fix(site-runtime): reuse exited detached worker slots in script scheduler (review 2026-10-21 01/SR-M1)

src/ZB.MOM.WW.ScadaBridge.SiteRuntime/Scripts/ScriptExecutionScheduler.cs

@@ -307,17 +307,43 @@ public sealed class ScriptExecutionScheduler : TaskScheduler, IDisposable
         }
     }
 
-    /// <summary>Appends <paramref name="count"/> new worker slots + threads. Caller holds <see cref="_growLock"/>.</summary>
+    /// <summary>
+    /// Adds <paramref name="count"/> workers. Slots vacated by an exited detached worker are
+    /// reused first (review 2026-10-21 01/SR-M1: otherwise a script that wedges on every run
+    /// grows <see cref="_slots"/> and starts a fresh slot per detach for the process lifetime);
+    /// only the remainder appends new slots. Caller holds <see cref="_growLock"/>.
+    /// </summary>
     private void GrowTo(int count)
     {
         var existing = _slots;
-        var grown = new WorkerSlot[existing.Length + count];
+        var reuse = new List<int>();
-        Array.Copy(existing, grown, existing.Length);
+        for (var i = 0; i < existing.Length && reuse.Count < count; i++)
+            if (Volatile.Read(ref existing[i].Free) != 0) reuse.Add(i);
 
-        for (var i = 0; i < count; i++)
+        var appended = count - reuse.Count;
+        var grown = existing;
+        if (appended > 0)
         {
-            var index = existing.Length + i;
+            grown = new WorkerSlot[existing.Length + appended];
-            grown[index] = new WorkerSlot();
+            Array.Copy(existing, grown, existing.Length);
+            for (var i = 0; i < appended; i++)
+                grown[existing.Length + i] = new WorkerSlot();
+        }
+
+        var indices = new List<int>(count);
+        indices.AddRange(reuse);
+        for (var i = 0; i < appended; i++)
+            indices.Add(existing.Length + i);
+
+        // Reset reused slots BEFORE the thread starts; the exited worker never touches its slot
+        // again after marking it Free.
+        foreach (var index in reuse)
+        {
+            var slot = grown[index];
+            Volatile.Write(ref slot.Detached, 0);
+            Volatile.Write(ref slot.RunStamp, 0);
+            Volatile.Write(ref slot.BusySinceTicks, 0);
+            Volatile.Write(ref slot.Free, 0);
         }
 
         // Publish the array BEFORE starting the threads: a worker's very first action is to
@@ -325,19 +351,22 @@ public sealed class ScriptExecutionScheduler : TaskScheduler, IDisposable
         _slots = grown;
         _configuredCount += count;
 
-        for (var i = 0; i < count; i++)
+        foreach (var index in indices)
         {
-            var index = existing.Length + i;
+            var i = index;
-            var thread = new Thread(() => WorkerLoop(index))
+            var thread = new Thread(() => WorkerLoop(i))
             {
                 IsBackground = true,
-                Name = ThreadNamePrefix + index
+                Name = ThreadNamePrefix + i
             };
-            grown[index].Thread = thread;
+            grown[i].Thread = thread;
             thread.Start();
         }
     }
 
+    /// <summary>Total worker slots (live + detached-but-unexited + free). Diagnostic/test seam.</summary>
+    internal int SlotCount => _slots.Length;
+
     [ThreadStatic]
     private static ScriptExecutionScheduler? CurrentScheduler;
 
@@ -368,6 +397,14 @@ public sealed class ScriptExecutionScheduler : TaskScheduler, IDisposable
                 if (Volatile.Read(ref slot.Detached) != 0)
                 {
                     Interlocked.Decrement(ref _detachedLive);
+                    // Hand the slot back for reuse by the next GrowTo. Under the lock so a
+                    // concurrent GrowTo sees either "not free" or a fully vacated slot; the
+                    // thread touches no slot state after this.
+                    lock (_growLock)
+                    {
+                        slot.Thread = null;
+                        Volatile.Write(ref slot.Free, 1);
+                    }
                     return;
                 }
             }
@@ -435,6 +472,9 @@ public sealed class ScriptExecutionScheduler : TaskScheduler, IDisposable
         /// <summary>Non-zero once the worker has been detached; it exits after its current task.</summary>
         public int Detached;
 
+        /// <summary>Non-zero once a detached worker has exited and the slot may be reused by <see cref="GrowTo"/>.</summary>
+        public int Free;
+
         /// <summary>The worker thread, for <see cref="Dispose"/> to join.</summary>
         public Thread? Thread;
     }

tests/ZB.MOM.WW.ScadaBridge.SiteRuntime.Tests/Scripts/ScriptExecutionSchedulerTests.cs

@@ -118,4 +118,44 @@ public class ScriptExecutionSchedulerTests
             await Task.Delay(50);
         Assert.True(condition(), "condition not met within timeout");
     }
+
+    /// <summary>
+    /// Repeatedly wedging + detaching a worker must reuse the vacated slot rather than grow the
+    /// slot table without bound (review 2026-10-21 01/SR-M1).
+    /// </summary>
+    [Fact]
+    public async Task DetachedWorkerSlot_IsReused_SlotTableStaysBounded()
+    {
+        using var scheduler = new ScriptExecutionScheduler(1);
+
+        for (var round = 0; round < 6; round++)
+        {
+            var gate = new ManualResetEventSlim(false);
+            var started = new TaskCompletionSource<(int Slot, long Stamp)>();
+            var wedged = Task.Factory.StartNew(() =>
+            {
+                var slot = ScriptExecutionScheduler.CurrentWorkerSlot!.Value;
+                started.SetResult((slot, scheduler.CurrentRunStamp(slot)));
+                gate.Wait();
+            }, CancellationToken.None, TaskCreationOptions.None, scheduler);
+
+            var (s, stamp) = await started.Task;
+            Assert.Equal(WorkerDetachOutcome.Detached, scheduler.TryDetachWorker(s, stamp));
+            gate.Set();
+            await wedged;
+
+            // Wait for the detached worker to exit and hand its slot back.
+            var deadline = DateTime.UtcNow.AddSeconds(5);
+            while (scheduler.DetachedThreadCount != 0 && DateTime.UtcNow < deadline)
+                await Task.Delay(10);
+            Assert.Equal(0, scheduler.DetachedThreadCount);
+            await Task.Delay(50);
+        }
+
+        // 1 live + at most 1 free/in-flight detached slot, never 1 + 6.
+        Assert.True(scheduler.SlotCount <= 2, $"slot table grew to {scheduler.SlotCount}");
+        // The pool still works afterwards.
+        var ran = await Task.Factory.StartNew(() => 42, CancellationToken.None, TaskCreationOptions.None, scheduler);
+        Assert.Equal(42, ran);
+    }
 }