Merge pull request #818 from gastownhall/worktree-fix-772-wake-budget

config: make wake-per-tick budget configurable via [daemon] (#772)

Author: julianknutsen

cmd/gc/session_lifecycle_parallel.go

@@ -788,6 +788,7 @@ func executePlannedStartsTraced(
 	if len(candidates) == 0 {
 		return 0
 	}
+	maxWakes := cfg.Daemon.MaxWakesPerTickOrDefault()
 	waveByCandidate, ok := candidateWaveOrder(candidates, cfg, desiredState, sp, cityName, store)
 	if !ok {
 		fmt.Fprintln(stderr, "session reconciler: dependency graph fallback to serial start order") //nolint:errcheck
@@ -810,7 +811,7 @@ func executePlannedStartsTraced(
 		if len(waveCandidates) == 0 {
 			continue
 		}
-		if wakeCount >= defaultMaxWakesPerTick {
+		if wakeCount >= maxWakes {
 			for _, candidate := range waveCandidates {
 				logLifecycleOutcome(stderr, "start", wave, candidate.name(), candidate.logicalTemplate(cfg), "deferred_by_wake_budget", time.Time{}, time.Time{}, nil)
 			}
@@ -825,13 +826,13 @@ func executePlannedStartsTraced(
 			ready = append(ready, candidate)
 		}
 		for offset := 0; offset < len(ready); {
-			if wakeCount >= defaultMaxWakesPerTick {
+			if wakeCount >= maxWakes {
 				for _, candidate := range ready[offset:] {
 					logLifecycleOutcome(stderr, "start", wave, candidate.name(), candidate.logicalTemplate(cfg), "deferred_by_wake_budget", time.Time{}, time.Time{}, nil)
 				}
 				break
 			}
-			batchSize := min(defaultMaxParallelStartsPerWave, defaultMaxWakesPerTick-wakeCount)
+			batchSize := min(defaultMaxParallelStartsPerWave, maxWakes-wakeCount)
 			end := min(offset+batchSize, len(ready))
 			var prepared []preparedStart
 			for _, candidate := range ready[offset:end] {

cmd/gc/session_lifecycle_parallel_test.go

@@ -741,6 +741,55 @@ func TestReconcileSessionBeads_BlockedCandidatesDoNotConsumeWakeBudget(t *testin
 	}
 }
 
+// TestReconcileSessionBeads_DaemonMaxWakesPerTickOverride covers the fix for
+// issue #772: [daemon].max_wakes_per_tick = N raises (or lowers) the wake
+// budget away from the 5-per-tick default. Cities with slow cold-starts
+// need this to drain the candidate queue.
+func TestReconcileSessionBeads_DaemonMaxWakesPerTickOverride(t *testing.T) {
+	env := newReconcilerTestEnv()
+	override := 8
+	env.cfg = &config.City{
+		Daemon: config.DaemonConfig{MaxWakesPerTick: &override},
+		Agents: []config.Agent{
+			{Name: "ready-1"},
+			{Name: "ready-2"},
+			{Name: "ready-3"},
+			{Name: "ready-4"},
+			{Name: "ready-5"},
+			{Name: "ready-6"},
+			{Name: "ready-7"},
+			{Name: "ready-8"},
+			{Name: "ready-9"},
+		},
+	}
+	names := []string{"ready-1", "ready-2", "ready-3", "ready-4", "ready-5", "ready-6", "ready-7", "ready-8", "ready-9"}
+	for _, name := range names {
+		env.addDesired(name, name, false)
+	}
+
+	var seeded []beads.Bead
+	for _, name := range names {
+		b := env.createSessionBead(name, name)
+		env.markSessionCreating(&b)
+		seeded = append(seeded, b)
+	}
+
+	woken := env.reconcile(seeded)
+
+	if woken != override {
+		t.Fatalf("woken = %d, want %d (overridden wake budget)", woken, override)
+	}
+	// First 8 run, 9th is deferred_by_wake_budget.
+	for _, name := range names[:override] {
+		if !env.sp.IsRunning(name) {
+			t.Fatalf("%s should have started under override=%d", name, override)
+		}
+	}
+	if env.sp.IsRunning(names[override]) {
+		t.Fatalf("%s should have been deferred once budget hit", names[override])
+	}
+}
+
 func TestPrepareStartCandidate_NoneModeInitialMessageStaysInNudge(t *testing.T) {
 	store := beads.NewMemStore()
 	bead, err := store.Create(beads.Bead{

cmd/gc/session_types.go

@@ -3,6 +3,8 @@ package main
 import (
 	"sync"
 	"time"
+
+	"github.com/gastownhall/gascity/internal/config"
 )
 
 // WakeReason describes why a session should be awake.
@@ -185,9 +187,11 @@ const (
 	// before it's considered stable (not a rapid exit / crash).
 	stabilityThreshold = 30 * time.Second
 
-	// maxWakesPerTick limits how many sessions can be woken per reconciler
-	// tick to prevent thundering herd after controller restart.
-	defaultMaxWakesPerTick = 5
+	// defaultMaxWakesPerTick mirrors config.DefaultMaxWakesPerTick (kept
+	// here so tests and non-config call sites don't need to take a
+	// dependency on internal/config just for the default). Configurable
+	// per city via [daemon].max_wakes_per_tick; see issue #772.
+	defaultMaxWakesPerTick = config.DefaultMaxWakesPerTick
 
 	// defaultTickBudget is the wall-clock budget per reconciler tick.
 	// Remaining work is deferred to the next tick.