Backend Developer interview prep for Accenture.

1) Optimistic Locking: Assumes data conflicts are rare. The database does not place locks on rows when reading. Instead, each row includes a `@Version` timestamp or integer column. When updating, the query checks `WHERE id = ? AND version = ?`. If another transaction updated the row in the interim, the version does not match, the update affects 0 rows, and an `OptimisticLockException` is thrown. Ideal for high-concurrency read-heavy applications with low write collision; 2) Pessimistic Locking: Assumes data conflicts will happen. When reading, it acquires an explicit row lock (`SELECT ... FOR UPDATE`), preventing any other transaction from reading (in some DBs) or writing to that row until the locking transaction commits. Ideal for financial transactions, stock inventory deductions, or workflows where collisions are frequent and cannot be tolerated.

1) Algorithm Selection: Use the Token Bucket or Sliding Window Counter algorithm. The Token Bucket allows configurable burst traffic while enforcing an average rate; Sliding Window Counter provides smooth rate enforcement without boundary spikes; 2) Distributed State: In a distributed system with multiple gateway or service instances, store client request counts in a centralized, in-memory datastore (Redis) using atomic Lua scripts to prevent race conditions during increments; 3) Identification & Headers: Key rate limits by client API key, OAuth client ID, or IP address. Return standard HTTP headers: `X-RateLimit-Limit`, `X-RateLimit-Remaining`, and `X-RateLimit-Reset`; 4) Rejection: When limits are exceeded, immediately return HTTP 429 Too Many Requests with a `Retry-After` header.