Error Budget and Burn Rate

Problem Statement

An SLO of 99.9% availability is really a budget: over the window you are allowed to fail 0.1% of requests. Spending that budget slowly is normal operation; spending it fast is an incident. Compute how much budget a service has left and how fast it is burning.

Background

For an availability SLO over one window:

budget_rate         = 1 - slo_target                       # fraction you may fail
allowed_failures    = budget_rate * total_requests
error_rate          = failed_requests / total_requests
burn_rate           = error_rate / budget_rate
budget_consumed_pct = 100 * failed_requests / allowed_failures
budget_remaining_pct= 100 - budget_consumed_pct

Burn rate is the number that pages someone. A burn rate of 1.0 means you will spend exactly the whole budget over exactly the window — sustainable. A burn rate of 10 means the month's budget is gone in three days, which is why fast-burn alerts fire on burn rate rather than on raw error rate.

Projecting forward at the current burn rate:

hours_to_exhaustion = (budget_remaining_pct / 100) * window_hours / burn_rate

Two clamps: once the budget is exhausted (budget_remaining_pct <= 0), report 0.0 for both the remaining percentage and the hours; and with a burn rate of 0 the budget never runs out, so report the full window_hours.

Your Task

Implement:

def error_budget(slo_target, total_requests, failed_requests, window_hours):

Return a dict with keys "allowed_failures", "burn_rate", "budget_consumed_pct", "budget_remaining_pct", "hours_to_exhaustion", in that order, each rounded to 2 decimal places.

Input Format

• slo_target: float in (0, 1), e.g. 0.999.
• total_requests, failed_requests: ints.
• window_hours: number of hours in the SLO window (720 for 30 days).

Output Format

• A dict of five values, each rounded to 2 decimals.

Sample

print(error_budget(0.999, 2000000, 1200, 720))

Output:

{'allowed_failures': 2000.0, 'burn_rate': 0.6, 'budget_consumed_pct': 60.0, 'budget_remaining_pct': 40.0, 'hours_to_exhaustion': 480.0}

A 99.9% SLO over 2M requests buys 2000 failures. 1200 are spent, so 40% of the budget is left and the service is burning at 0.6x — under 1.0, so it will finish the window inside budget.