ByteDance Backend Interview Experience: Key Topics and Knowledge Points

Interview Overview

Today I share a ByteDance backend interview experience covering three rounds; the candidate passed the first two but failed the third.

First Round Topics

Java: HashMap, exceptions, Spring circular dependencies, design patterns, synchronized

MySQL: indexes, SQL optimization, B+Tree

Operating System: processes and threads, soft/hard links, I/O multiplexing, fork principle, heap vs stack

Network: TCP/UDP, HTTPS handshake, HTTP/2, breakpoint resume, header fields

Hand‑written: algorithms (present in every round), Singleton pattern (appeared in third round)

HashMap Deep Dive

Difference between JDK 1.7 (array + linked list) and JDK 1.8 (array + linked list + red‑black tree when bucket size > 8, revert when < 6).

HashMap is not thread‑safe; to make it safe use synchronized, ConcurrentHashMap, ReentrantLock, ThreadLocal, etc.

In multithreaded environments JDK 1.7 may suffer linked‑list loops and data loss during resize; JDK 1.8 mitigates with tree bins but still has write‑overlap issues.

TLS Handshake Overview

Four‑step RSA‑based TLS handshake: ClientHello, ServerHello (certificate, randoms, cipher suite), ClientKeyExchange (pre‑master secret encrypted with server public key), ChangeCipherSpec and Finished messages. Illustrated with Wireshark captures.

Man‑in‑the‑middle attack description: attacker acts as proxy, presents forged certificate, if user accepts the warning the whole HTTPS session can be decrypted.

MySQL Indexes

Indexes accelerate data retrieval; types include B+Tree, Hash, Full‑text; physical storage: clustered (primary) vs secondary; field characteristics: primary, unique, normal, prefix; single‑column vs composite.

InnoDB uses B+Tree as default. Optimization tips: use EXPLAIN, create appropriate (composite) indexes, avoid leading wildcard, functions on indexed columns, drive large tables with small tables, pagination tricks, table partitioning.

Spring Circular Dependency

Three cases: constructor injection, setter injection in prototype scope, setter injection in singleton scope. Only the singleton setter case is resolved by Spring using a three‑level cache (singletonObjects, earlySingletonObjects, singletonFactories).

Cache structures are simple Map objects stored in DefaultSingletonBeanRegistry.

Operating System Links

Hard link: multiple directory entries point to same inode, cannot cross file systems, file is removed only when all links are deleted.

Soft link: separate inode containing path to target, can cross file systems, remains after target deletion (dangling).

Process vs Thread

Process is resource‑allocation unit; thread is execution unit. Processes have higher context‑switch cost and isolation; threads share memory, are lighter weight.

Fork Mechanism

Fork copies parent page table (virtual memory) using copy‑on‑write; physical pages are shared until a write triggers duplication.

Heap vs Stack

Heap: dynamic allocation, manual management, larger, slower. Stack: static allocation, automatic management, limited size, fast.

I/O Multiplexing

Single process can monitor many sockets using select/poll/epoll, handling each event within ~1 ms to achieve high concurrency.

TCP vs UDP

TCP is connection‑oriented, reliable, with flow control and congestion control; UDP is connectionless, faster, with no reliability guarantees.

HTTP/2 Features

Header compression (HPACK), binary framing, multiplexed streams, server push.

Additional Topics

Exception hierarchy in Java (Throwable → Error / Exception, checked vs unchecked)

Common design patterns: Singleton, Factory Method, Abstract Factory

Thread‑synchronization alternatives: ReentrantLock, volatile, atomic classes

HTTP range requests for breakpoint resume

Hand‑written coding tasks: rotate linked list, longest substring without repeat, singleton implementation, dynamic array‑based queue