Merge pull request #360 from golemfactory/az/async-task-generator

Asynchronous task iterator as argument to `Executor.submit()`

.github/workflows/integration.yml

@@ -36,11 +36,11 @@ jobs:
           poetry install -E integration-tests
 
       - name: Disconnect Docker containers from default network
-       continue-on-error: true
-       run: |
-         docker network inspect docker_default
-         sudo apt-get install -y jq
-         docker network inspect docker_default | jq ".[0].Containers | map(.Name)[]" | tee /dev/stderr | xargs --max-args 1 -- docker network disconnect -f docker_default
+        continue-on-error: true
+        run: |
+          docker network inspect docker_default
+          sudo apt-get install -y jq
+          docker network inspect docker_default | jq ".[0].Containers | map(.Name)[]" | tee /dev/stderr | xargs --max-args 1 -- docker network disconnect -f docker_default
 
       - name: Remove Docker containers
         continue-on-error: true

tests/executor/test_peekable_async_iterator.py

@@ -0,0 +1,34 @@
+from typing import AsyncIterable
+
+import pytest
+
+from yapapi.executor._smartq import PeekableAsyncIterator
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize(
+    "input",
+    [
+        [],
+        [1],
+        [1, 2],
+        [1, 2, 3],
+        [1, 2, 3, 4],
+    ],
+)
+async def test_iterator(input):
+    async def iterator():
+        for item in input:
+            yield item
+
+    it = PeekableAsyncIterator(iterator())
+    assert (await it.has_next()) == bool(input)
+
+    output = []
+
+    async for item in it:
+        output.append(item)
+        assert (await it.has_next()) == (len(output) < len(input))
+
+    assert not await it.has_next()
+    assert input == output

tests/executor/test_smartq.py

@@ -6,10 +6,16 @@
 import asyncio
 
 
+async def async_iter(iterable):
+    for item in iterable:
+        yield item
+
+
 @pytest.mark.asyncio
 @pytest.mark.parametrize("length", [0, 1, 100])
 async def test_smart_queue(length: int):
-    q = SmartQueue(range(length))
+
+    q = SmartQueue(async_iter(range(length)))
 
     async def worker(i, queue):
         print(f"worker {i} started")
@@ -46,40 +52,40 @@ async def stats():
 
 @pytest.mark.asyncio
 async def test_smart_queue_empty():
-    q: SmartQueue[None] = SmartQueue([])
+
+    q: SmartQueue = SmartQueue(async_iter([]))
     with q.new_consumer() as c:
-        async for item in c:
+        async for _item in c:
             assert False, "Expected empty list"
-    # done
 
 
 @pytest.mark.asyncio
 async def test_unassigned_items():
-    q = SmartQueue([1, 2, 3])
+    q = SmartQueue(async_iter([1, 2, 3]))
     with q.new_consumer() as c:
         async for handle in c:
-            assert q.has_new_items() == q.has_unassigned_items()
-            if not q.has_unassigned_items():
+            assert await q.has_new_items() == await q.has_unassigned_items()
+            if not await q.has_unassigned_items():
                 assert handle.data == 3
                 break
-        assert not q.has_unassigned_items()
+        assert not await q.has_unassigned_items()
         await q.reschedule_all(c)
-        assert q.has_unassigned_items()
-        assert not q.has_new_items()
+        assert await q.has_unassigned_items()
+        assert not await q.has_new_items()
 
 
 @pytest.mark.asyncio
 async def test_smart_queue_retry(caplog):
     loop = asyncio.get_event_loop()
 
     caplog.set_level(logging.DEBUG)
-    q = SmartQueue([1, 2, 3])
+    q = SmartQueue(async_iter([1, 2, 3]))
 
     async def invalid_worker(q):
         print("w start")
         with q.new_consumer() as c:
             async for item in c:
-                print("item=", item.data)
+                print("item =", item.data)
         print("w end")
 
     try:
@@ -96,11 +102,10 @@ async def invalid_worker(q):
     print("w start", q.stats())
     with q.new_consumer() as c:
         async for item in c:
-            print("item2=", item.data)
+            print("item2 =", item.data)
             assert c.current_item == item.data
             outputs.add(item.data)
             await q.mark_done(item)
     print("w end")
 
-    assert outputs == set([1, 2, 3])
-    # done
+    assert outputs == {1, 2, 3}

tests/goth/test_async_task_generation/requestor.py

@@ -0,0 +1,67 @@
+#!/usr/bin/env python3
+"""A requestor script for testing asynchronous generation of input tasks."""
+import asyncio
+from datetime import timedelta
+import pathlib
+import sys
+from typing import AsyncGenerator
+
+from yapapi import Executor, Task
+from yapapi.log import enable_default_logger, log_event_repr
+from yapapi.package import vm
+
+
+async def main():
+
+    vm_package = await vm.repo(
+        image_hash="9a3b5d67b0b27746283cb5f287c13eab1beaa12d92a9f536b747c7ae",
+        min_mem_gib=0.5,
+        min_storage_gib=2.0,
+    )
+
+    async def worker(work_ctx, tasks):
+        async for task in tasks:
+            print("task data:", task.data, file=sys.stderr)
+            work_ctx.run("/bin/sleep", "1")
+            yield work_ctx.commit()
+            task.accept_result(result=task.data)
+
+    async with Executor(
+        budget=10.0,
+        package=vm_package,
+        max_workers=1,
+        subnet_tag="goth",
+        timeout=timedelta(minutes=6),
+        event_consumer=log_event_repr,
+    ) as executor:
+
+        # We use an async task generator that yields tasks removed from
+        # an async queue. Each computed task will potentially spawn
+        # new tasks -- this is made possible thanks to using async task
+        # generator as an input to `executor.submit()`.
+
+        task_queue = asyncio.Queue()
+
+        # Seed the queue with the first task:
+        await task_queue.put(Task(data=3))
+
+        async def input_generator():
+            """Task generator yields tasks removed from `queue`."""
+            while True:
+                task = await task_queue.get()
+                if task.data == 0:
+                    break
+                yield task
+
+        async for task in executor.submit(worker, input_generator()):
+            print("task result:", task.result, file=sys.stderr)
+            for n in range(task.result):
+                await task_queue.put(Task(data=task.result - 1))
+
+        print("all done!", file=sys.stderr)
+
+
+if __name__ == "__main__":
+    test_dir = pathlib.Path(__file__).parent.name
+    enable_default_logger(log_file=f"{test_dir}.log")
+    asyncio.run(main())

