Sliding Window Error Counter

Build a deterministic counter for recent errors in a rolling time window.

Requirements

• Define ErrorCounter(window_seconds, now).
• record(message) stores one error occurrence at the current time.
• count() returns the number of non-expired errors.
• top(k) returns (message, count) pairs for the top k messages, ordered by count descending then message ascending.
• If k <= 0, top(k) returns an empty list.
• Every read path must treat expired events as missing. record may also purge expired events, preventing unbounded storage growth.
• An event expires when now() - event_time >= window_seconds.
• Ties in top(k) sort alphabetically by message (already implied by count-desc then message-asc).

Example

clock = {"t": 0}
counter = ErrorCounter(10, now=lambda: clock["t"])
counter.record("timeout")
counter.record("timeout")
counter.record("rate_limit")
assert counter.count() == 3
assert counter.top(2) == [("timeout", 2), ("rate_limit", 1)]
clock["t"] = 10
assert counter.count() == 0

Constraints

• Use deterministic injected time.
• Assume injected timestamps move forward monotonically; an ordered deque can't safely expire events after the clock moves backward.
• window_seconds is finite and positive. now() returns finite numeric timestamps in seconds and may repeat a timestamp. Use one clock snapshot per operation. Messages are strings counted exactly, including empty strings; k is an integer.
• Use standard-library Python only.
• Don't scan expired data forever if many events are recorded.
• Retain events in the active window rather than the entire history. Memory still grows with the number of events inside that window; a high event rate requires a separate limit or aggregation policy.