems elect and maintain a leader — including split-brain prevention, epoch numbers, fencing, and the role of external coordination services

Multi-datacenter replication topologies, circular and star topologies, write conflicts, and leaderless systems (Dynamo-style) with tunable quorums

Data partitioning strategies — range partitioning, hash partitioning, and directory-based — including hotspot avoidance, rebalancing, and cross-shard query challenges

How consistent hashing distributes load and data across nodes, virtual nodes for improved balance, and its application in distributed caches and databases

Load balancing algorithms — round-robin, least-connections, power-of-two-choices — health checking, session affinity, and L4 vs L7 load balancing trade-offs

Cache invalidation strategies, cache stampede prevention, write-through vs write-behind vs cache-aside patterns, and distributed cache consistency challenges

Designing systems that tolerate node failures, network partitions, and partial failures — including redundancy, graceful degradation, bulkheads, and failure mode analysis

Byzantine failures where nodes send conflicting or malicious information, the Byzantine Generals Problem, BFT consensus algorithms, and their applicability in production systems

Circuit breaker state machines (closed/open/half-open), timeout design, retry strategies with exponential backoff, and jitter to prevent thundering herd on recovery

Backpressure mechanisms to prevent producer/consumer overload, rate limiting strategies, load shedding, and reactive streams for flow-controlled pipelines

B-tree vs LSM-tree storage engine trade-offs, read amplification, write amplification, space amplification, and why each engine suits different workload profiles

How write-ahead logging provides crash recovery guarantees, WAL-based replication, checkpointing, and the interplay between WAL and storage engine durability

LSM-tree compaction strategies — size-tiered, leveled, and FIFO — their impact on read/write amplification, space usage, and latency spikes in production

Column-oriented storage layouts, compression techniques (run-length encoding, bitmap indexes), vectorized execution, and why columnar formats (Parquet, ORC) accelerate analytical queries

ACID transaction properties (atomicity, consistency, isolation, durability) contrasted with BASE (Basically Available, Soft state, Eventual consistency) and the design implications of each

Two-phase commit protocol for distributed atomicity, its failure modes (coordinator crash, blocking), and saga pattern as a choreography-based alternative for long-running transactions

Paxos and Raft consensus protocols for achieving agreement in the presence of failures, log replication, leader-based coordination, and production systems that implement them

Gossip-based dissemination for cluster membership, failure detection (SWIM protocol), and epidemic broadcast — how decentralized protocols achieve eventual consistency across large clusters