Exercise 2: Two Servers Going Down at Different Points — Possible Solution
====================================================================

THE CHANGE
------------------------------
    if i == 10:
        server_b.healthy = False
    if i == 20:
        server_c.healthy = False

Server B goes down at request 10 (matching this chapter's own original
timing), and server C goes down at request 20 - leaving only server A
healthy for the final third of the run. This chapter's own
HealthCheckingBalancer and send_request() logic are used completely
unchanged.

VERIFYING ZERO FAILURES STILL HOLDS
------------------------------
    failures out of 30 requests: 0

Even with two separate, independent failures during the run, the
health-checking balancer produced zero failed requests - exactly
matching this chapter's own single-failure result. Each time a server
went down, the balancer's own healthy_servers filter simply excluded
it from the very next routing decision onward.

DISTRIBUTION ACROSS THE THREE SERVERS
------------------------------
    distribution: {'A': 19, 'B': 3, 'C': 8}

B received 3 requests before going down at request 10 (its share of
the first 10 requests, split with A and C). C received 8 requests
total - its share of requests 0-9 alongside A and B, plus its share of
requests 10-19 alongside only A, before going down itself at request
20. A received the remaining 19 requests - its share of both earlier
windows, plus the ENTIRE final 10 requests (20-29) once it was the only
healthy server left standing.

WHY THIS CONFIRMS HEALTH CHECKING SCALES TO MULTIPLE FAILURES
------------------------------
This chapter's own single-failure scenario already showed health
checking prevents failed requests from one dead server. This exercise
confirms the same mechanism handles a SECOND, independent failure
happening later in the same run, with zero additional logic needed -
the balancer re-evaluates which servers are healthy on every single
routing decision, so it doesn't matter whether one server goes down or
several, or when.

WHY THIS WORKS AS AN ANSWER
------------------------------
The scenario extends this chapter's own single-failure test with a
second, independently-timed failure using the exact same balancer
code, verifies the zero-failure result still holds, and traces the
resulting server distribution back to exactly which servers were
healthy during each segment of the run rather than only reporting the
final totals.