Backend Systems Architecture Flashcards

Question 1

What is the difference between a monolithic and a microservices architecture?

Accepted Answer

A monolithic architecture packages an application's functions into a single deployable unit that shares one codebase and one process. A microservices architecture splits the same functionality into small, independently deployable services that communicate over the network. Each approach trades simplicity of deployment against the operational complexity of managing many moving parts.

Question 2

What does it mean for a component to be loosely coupled?

Accepted Answer

A loosely coupled component depends on other components only through minimal, stable interfaces, so it knows little about their internal details. Changes inside one component can then be made without forcing changes elsewhere. Loose coupling makes systems easier to test, replace, and evolve over time.

Question 3

What is cohesion, and why does high cohesion matter in backend design?

Accepted Answer

Cohesion describes how closely the responsibilities inside one module or service relate to a single purpose. High cohesion means each module does one well-defined job, which makes its code easier to understand, change, and reuse. Low cohesion tends to produce tangled modules where unrelated logic must be modified together.

Question 4

What is the single responsibility principle, and how does it apply to services?

Accepted Answer

The single responsibility principle states that a unit of code should have one reason to change. When applied to services, each service should own one business capability and its data, so that changes to that capability stay contained. This idea supports both cohesion and independent deployment.

Question 5

What is an abstraction boundary, and why do architects define one?

Accepted Answer

An abstraction boundary is a line between parts of a system where one side exposes a defined interface and hides its internal implementation. Architects define these boundaries so teams can change internals without breaking callers. Clear boundaries also limit how far a bug or design change can spread through the codebase.

Question 6

What is the difference between a layered architecture and a hexagonal architecture?

Accepted Answer

A layered architecture organizes code into stacked tiers, such as presentation, business logic, and data access, each calling the tier below. A hexagonal architecture places the core business logic at the center and connects it to outside concerns through ports and adapters. Hexagonal designs make the core easier to test in isolation.

Question 7

What is the role of an API gateway in a backend system?

Accepted Answer

An API gateway is a single entry point that receives client requests and routes them to the appropriate backend services. It handles cross-cutting tasks such as authentication, rate limiting, request logging, and response aggregation, so individual services do not repeat that logic. The gateway can hide the internal service layout from external clients.

Question 8

Why do backend architects often prefer stateless services over stateful ones?

Accepted Answer

A stateless service does not keep client session data in memory between requests, so each request carries or references all the context it needs. This lets any instance handle any request, which makes horizontal scaling and failover simpler. The trade-off is that session context must be stored elsewhere, in a database or cache.

Question 9

What is the difference between synchronous and asynchronous communication between services?

Accepted Answer

In synchronous communication, a caller sends a request and waits for the response before continuing, such as HTTP or gRPC calls. In asynchronous communication, the caller sends a message and continues without waiting, while the receiver processes it later. Asynchronous messaging reduces coupling in time but makes responses and error handling harder to follow.

Question 10

What is a REST API, and what does it mean for it to be resource oriented?

Accepted Answer

A REST API is an interface style that exposes data as resources, each identified by a URL and manipulated through standard HTTP methods such as GET, POST, PUT, and DELETE. Being resource oriented means clients interact with nouns like orders or customers rather than with remote procedure names. It relies on stateless requests and predictable conventions.

Question 11

What is idempotency, and why does it matter for retried requests?

Accepted Answer

An operation is idempotent when performing it several times produces the same result as performing it once. This matters for retried requests because network timeouts often leave a client unsure whether the first attempt succeeded. Designing payment or order endpoints to be idempotent, often with a client-supplied key, prevents duplicate charges or duplicate records.

Question 12

What is an API versioning strategy, and why is it needed?

Accepted Answer

An API versioning strategy is a planned method for releasing changed interfaces while keeping existing clients working. Common approaches place a version in the URL path, in a request header, or in a media type. Versioning is needed because clients update on their own schedules, so breaking changes can disrupt users who have not migrated yet.

Question 13

What is a message queue, and how does it help decouple producers from consumers?

Accepted Answer

A message queue is a buffer that stores messages from producers until consumers are ready to process them. It decouples the two sides, so a producer can keep sending work even when a consumer is slow, offline, or being scaled up. Queues also smooth traffic spikes and let several consumers share the workload.

Question 14

What is the difference between at-least-once and exactly-once message delivery?

