async def main():
async with Golem(budget=1.0, subnet_tag="devnet-beta") as golem:
cluster = await golem.run_service(DateService, num_instances=1)
start_time = datetime.now()
while datetime.now() < start_time + timedelta(minutes=1):
for num, instance in enumerate(cluster.instances):
print(
f"Instance {num} is {instance.state.value} on {instance.provider_name}"
)
await asyncio.sleep(REFRESH_INTERVAL_SEC)
async def test_emit_event_class(dummy_yagna_engine):
got_events_1 = []
got_events_2 = []
got_events_3 = []
def event_consumer_1(event: events.Event):
got_events_1.append(event)
def event_consumer_2(event: events.Event):
got_events_2.append(event)
def event_consumer_3(event: events.Event):
got_events_3.append(event)
golem = Golem(budget=1, app_key="NOT_A_REAL_APPKEY")
golem.add_event_consumer(event_consumer_1, [events.SubscriptionEvent])
golem.add_event_consumer(event_consumer_2,
[events.TaskEvent, "ServiceEvent"])
golem.add_event_consumer(event_consumer_3)
async with golem:
invoice_received = golem._engine.emit(events.InvoiceReceived,
job="foo",
agreement="bar",
invoice="baz")
subscription_created = golem._engine.emit(events.SubscriptionCreated,
job="foo",
subscription="bar")
service_finished = golem._engine.emit(events.ServiceFinished,
job="foo",
agreement="bar",
activity="baz",
service="???")
task_started = golem._engine.emit(events.TaskStarted,
job="foo",
agreement="bar",
activity="baz",
task="???")
assert got_events_1 == [subscription_created]
assert got_events_2 == [service_finished, task_started]
# Additional ShutdownFinished event is passed to the catchall consumer only
assert got_events_3[:4] == [
invoice_received,
subscription_created,
service_finished,
task_started,
]
assert len(got_events_3) == 5 and isinstance(got_events_3[4],
events.ShutdownFinished)
async def main():
package = await vm.repo(
image_hash="d646d7b93083d817846c2ae5c62c72ca0507782385a2e29291a3d376",
)
tasks = [Task(data=None)]
async with Golem(budget=1.0, subnet_tag="devnet-beta") as golem:
async for completed in golem.execute_tasks(worker,
tasks,
payload=package):
print(completed.result.stdout)
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:
script = work_ctx.new_script()
print("task data:", task.data, file=sys.stderr)
script.run("/bin/sleep", "1")
yield script
task.accept_result(result=task.data)
async with Golem(
budget=10.0,
subnet_tag="goth",
event_consumer=log_event_repr,
) as golem:
# 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 golem.execute_tasks(
worker,
input_generator(),
vm_package,
max_workers=1,
timeout=timedelta(minutes=6),
):
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)
async def test_emit_event(dummy_yagna_engine):
got_events_1 = []
got_events_2 = []
got_events_3 = []
got_events_4 = []
def event_consumer_1(event: events.Event) -> None:
# Event consumer passed to Golem()
got_events_1.append(event)
def event_consumer_2(event: events.Event) -> None:
# Event consumer passed via add_event_consumer to not-yet-started Golem
got_events_2.append(event)
def event_consumer_3(event: events.Event) -> None:
# Event consumer passed to working Golem
got_events_3.append(event)
golem = Golem(budget=1,
event_consumer=event_consumer_1,
app_key="NOT_A_REAL_APPKEY")
golem.add_event_consumer(event_consumer_2)
async with golem:
golem.add_event_consumer(event_consumer_3)
for sample_event in sample_events:
event_class, event_kwargs = type(sample_event), asdict(
sample_event)
del event_kwargs["timestamp"] # timestamp is set only internally
emitted_event = golem._engine.emit(event_class, **event_kwargs)
got_events_4.append(emitted_event)
assert got_events_1 == emitted_events
assert got_events_2 == emitted_events
assert got_events_3 == emitted_events
assert got_events_4 == sample_events
# When exiting from golem contextmanager, new `_Engine` is created -> we must ensure
# event consumers are not lost when `_Engine` exits -> so we repeat the same test again
got_events_1.clear()
got_events_2.clear()
got_events_3.clear()
got_events_4.clear()
async with golem:
for sample_event in sample_events:
event_class, event_kwargs = type(sample_event), asdict(
sample_event)
del event_kwargs["timestamp"] # timestamp is set only internally
emitted_event = golem._engine.emit(event_class, **event_kwargs)
got_events_4.append(emitted_event)
assert got_events_1 == emitted_events
assert got_events_2 == emitted_events
assert got_events_3 == emitted_events
assert got_events_4 == sample_events
async def main():
async with Golem(budget=1.0, subnet_tag="goth") as golem:
