System Design — Queue With Visibility Timeout

How SQS-style queues survive crashing workers. A consumer that receives a message does not
remove it: the message is only hidden for a while. If the consumer deletes it in time the
work is done; if it crashes, the message reappears and another worker gets it.
That is *at-least-once* delivery.

solve(args, ops) builds new VisibilityQueue(timeout) — args = [timeout] — and replays
ops:

• send(body, now) — enqueue a message and return its id (1, 2, 3, … in send order).
• receive(now) — return [id, body] of the visible message with the smallest id, and

hide it until now + timeout; return null if no message is visible.

• delete(id) — remove the message permanently (even while hidden). Returns true if it

existed, else false.

A message is visible when now >= hiddenUntil (new messages are visible at once). A
redelivered message keeps its id, so it is served before newer messages.

Example (timeout = 10)

send("a",0) → 1   send("b",1) → 2   receive(2) → [1,"a"]   receive(4) → [2,"b"]   receive(5) → null
receive(12) → [1,"a"]      // the first lease expired