Design an async, batched key-value fetcher with a 100 ms window

Context

You have a single-threaded, event-loop environment (e.g., JS/TS). A backend endpoint serves numeric values for keys via HTTP GET:

• Endpoint: GET /read?keys=k1,k2,...
• Response: a JSON object mapping each requested key to its numeric value, for example: { "foo": 12, "bar": 100 } .

Available primitives:

• httpGetAsync(urlPath, callback) -> void where callback(err, jsonString) .
• parseKvs(jsonString) -> Map<string, number> that parses the JSON string into a map of key → value.

Task

Implement a single-threaded (event-loop) interface:

• getKeyAsync(key, callback)

that supports concurrent I/O and batches multiple incoming requests.

Requirements

1. Batch all requests that arrive within a 100 ms window into a single HTTP call to /read?keys=... .
2. Deduplicate keys within a batch; if multiple requests ask for the same key, invoke each request’s callback exactly once.
3. Enforce consistency: a batch response may only satisfy callbacks for requests that arrived before that batch was dispatched. Requests arriving afterward must be served by a subsequent batch.
4. Allow multiple in-flight batches without mixing responses.
5. Define the data structures (queues, maps, sets), control flow (timers, scheduling), and how responses are correlated to waiting callbacks.
6. Specify behavior for missing keys, partial failures, timeouts/retries, and cancellation/backpressure.
7. Discuss complexity (time/space), single-threaded ordering pitfalls, and a brief test plan.