tests/goth/test_async_task_generation/test_async_task_generation.py

@@ -0,0 +1,50 @@
+import logging
+import os
+from pathlib import Path
+
+import pytest
+
+import goth.configuration
+from goth.runner import Runner
+from goth.runner.log import configure_logging
+from goth.runner.probe import RequestorProbe
+
+
+logger = logging.getLogger("goth.test.async_task_generation")
+
+
+@pytest.mark.asyncio
+async def test_async_task_generation(project_dir: Path, log_dir: Path, config_overrides) -> None:
+    """Run the `requestor.py` and make sure that it's standard output is as expected."""
+
+    configure_logging(log_dir)
+
+    # Override the default test configuration to create only one provider node
+    nodes = [
+        {"name": "requestor", "type": "Requestor"},
+        {"name": "provider-1", "type": "VM-Wasm-Provider", "use-proxy": True},
+    ]
+    config_overrides.append(("nodes", nodes))
+    goth_config = goth.configuration.load_yaml(
+        project_dir / "tests" / "goth" / "assets" / "goth-config.yml",
+        config_overrides,
+    )
+
+    runner = Runner(base_log_dir=log_dir, compose_config=goth_config.compose_config)
+
+    async with runner(goth_config.containers):
+
+        requestor = runner.get_probes(probe_type=RequestorProbe)[0]
+
+        async with requestor.run_command_on_host(
+            str(Path(__file__).parent / "requestor.py"), env=os.environ
+        ) as (_cmd_task, cmd_monitor):
+            # The requestor should print "task result: 3" once ...
+            await cmd_monitor.wait_for_pattern("task result: 3", timeout=60)
+            # ... then "task result: 2" twice ...
+            for _ in range(3):
+                await cmd_monitor.wait_for_pattern("task result: 2", timeout=10)
+            # ... and "task result: 1" six times.
+            for _ in range(6):
+                await cmd_monitor.wait_for_pattern("task result: 1", timeout=10)
+            await cmd_monitor.wait_for_pattern("all done!", timeout=10)

yapapi/executor/__init__.py

@@ -206,7 +206,7 @@ def strategy(self) -> MarketStrategy:
     async def submit(
         self,
         worker: Callable[[WorkContext, AsyncIterator[Task[D, R]]], AsyncGenerator[Work, None]],
-        data: Iterable[Task[D, R]],
+        data: Union[AsyncIterator[Task[D, R]], Iterable[Task[D, R]]],
     ) -> AsyncIterator[Task[D, R]]:
         """Submit a computation to be executed on providers.
 
@@ -388,7 +388,7 @@ async def _find_offers(self, state: "Executor.SubmissionState") -> None:
     async def _submit(
         self,
         worker: Callable[[WorkContext, AsyncIterator[Task[D, R]]], AsyncGenerator[Work, None]],
-        data: Iterable[Task[D, R]],
+        data: Union[AsyncIterator[Task[D, R]], Iterable[Task[D, R]]],
         services: Set[asyncio.Task],
         workers: Set[asyncio.Task],
     ) -> AsyncGenerator[Task[D, R], None]:
@@ -415,7 +415,6 @@ async def _submit(
 
         state = Executor.SubmissionState(builder, agreements_pool)
 
-        market_api = self._market_api
         activity_api: rest.Activity = self._activity_api
 
         done_queue: asyncio.Queue[Task[D, R]] = asyncio.Queue()
@@ -426,10 +425,15 @@ def on_task_done(task: Task[D, R], status: TaskStatus) -> None:
             if status == TaskStatus.ACCEPTED:
                 done_queue.put_nowait(task)
 
-        def input_tasks() -> Iterable[Task[D, R]]:
-            for task in data:
-                task._add_callback(on_task_done)
-                yield task
+        async def input_tasks() -> AsyncIterator[Task[D, R]]:
+            if isinstance(data, AsyncIterator):
+                async for task in data:
+                    task._add_callback(on_task_done)
+                    yield task
+            else:
+                for task in data:
+                    task._add_callback(on_task_done)
+                    yield task
 
         work_queue = SmartQueue(input_tasks())
 
@@ -607,7 +611,10 @@ async def worker_starter() -> None:
             while True:
                 await asyncio.sleep(2)
                 await agreements_pool.cycle()
-                if len(workers) < self._conf.max_workers and work_queue.has_unassigned_items():
+                if (
+                    len(workers) < self._conf.max_workers
+                    and await work_queue.has_unassigned_items()
+                ):
                     new_task = None
                     try:
                         new_task = await agreements_pool.use_agreement(