Performance traps: interview questions

line 4 · Performance trap

Each request waits for the previous one to finish, so total time is the sum of all latencies (50 × ~200ms). The requests are independent and can run concurrently.

export async function loadAvatars(userIds: string[]): Promise<string[]> {  return Promise.all(    userIds.map(async (id) => {      const res = await fetch(`/api/users/${id}`);      const user: { avatarUrl: string } = await res.json();      return user.avatarUrl;    }),  );}

Interview follow-up: Strong candidates add that very large lists need a concurrency limit, or a batch endpoint.

line 4 · Performance trap

Strings are immutable, so each += copies everything built so far into a new string. Over n rows that is O(n²) copying and a huge amount of garbage.

    StringBuilder csv = new StringBuilder();    for (String[] row : rows) {        csv.append(String.join(",", row)).append('\n');    }    return csv.toString();

Interview follow-up: Mention that the compiler's StringBuilder optimization applies within a single expression, not across loop iterations.

line 3 · Performance trap

A B-tree index is ordered by the start of the value, so a pattern that starts with % can't use it. The database scans all 30M rows.

-- Store the domain separately (or use a generated column) and index it:ALTER TABLE users ADD COLUMN email_domain TEXT    GENERATED ALWAYS AS (split_part(email, '@', 2)) STORED;CREATE INDEX ON users (email_domain); SELECT COUNT(*) FROM users WHERE email_domain = 'gmail.com';

Interview follow-up: Alternatives worth naming: an index on reverse(email), or a trigram (pg_trgm) index for general substring search.