Accepted Answer

At-least-once delivery guarantees that every message arrives, but it may arrive more than once after retries, so consumers must tolerate duplicates. Exactly-once delivery means each message takes effect only one time, which is much harder to guarantee across networks and separate systems. In practice, many teams accept at-least-once delivery and make their consumers idempotent.

Question 15

What is a circuit breaker pattern, and what problem does it solve?

Accepted Answer

A circuit breaker wraps calls to a remote service and stops sending requests once failures pass a threshold, returning a fast error or fallback instead. It solves the problem of cascading failure, where one slow dependency ties up threads and resources. After a cooldown period, the breaker allows trial requests to test whether the dependency has recovered.

Question 16

Why do teams write explicit schemas for service contracts before implementing them?

Accepted Answer

Explicit schemas define the request fields, response shapes, and error formats a service promises to offer. Teams agree on these contracts first so that consumers and providers can build in parallel without guessing. Machine-readable schemas also enable automatic validation, generated client code, and documentation that stays closer to the real behavior.

Question 17

What is the difference between a relational database and a NoSQL database?

Accepted Answer

A relational database stores data in tables with defined schemas and relationships, and it typically supports SQL queries and transactions across rows. A NoSQL database is a family of systems, such as document or key-value stores, that often trade rigid schemas for flexibility or scaling. The right choice depends on access patterns and consistency needs.

Question 18

What does ACID mean in the context of database transactions?

Accepted Answer

ACID is an acronym for atomicity, consistency, isolation, and durability, the four properties that reliable database transactions aim to provide. Atomicity means a transaction completes fully or not at all, and durability means committed changes survive crashes. Consistency and isolation ensure valid rules are preserved and concurrent transactions do not interfere in harmful ways.

Question 19

What is eventual consistency, and where is it commonly accepted?

Accepted Answer

Eventual consistency means that replicas of data may temporarily disagree after an update, but they converge to the same value if no new updates arrive. It is commonly accepted in systems where brief staleness is tolerable, such as social media feeds or caches. It trades immediate agreement for higher availability and lower latency.

Question 20

What is a database index, and what is its main cost?

Accepted Answer

A database index is a separate data structure that lets the engine locate rows matching a condition without scanning the whole table. Its main cost is extra storage plus slower writes, because every insert, update, or delete must maintain the index. Indexes should be added based on real query patterns rather than on every column.

Question 21

What is database sharding, and what challenge does it introduce?

Accepted Answer

Database sharding splits a large dataset horizontally across multiple database instances, with each shard holding a subset of the rows. It lets storage and write load scale beyond a single machine. The main challenge is that queries and transactions spanning several shards become harder, and rebalancing data later can be complex.

Question 22

What is a read replica, and what risk does it bring for reads right after writes?

Accepted Answer

A read replica is a copy of a primary database that receives changes asynchronously and serves read queries to spread the load. Because replication takes time, a read sent to a replica right after a write may return stale data. Applications often route such reads to the primary or wait for replication to catch up.

Question 23

What is the CAP theorem, and what does it say about network partitions?

Accepted Answer

The CAP theorem states that a distributed data system facing a network partition must choose between consistency and availability, because it cannot fully guarantee both at once. Partition tolerance is not optional in a real distributed network, so the practical choice is between returning consistent answers and remaining responsive. The theorem is often oversimplified, so designers should consider latency and failure modes as well.

Question 24

How do database transactions help an order service avoid partial updates?

Accepted Answer

Transactions group several writes into one unit, so they either all commit or all roll back on failure. For an order, the service can insert the order record, reserve inventory, and record a payment row inside one transaction. If any step fails, the database discards the earlier changes.

Question 25

What is the difference between vertical and horizontal scaling?

Accepted Answer

Vertical scaling adds more resources, such as CPU, memory, or storage, to a single machine. Horizontal scaling adds more machines or instances that share the workload. Vertical scaling is simpler but has a hard upper limit and often a single point of failure, while horizontal scaling usually requires stateless design or careful data partitioning.

Question 26

What is load balancing, and which algorithms are commonly used?

Accepted Answer

Load balancing distributes incoming requests across a pool of servers so that no single server becomes overloaded. Common algorithms include round robin, which rotates through servers in order, and least connections, which sends new requests to the server with the fewest active connections. Balancers also use health checks to skip failing instances.

Question 27

What is caching, and what is the difference between cache-aside and write-through strategies?

Accepted Answer

Caching keeps copies of frequently read data in a faster layer. With cache-aside, the application checks the cache first, loads from the database on a miss, and fills the cache itself. With write-through, every write goes to the cache and the database together, which keeps them aligned at the cost of slower writes.

