OPDATA-6745: More test coverage for Requester

[dskloetc]

OPDATA-6745

Description

There is a memory leak in Requester.

If requests come in frequently enough that the requester queue is never empty, even if it stays at a small size, the recursive processNext function will chain Promise objects forever. Only when the request queue is empty, does the promise stack unravel to release the memory.

This happens in por-address-list, the-network-firm, tiingo and to a lesser extent finnhub-secondary because they have multiple endpoint making multiple requests. This means that even though one endpoint waits for all its requests to finish before sending a new batch, there is always a batch from another endpoint still in the queue to keep the promise chain from unraveling.

I plan to significantly refactor the Requester to simplify how it queues requests to fix this issue. But before I do, I want to make sure there is extensive test coverage to make sure I don't break anything with the refactoring.

This PR adds that test coverage.

Another side effect of how the queue is processed is that some logs that are made by the requester seem to come from a different endpoint than the endpoint that made the request. The logger uses AsyncLocalStorage to enrich log statements with additional context about the origin of the requests. But because Requester.request will keep processing queued requests until the queue is empty, requests end up being attributed to the wrong endpoint.

While the memory leak is difficult to observe in a unit test, we can observe this AsyncLocalStorage usage so this will be used as a proxy unit test for whether the memory leak is fixed before confirm it on staging.

There are also a few other small bugs observed by unit tests:

1. When retrying a failed request, it waits some time before sending the request back into the queue. This wait time is incorrectly attributed to the request duration metric.
2. Before making a request, it waits for the rate limiter to give the green light. But the request is already removed from the queue before it starts waiting for the rate limited.

None of these bugs are fixed in this PR but they are demonstrated in the tests. They should all be fixed by the follow-up refactoring.

Changes

1. Fix a small bug in FixedIntervalRateLimiter where it wouldn't rate limit as long as Date.now() equal zero. This was observed in the unit test because it uses fake timer that starts at 0.
2. Fix a small issue in runAllUntil where it could advance time even if it wasn't necessary in order to satisfy isComplete().
3. Add unit tests for Requester.

[github-actions]

NPM Publishing labels 🏷️

🔵 This PR has the none label set and it will not cause a version bump.

[github-actions]

❌ failed to create version bump PR
Failure message: No message provided
Failed job: createVersionBumpPR