Software engineering concepts explained with Mermaid diagrams — updated daily. Inspired by ByteByteGo.
The CAP Theorem states that a distributed system can only guarantee two of three properties: Consistency (all nodes see same data), Availability (responsiveness), and Partition tolerance (surviving network failures). Understanding CAP helps you make trade-off decisions when designing databases and services—you must accept that perfect consistency and availability cannot coexist during network outages.
Database replication is the process of copying and synchronizing data across multiple database instances, either synchronously or asynchronously. It matters because it enables high availability, fault tolerance, and geographic redundancy—so if one database fails, others continue serving requests and your system stays online.
A CDN caches your content on geographically distributed servers worldwide, so when a user requests a file, the closest server delivers it instead of the origin. This reduces latency for users globally and decreases load on your origin server, making your application faster and more scalable.
OAuth 2.0 is an authorization protocol that lets users grant third-party applications access to their resources on another service without sharing passwords directly. It matters because it decouples authentication from authorization, limits credential exposure, and enables seamless single-sign-on while reducing the blast radius if any application is compromised.
Sharding splits a single large database into smaller pieces (shards) spread across multiple servers. A shard router uses a key (e.g. user_id) to decide which shard holds the data. Reads and writes go only to the relevant shard — no full-table scans across all data.
A load balancer sits in front of your servers and distributes incoming requests so no single server gets overwhelmed. Round-robin sends requests in turn; least-connections routes to whichever server is least busy.
DNS turns human-readable names into IP addresses. The recursive resolver does the legwork — it asks the root, then the TLD (.com), then the authoritative nameserver, then caches the answer.
TLS negotiates a symmetric session key using asymmetric cryptography. The certificate proves the server's identity; the handshake produces a shared secret that neither side transmits directly.
When you type a URL, four parties cooperate to deliver the page in milliseconds. The DNS resolver does the work most people forget.
A language model never "writes" — it predicts the next token, one at a time, sampling from a probability distribution.
Modern software is a loop, not a line. The faster the loop runs, the smaller and safer each change is.
A typical 2026 fullstack request: cached at the edge, rendered close to the user, and only hits the database when nothing else can answer.