# Start a cluster with a single service.
# The first instance will fail before reaching the `running` state
# due to an error. Another instance should be spawned in its place.
log("Starting cluster...")
await golem.run_service(FirstInstanceFailsToStart)
# This another instance should get to `running` state.
while not instances_running:
log("Waiting for an instance...")
await asyncio.sleep(2)
# And then stop.
while not instances_stopped:
log("Waiting until the instance stops...")
await asyncio.sleep(2)
# Then we start another cluster with a single instance.
# This time the instance is stopped by a signal before it reaches the `running` state,
# but in that case the cluster should not spawn another instance.
log("Starting another cluster...")
cluster = await golem.run_service(FirstInstanceFailsToStart)
while instances_started < 3:
log("Waiting for another instance...")
await asyncio.sleep(2)
assert [i for i in cluster.instances if i.is_available]
log("Closing the second cluster...")
cluster.stop()
while [i for i in cluster.instances if i.is_available]:
log("Waiting for the cluster to stop...")
await asyncio.sleep(2)
log("Cluster stopped")
async def main(running_time_sec, subnet_tag, driver=None, network=None):
# necessary to set maximum price for one unit of custom counter,
# default strategy does not take custom counter prices into account
strategy = LeastExpensiveLinearPayuMS(
max_price_for={
com.Counter.CPU.value: Decimal("0.2"),
com.Counter.TIME.value: Decimal("0.1"),
"golem.usage.custom.counter": Decimal("0.1"),
})
async with Golem(
budget=10.0,
subnet_tag=subnet_tag,
payment_driver=driver,
payment_network=network,
strategy=strategy,
) as golem:
instance_params = [{"running_time_sec": running_time_sec}]
cluster = await golem.run_service(CustomCounterService,
instance_params=instance_params)
def print_instances():
print(
f"instances: {[{s.id, s.state.value} for s in cluster.instances]}"
)
was_running = False
while True:
await asyncio.sleep(3)
if was_running and all([
s.state == ServiceState.terminated
for s in cluster.instances
]):
print_instances()
print("All services were successfully terminated")
break
if len(cluster.instances) > 0:
print_instances()
was_running = True
async def main():
package = await vm.repo(
image_hash="d646d7b93083d817846c2ae5c62c72ca0507782385a2e29291a3d376")
tasks = [Task(data=None)]
timeout = timedelta(hours=24)
async with Golem(
budget=10.0,
strategy=ShortDebitNoteIntervalAndPaymentTimeout(),
subnet_tag="goth",
event_consumer=log_event_repr,
) as golem:
logger = logging.getLogger("yapapi")
logger.handlers[0].setLevel(logging.DEBUG)
async for completed in golem.execute_tasks(worker,
tasks,
payload=package,
max_workers=1,
timeout=timeout):
print(f"Task finished: {completed}.")
async def main(subnet_tag,
payment_driver=None,
payment_network=None,
num_instances=2):
