Google Software Engineer Interview Questions

A stack is a Last-In-First-Out (LIFO) structure, optimized for push/pop operations at a single end. Dynamic arrays (like std::vector) implement stacks with amortized O(1) time and high cache locality due to contiguous memory. A queue is a First-In-First-Out (FIFO) structure. Dynamic array queues require circular buffers to prevent O(n) element shifting overhead during de-queuing. Use stacks for tracking execution state (DFS, undo/redo mechanisms, compiler parsing) and queues for asynchronous decoupling (BFS graph traversal, thread-pool task scheduling, rate-limiting buffers).

Binary search runs in O(log n) time and O(1) auxiliary space complexity. The prerequisite is that the input data must exhibit strict monotonic order (be sorted) across a contiguous random-access sequence to correctly halve the search space at each iteration step. At massive data scale, the midpoint index evaluation must be implemented mathematically as low + (high - low) / 2 instead of (low + high) / 2 to avoid integer overflow bugs inside the runtime execution stack.

The four pillars are: 1. Encapsulation: Binding data states and mutating methods into a protected class unit, exposing changes exclusively via strict access control modifiers (private/protected). 2. Abstraction: Concealing low-level architectural complexity behind clean, minimal public interfaces (e.g., exposing simple Get/Set APIs while hiding distributed sharding or disk I/O logic). 3. Inheritance: Sharing behavioral code structurally from generic parent classes to specialized child subclasses. 4. Polymorphism: Processing different data objects using a uniform interface via dynamic runtime method dispatch or compile-time static overloading.

A deadlock is an absolute execution stall occurring when a set of concurrent threads cannot proceed because each holds a resource while waiting for another thread to release a different resource. It mathematically requires all four Coffman conditions simultaneously: mutual exclusion, hold-and-wait, no resource preemption, and a circular wait dependency chain. Prevention strategies require breaking at least one condition, typically by enforcing a strict global resource allocation hierarchy or utilizing non-blocking lock-free synchronization primitives like Compare-And-Swap (CAS).

TCP is a connection-oriented protocol that guarantees reliable, ordered byte stream delivery using a 3-way handshake, sequence numbers, checksum acknowledgments, sliding-window flow control, and window-based congestion avoidance. UDP is a connectionless, lightweight datagram protocol that prioritizes speed and low-overhead delivery with no retransmissions or packet ordering guarantees. Google leverages modern protocols like QUIC running over UDP to implement low-latency connection multiplexing and customized packet loss recovery, entirely avoiding traditional TCP head-of-line blocking.