Question 28

What is a rate limiter, and why do backend systems use one?

Accepted Answer

A rate limiter restricts how many requests a client, key, or endpoint can make within a time window. Backend systems use one to protect shared resources from overload, abusive traffic, and runaway clients. When the limit is exceeded, the server rejects extra requests with an error response, so that well-behaved clients can still be served.

Question 29

What is a health check, and how does it support reliability?

Accepted Answer

A health check is a simple probe that asks a service whether it is running and able to handle work. Load balancers and orchestrators use these results to stop sending traffic to unhealthy instances and to replace them. This supports reliability by isolating failures quickly, before users are routed to a broken instance.

Question 30

What is the difference between retries with exponential backoff and simple retries?

Accepted Answer

Simple retries resend a failed request immediately or at a fixed interval, which can overwhelm a service that is already struggling. Retries with exponential backoff wait progressively longer between attempts, giving the dependency time to recover. Adding random jitter to those waits further prevents many clients from retrying in synchronized waves.

Question 31

What do the terms availability and durability mean in reliability engineering?

Accepted Answer

Availability is the proportion of time a system is able to respond to valid requests. Durability is the assurance that data, once accepted, is not lost, even after hardware failures or crashes. A system can be highly available yet still lose data, so the two properties must be designed and measured separately.

Question 32

Why do teams design graceful degradation for partial outages in backend systems?

Accepted Answer

Graceful degradation means a system keeps its core functions working when some dependencies fail, instead of failing completely. For example, a shopping site might hide personalized recommendations and still allow checkout when the recommendation service is down. Teams design for it because partial service usually serves users better than a total outage.

Question 33

What are the three commonly cited pillars of observability in backend systems?

Accepted Answer

The three commonly cited pillars are logs, metrics, and traces. Logs are timestamped records of discrete events, metrics are numeric measurements aggregated over time, and traces follow a single request as it moves through multiple services. Together they help engineers understand what a system is doing and why it behaves as it does.

Question 34

What is a distributed trace, and what is a span?

Accepted Answer

A distributed trace records the path of one request across several services, identified by a shared trace identifier. A span represents a single timed operation within that trace, such as a database query or an outgoing HTTP call, and spans nest to show parent and child relationships. Comparing span durations reveals where latency is concentrated.

Question 35

What is the difference between a metric's counter and gauge types?

Accepted Answer

A counter is a value that only increases, such as the total number of requests served, and it is usually read as a rate over time. A gauge is a value that can rise and fall, such as current memory use or the number of open connections. Choosing the right type prevents misleading graphs.

Question 36

Why do alerts based on symptoms usually work better than alerts based only on causes?

Accepted Answer

Symptom-based alerts fire when users are affected, such as high error rates or slow responses. Cause-based alerts, such as one host's high CPU, may fire often without any user impact, which leads to alert fatigue. Pairing symptom alerts with cause-level dashboards lets responders move from what is failing to why it is failing.

Question 37

What is authentication, and how does it differ from authorization?

Accepted Answer

Authentication verifies who a caller is, usually by checking credentials such as a password, token, or certificate. Authorization decides what that verified caller is allowed to do, such as reading a particular record or calling an admin endpoint. A request can pass authentication and still be denied by authorization.

Question 38

What is the principle of least privilege, and how is it applied to service accounts?

Accepted Answer

The principle of least privilege means every user, process, or service receives only the permissions it needs to perform its function. For service accounts, this means granting access to specific tables, queues, or APIs rather than broad administrative rights. Limiting permissions reduces the damage a compromised credential can cause.

Question 39

What is input validation, and why should it happen on the server side?

Accepted Answer

Input validation checks that data received from a client has the expected type, format, length, and range before the system uses it. Server-side validation is essential because client-side checks can be bypassed by anyone sending requests directly. Validating on the server also protects against injection attacks, where malicious input is interpreted as commands.

Question 40

Why should secrets such as database passwords stay out of source code?

Accepted Answer

Source code is shared widely across developers, build tools, and version history, so secrets stored there can leak easily and remain exposed even after deletion. Secrets should instead be injected at runtime from a dedicated secrets manager or environment configuration. Rotating a credential is also far simpler when it lives outside the code.

Backend Systems Architecture Flashcards

1Core Architecture Principles

2Service Design and Communication

3Data Storage and Consistency

4Scalability and Reliability

5Observability and Security

Learning Resources