# By passing `event_consumer=log_summary()` we enable summary logging.
# See the documentation of the `yapapi.log` module on how to set
# the level of detail and format of the logged information.
async with Golem(
budget=1.0,
subnet_tag=subnet_tag,
payment_driver=payment_driver,
payment_network=payment_network,
) as golem:
print_env_info(golem)
network = await golem.create_network("192.168.0.1/24")
async with network:
cluster = await golem.run_service(SshService,
network=network,
num_instances=num_instances)
instances = cluster.instances
while True:
print(instances)
try:
await asyncio.sleep(5)
except (KeyboardInterrupt, asyncio.CancelledError):
break
cluster.stop()
cnt = 0
while cnt < 3 and any(s.is_available for s in instances):
print(instances)
await asyncio.sleep(5)
cnt += 1
async def main(subnet_tag, payment_driver=None, payment_network=None):
package = await vm.repo(
image_hash="a23ce2c0c29ea9711e4a293a2805700e2f0cb6450fddf9506812eb1b",
min_mem_gib=0.5,
min_storage_gib=2.0,
)
tasks = [() for _ in range(10)]
async def worker(ctx: WorkContext, tasks):
async for task in tasks:
output_file = f"output_{datetime.now()}_{random.random()}.txt"
script = ctx.new_script(timeout=timedelta(minutes=10))
script.run("/usr/bin/stress-ng", "--cpu", "1", "--timeout", "1")
script.run("/golem/task.sh", "-o", "1024", "-t", "5")
script.run("/golem/task.sh", "-f",
"/golem/output/output.txt,1048576")
script.download_file(f"/golem/output/output.txt", output_file)
script.run("/golem/task.sh", "-e", "1024", "-t", "5")
try:
yield script
task.accept_result(result=output_file)
except BatchTimeoutError:
print(
f"{TEXT_COLOR_RED}"
f"Task {task} timed out on {ctx.provider_name}, time: {task.running_time}"
f"{TEXT_COLOR_DEFAULT}")
raise
async with Golem(
budget=10.0,
subnet_tag=subnet_tag,
payment_driver=payment_driver,
payment_network=payment_network,
) as golem:
print(
f"yapapi version: {TEXT_COLOR_YELLOW}{yapapi_version}{TEXT_COLOR_DEFAULT}\n"
f"Using subnet: {TEXT_COLOR_YELLOW}{subnet_tag}{TEXT_COLOR_DEFAULT}, "
f"payment driver: {TEXT_COLOR_YELLOW}{golem.payment_driver}{TEXT_COLOR_DEFAULT}, "
f"and network: {TEXT_COLOR_YELLOW}{golem.payment_network}{TEXT_COLOR_DEFAULT}\n"
)
num_tasks = 0
start_time = datetime.now()
completed_tasks = golem.execute_tasks(
worker,
[Task(data=data) for data in tasks],
payload=package,
max_workers=50,
timeout=timedelta(minutes=30),
)
async for task in completed_tasks:
num_tasks += 1
print(
f"{TEXT_COLOR_CYAN}"
f"Task computed: {task}, result: {task.result}, time: {task.running_time}"
f"{TEXT_COLOR_DEFAULT}")
print(
f"{TEXT_COLOR_CYAN}"
f"{num_tasks} tasks computed, total time: {datetime.now() - start_time}"
f"{TEXT_COLOR_DEFAULT}")
if __name__ == "__main__":
parser = build_parser("Render a Blender scene")
parser.add_argument("--show-usage", action="store_true", help="show activity usage and cost")
parser.add_argument(
"--min-cpu-threads",
type=int,
default=2,
help="require the provider nodes to have at least this number of available CPU threads",
)
now = datetime.now().strftime("%Y-%m-%d_%H.%M.%S")
parser.set_defaults(log_file=f"blender-yapapi-{now}.log")
args = parser.parse_args()
golem = Golem(
budget=10,
subnet_tag=args.subnet_tag,
payment_driver=args.payment_driver,
payment_network=args.payment_network,
)
run_golem_example(
main(
golem,
show_usage=args.show_usage,
min_cpu_threads=args.min_cpu_threads,
),
log_file=args.log_file,
)
async def main(subnet_tag, payment_driver, payment_network, port):
async with Golem(
budget=1.0,
subnet_tag=subnet_tag,
payment_driver=payment_driver,
payment_network=payment_network,
) as golem:
print_env_info(golem)
network = await golem.create_network("192.168.0.1/24")
async with network:
db_cluster = await golem.run_service(DbService, network=network)
db_instance = db_cluster.instances[0]
def still_starting(cluster):
return any(i.state in (ServiceState.pending,
ServiceState.starting)
