Design out-of-order windowed stream processor
Company: Box
Role: Software Engineer
Category: Coding & Algorithms
Difficulty: medium
Interview Round: Onsite
##### Question
Design an event processor for an unbounded, infinite stream of events. Each event has the fields `id`, `timestamp`, `payload` (a string), and `checksum`. Events may arrive out of timestamp order, and they should drive a rolling aggregate over a 1-minute (60-second) sliding window.
Your processor must:
1. **Validate each event by its checksum** and discard any event whose checksum does not match its payload.
2. **Drop late and out-of-window events** — any valid event whose timestamp falls more than 60 seconds behind the current window (the latest observed timestamp) is discarded.
3. **Handle duplicates** — if the same event `id` is seen more than once, it must only be counted once.
4. **Maintain and report the average payload length over the latest 60 seconds**, updating continuously as new events arrive and as old events fall out of the window.
5. Keep the **overall time complexity no worse than O(n log n)** across `n` events.
Describe the data structures you would use, how you keep the window current as the clock advances, and how you handle late arrivals and duplicates.
Quick Answer: Design out-of-order windowed stream processor evaluates algorithm design, data structures, correctness, complexity, edge cases, and implementation details in a realistic interview setting. A strong answer states assumptions, handles edge cases, explains trade-offs, and shows how to validate the result clearly.
Solution
# Solution Alignment
The prompt asks for an implementation-level answer. The safest way to present it is to define the state, maintain clear invariants, then walk through complexity and tests.
## Problem Restatement
##### Question Design an event processor for an unbounded, infinite stream of events. Each event has the fields `id`, `timestamp`, `payload` (a string), and `checksum`. Events may arrive out of timestamp order, and they should drive a rolling aggregate over a 1-minute (60-second) sliding window. Your processor must: 1. **Validate each event by its checksum** and discard any event whose checksum does not match its payload. 2. **Drop late and out-of-window events** — any valid event whose timestamp falls more than 60 seconds behind the current window (the latest observed timestamp) is discarded. 3. **Handle duplicates** — if the same event `id` is seen more than once, it must only be counted o...
## Recommended Approach
Use the string constraints to choose between two pointers, a stack, frequency counts, prefix/suffix state, or dynamic programming. Maintain the invariant that processed characters have already been normalized, counted, or matched according to the operation.
## Correctness
The implementation should maintain an invariant after each loop or operation that directly matches the problem statement. At termination, that invariant implies the returned value has considered every valid candidate exactly once, or has preserved the required data-structure state after every API call.
## Complexity
Most direct string scans are O(n) time. Space ranges from O(1) for two pointers to O(n) for stacks, maps, or DP tables.
## Edge Cases and Tests
Empty string, length 1, repeated characters, invalid characters, case sensitivity, Unicode vs ASCII, and very long input.
Explanation
Streaming validate -> dedupe -> admit -> evict pipeline. Keep a timestamp-ordered window (min-heap or balanced BST) plus a running sum of payload lengths and a count, so the average is O(1) per report. Out-of-order arrivals are handled by the ordered structure; events older than (max_ts - 60s) are dropped; duplicate ids are rejected via a hash set; corrupt events fail checksum. Each event is inserted/removed at most once at O(log n), giving O(n log n) overall.