AI-assisted engineering/Reviewing AI code

Fix the secret cache in a partner API Lambda

Review the AI's codeEasy10 minAmazon style

Your task

Your Lambda calls a partner API with a key it reads from Secrets Manager. The first draft fetched the key on every request and fell back to a hardcoded test key when the secret was empty. The new version drops the fallback, caches the key in the warm container, and fetches it again once it is older than a time to live (TTL).

An assistant wrote secret_fetch_times(invocation_times, ttl_seconds), which models when the cache fetches. It is in the editor. It has a bug. Find it and fix it.

Input and output:

invocation_times: the time of each invocation of one warm container, in seconds, in the order they ran. Times are whole numbers, 0 or more. They usually increase, but not always.
ttl_seconds: how long a fetched key stays fresh, 0 or more.
Return the times at which the key is fetched, in invocation order.

Rules the tests check:

The first invocation always fetches, because nothing is cached yet.
A later invocation fetches when its time minus the time of the last fetch is at least ttl_seconds. Otherwise it reuses the cached key.
The TTL clock restarts at each fetch, not at each invocation.
When times go backwards, the difference is negative. That is less than any TTL, so there is no fetch, even with a TTL of 0.

Examples

secret_fetch_times(invocation_times: list[int], ttl_seconds: int) -> list[int]

Example 1

Inputsecret_fetch_times([0, 10, 300, 305, 700], 300)

Output[0, 300, 700]

Example 2

Inputsecret_fetch_times([1, 2, 3], 0)

Output[1, 2, 3]