AMM: scatter and replicate to avoid paused workers

distributed/client.py

@@ -2048,9 +2048,10 @@ async def _scatter(
                         await asyncio.sleep(0.1)
                     if time() > start + timeout:
                         raise TimeoutError("No valid workers found")
-                    nthreads = await self.scheduler.ncores(workers=workers)
+                    # Exclude paused and closing_gracefully workers
+                    nthreads = await self.scheduler.ncores_running(workers=workers)
                 if not nthreads:
-                    raise ValueError("No valid workers")
+                    raise ValueError("No valid workers found")
 
                 _, who_has, nbytes = await scatter_to_workers(
                     nthreads, data2, report=False, rpc=self.rpc

distributed/scheduler.py

@@ -3826,6 +3826,7 @@ def __init__(
             "broadcast": self.broadcast,
             "proxy": self.proxy,
             "ncores": self.get_ncores,
+            "ncores_running": self.get_ncores_running,
             "has_what": self.get_has_what,
             "who_has": self.get_who_has,
             "processing": self.get_processing,
@@ -5710,18 +5711,24 @@ async def scatter(
         Scheduler.broadcast:
         """
         parent: SchedulerState = cast(SchedulerState, self)
+        ws: WorkerState
+
         start = time()
-        while not parent._workers_dv:
-            await asyncio.sleep(0.2)
+        while True:
+            if workers is None:
+                wss = parent._running
+            else:
+                workers = [self.coerce_address(w) for w in workers]
+                wss = {parent._workers_dv[w] for w in workers}
+                wss = {ws for ws in wss if ws._status == Status.running}
+
+            if wss:
+                break
             if time() > start + timeout:
-                raise TimeoutError("No workers found")
+                raise TimeoutError("No valid workers found")
+            await asyncio.sleep(0.1)
 
-        if workers is None:
-            ws: WorkerState
-            nthreads = {w: ws._nthreads for w, ws in parent._workers_dv.items()}
-        else:
-            workers = [self.coerce_address(w) for w in workers]
-            nthreads = {w: parent._workers_dv[w].nthreads for w in workers}
+        nthreads = {ws._address: ws.nthreads for ws in wss}
 
         assert isinstance(data, dict)
 
@@ -5732,10 +5739,7 @@ async def scatter(
         self.update_data(who_has=who_has, nbytes=nbytes, client=client)
 
         if broadcast:
-            if broadcast == True:  # noqa: E712
-                n = len(nthreads)
-            else:
-                n = broadcast
+            n = len(nthreads) if broadcast is True else broadcast
             await self.replicate(keys=keys, workers=workers, n=n)
 
         self.log_event(
@@ -6451,7 +6455,12 @@ async def replicate(
 
         assert branching_factor > 0
         async with self._lock if lock else empty_context:
-            workers = {parent._workers_dv[w] for w in self.workers_list(workers)}
+            if workers is not None:
+                workers = {parent._workers_dv[w] for w in self.workers_list(workers)}
+                workers = {ws for ws in workers if ws._status == Status.running}
+            else:
+                workers = parent._running
+
             if n is None:
                 n = len(workers)
             else:
@@ -6989,6 +6998,15 @@ def get_ncores(self, comm=None, workers=None):
         else:
             return {w: ws._nthreads for w, ws in parent._workers_dv.items()}
 
+    def get_ncores_running(self, comm=None, workers=None):
+        parent: SchedulerState = cast(SchedulerState, self)
+        ncores = self.get_ncores(workers=workers)
+        return {
+            w: n
+            for w, n in ncores.items()
+            if parent._workers_dv[w].status == Status.running
+        }
+
     async def get_call_stack(self, comm=None, keys=None):
         parent: SchedulerState = cast(SchedulerState, self)
         ts: TaskState

distributed/tests/test_client.py

@@ -5734,6 +5734,33 @@ def bad_fn(x):
     assert y.status == "error"  # not cancelled
 
 
+@pytest.mark.parametrize("workers_arg", [False, True])
+@pytest.mark.parametrize("direct", [False, True])
+@pytest.mark.parametrize("broadcast", [False, True, 10])
+@gen_cluster(client=True, nthreads=[("", 1)] * 10)
+async def test_scatter_and_replicate_avoid_paused_workers(
+    c, s, *workers, workers_arg, direct, broadcast
+):
+    paused_workers = [w for i, w in enumerate(workers) if i not in (3, 7)]
+    for w in paused_workers:
+        w.memory_pause_fraction = 1e-15
+    while any(s.workers[w.address].status != Status.paused for w in paused_workers):
+        await asyncio.sleep(0.01)
+
+    f = await c.scatter(
+        {"x": 1},
+        workers=[w.address for w in workers[1:-1]] if workers_arg else None,
+        broadcast=broadcast,
+        direct=direct,
+    )
+    if not broadcast:
+        await c.replicate(f, n=10)
+
+    expect = [i in (3, 7) for i in range(10)]
+    actual = [("x" in w.data) for w in workers]
+    assert actual == expect
+
+
 @pytest.mark.xfail(reason="GH#5409 Dask-Default-Threads are frequently detected")
 def test_no_threads_lingering():
     if threading.active_count() < 40:

distributed/tests/test_scheduler.py

@@ -784,20 +784,27 @@ async def test_story(c, s, a, b):
     assert s.story(x.key) == s.story(s.tasks[x.key])
 
 
-@gen_cluster(nthreads=[], client=True)
-async def test_scatter_no_workers(c, s):
+@pytest.mark.parametrize("direct", [False, True])
+@gen_cluster(client=True, nthreads=[])
+async def test_scatter_no_workers(c, s, direct):
     with pytest.raises(TimeoutError):
         await s.scatter(data={"x": 1}, client="alice", timeout=0.1)
 
     start = time()
     with pytest.raises(TimeoutError):
-        await c.scatter(123, timeout=0.1)
+        await c.scatter(123, timeout=0.1, direct=direct)
     assert time() < start + 1.5
 
-    w = Worker(s.address, nthreads=3)
-    await asyncio.gather(c.scatter(data={"y": 2}, timeout=5), w)
-
-    assert w.data["y"] == 2
+    fut = c.scatter({"y": 2}, timeout=5, direct=direct)
+    await asyncio.sleep(0.1)
+    async with Worker(s.address) as w:
+        await fut
+        assert w.data["y"] == 2
+
+    # Test race condition between worker init and scatter
+    w = Worker(s.address)
+    await asyncio.gather(c.scatter({"z": 3}, timeout=5, direct=direct), w)
+    assert w.data["z"] == 3
     await w.close()