package server

import (
	"fmt"
	"io"

	"github.com/mostlygeek/llama-swap/internal/router/scheduler"
	"github.com/mostlygeek/llama-swap/internal/shared"
)

// writeSchedulerMetrics emits the request queue in Prometheus text format.
//
// The reorder counters are the ones that matter for tuning: they say how often
// aging or swap affinity actually changed which request went next. Without
// them, choosing agingDivisor and swapAffinityBonus is guesswork.
func writeSchedulerMetrics(w io.Writer, stats scheduler.QueueStats) {
	fmt.Fprintf(w, "# HELP llamaswap_scheduler_queue_depth Requests waiting in the scheduler queue, by priority band\n")
	fmt.Fprintf(w, "# TYPE llamaswap_scheduler_queue_depth gauge\n")
	for _, band := range shared.PriorityBands {
		fmt.Fprintf(w, "llamaswap_scheduler_queue_depth{band=%q} %d\n", band, stats.Bands[band].Depth)
	}

	fmt.Fprintf(w, "# HELP llamaswap_scheduler_queue_oldest_wait_seconds How long the longest-waiting queued request has waited, by priority band\n")
	fmt.Fprintf(w, "# TYPE llamaswap_scheduler_queue_oldest_wait_seconds gauge\n")
	for _, band := range shared.PriorityBands {
		fmt.Fprintf(w, "llamaswap_scheduler_queue_oldest_wait_seconds{band=%q} %g\n", band, stats.Bands[band].OldestWait.Seconds())
	}

	fmt.Fprintf(w, "# HELP llamaswap_scheduler_dispatched_total Requests dispatched by the scheduler, by priority band\n")
	fmt.Fprintf(w, "# TYPE llamaswap_scheduler_dispatched_total counter\n")
	for _, band := range shared.PriorityBands {
		fmt.Fprintf(w, "llamaswap_scheduler_dispatched_total{band=%q} %d\n", band, stats.Bands[band].Dispatched)
	}

	fmt.Fprintf(w, "# HELP llamaswap_scheduler_reorders_total Dispatches where a scoring term changed which request was picked\n")
	fmt.Fprintf(w, "# TYPE llamaswap_scheduler_reorders_total counter\n")
	fmt.Fprintf(w, "llamaswap_scheduler_reorders_total{cause=\"aging\"} %d\n", stats.AgingReorders)
	fmt.Fprintf(w, "llamaswap_scheduler_reorders_total{cause=\"swap_affinity\"} %d\n", stats.AffinityReorders)
}