Schedules, failures and the WAIT rule

Schedules

A Schedule is piecewise-constant: points = [[duration, value], ...], each value holding for duration time units, optionally repeating forever (without repeat, the last value holds forever once the points run out). Declared in schedules, one of two kinds:

Kind	Values mean	Consumed by
arrival	Arrival rate per time unit	A CREATE block with arrival_mode = "Schedule" and matching schedule_name -- drawn as a non-homogeneous Poisson process via thinning (Lewis-Shedler): candidates come from an exponential at the schedule's maximum rate and are kept with probability rate(t) / max_rate.
capacity	Integer unit count	A resource's capacity_schedule field.

call_centre_day.json uses both kinds together (an arrival schedule for call volume, a capacity schedule for agent headcount).

Weekday schedules

When the model has a calendar, either kind can be written as a weekday grid in local wall-clock time -- {"mon": [["06:00", "14:00", 3]]} -- with dated exceptions for holidays and shutdowns, instead of points. It compiles to exactly the points above, once per run. See Dates and calendars.

Failures

A failure names one or more resources and an uptime/downtime distribution pair (clock time, not busy time). The controller fails a resource every uptime, taking whatever units the capacity schedule currently leaves available, holds for downtime, then restores them. machine_with_breakdowns.json is the worked example.

There is no per-unit arbitration between a capacity schedule and a failure competing for the same units at the same time (a known ceiling).

The WAIT rule

Capacity schedules and failures never change a resource's live SimPy capacity -- doing so would strand already-queued requests. Instead, the resource is always built at its maximum capacity, and a controller process holds invisible, top-priority "blocker" requests to take units out of service. A blocker outranks every entity request but still waits for a free unit, which is exactly Arena's WAIT rule: a job already in service always finishes before its resource goes off shift or fails.

Blockers are invisible to every statistic -- NQ/NR/MR and set selection see only the net values (units not currently blocked). Utilization (Usage % (snapshot)) divides busy time by the scheduled (available) capacity, so time a resource spends off-shift or down never dilutes it; use Scheduled Capacity Avg and Down % to see the schedule/failure effect directly.

Run length

A model with an unbounded CREATE (max_arrivals = 0), any failure, or a repeating capacity schedule never drains on its own -- it needs an explicit run length, or naga run / nagasim.api.run_model raises a clear ModelValidationError instead of hanging.