for i in cluster.instances)
def raise_exception_if_still_starting(cluster):
if still_starting(cluster):
raise Exception(
f"Failed to start {cluster} instances "
f"after {STARTING_TIMEOUT.total_seconds()} seconds")
commissioning_time = datetime.now()
while (still_starting(db_cluster)
and datetime.now() < commissioning_time + STARTING_TIMEOUT):
print(db_cluster.instances)
await asyncio.sleep(5)
raise_exception_if_still_starting(db_cluster)
print(
f"{TEXT_COLOR_CYAN}DB instance started, spawning the web server{TEXT_COLOR_DEFAULT}"
)
web_cluster = await golem.run_service(
HttpService,
network=network,
instance_params=[{
"db_address": db_instance.network_node.ip
}],
)
# wait until all remote http instances are started
while (still_starting(web_cluster)
and datetime.now() < commissioning_time + STARTING_TIMEOUT):
print(web_cluster.instances + db_cluster.instances)
await asyncio.sleep(5)
raise_exception_if_still_starting(web_cluster)
# service instances started, start the local HTTP server
proxy = LocalHttpProxy(web_cluster, port)
await proxy.run()
print(
f"{TEXT_COLOR_CYAN}Local HTTP server listening on:\nhttp://localhost:{port}{TEXT_COLOR_DEFAULT}"
)
# wait until Ctrl-C
while True:
print(web_cluster.instances + db_cluster.instances)
try:
await asyncio.sleep(10)
except (KeyboardInterrupt, asyncio.CancelledError):
break
# perform the shutdown of the local http server and the service cluster
await proxy.stop()
print(f"{TEXT_COLOR_CYAN}HTTP server stopped{TEXT_COLOR_DEFAULT}")
web_cluster.stop()
db_cluster.stop()
cnt = 0
while cnt < 3 and any(
s.is_available
for s in web_cluster.instances + db_cluster.instances):
print(web_cluster.instances + db_cluster.instances)
await asyncio.sleep(5)
cnt += 1
async def main(args):
package = await vm.repo(
image_hash="055911c811e56da4d75ffc928361a78ed13077933ffa8320fb1ec2db",
min_mem_gib=0.5,
min_storage_gib=2.0,
)
async with Golem(
budget=10.0,
subnet_tag=args.subnet_tag,
payment_driver=args.payment_driver,
payment_network=args.payment_network,
) as golem:
print_env_info(golem)
start_time = datetime.now()
completed = golem.execute_tasks(
compute_keyspace,
[Task(data="compute_keyspace")],
payload=package,
timeout=KEYSPACE_TIMEOUT_ALL_TASKS,
)
keyspace = 0
async for task in completed:
keyspace = task.result
print(
f"{TEXT_COLOR_CYAN}"
f"Task computed: keyspace size count. The keyspace size is {keyspace}"
f"{TEXT_COLOR_DEFAULT}"
)
data = (Task(data=c) for c in range(0, keyspace, args.chunk_size))
max_workers = args.max_workers or math.ceil(keyspace / args.chunk_size) // 2
completed = golem.execute_tasks(
perform_mask_attack,
data,
payload=package,
max_workers=max_workers,
timeout=MASK_ATTACK_TIMEOUT,
)
password = None
async for task in completed:
print(
f"{TEXT_COLOR_CYAN}Task computed: {task}, result: {task.result}{TEXT_COLOR_DEFAULT}"
)
result = _parse_result(task.result)
if result:
password = result
if password:
print(f"{TEXT_COLOR_GREEN}Password found: {password}{TEXT_COLOR_DEFAULT}")
else:
print(f"{TEXT_COLOR_RED}No password found{TEXT_COLOR_DEFAULT}")
print(f"{TEXT_COLOR_CYAN}Total time: {datetime.now() - start_time}{TEXT_COLOR_DEFAULT}")
async def main(subnet_tag, payment_driver, payment_network, num_instances, port, running_time):
async with Golem(
budget=1.0,
subnet_tag=subnet_tag,
payment_driver=payment_driver,
payment_network=payment_network,
) as golem:
print_env_info(golem)
commissioning_time = datetime.now()
network = await golem.create_network("192.168.0.1/24")
cluster = await golem.run_service(
HttpService,
network=network,
num_instances=num_instances,
expiration=datetime.now(timezone.utc)
+ STARTING_TIMEOUT
+ EXPIRATION_MARGIN
+ timedelta(seconds=running_time),
)
instances = cluster.instances
def still_starting():
return any(i.state in (ServiceState.pending, ServiceState.starting) for i in instances)
# wait until all remote http instances are started
while still_starting() and datetime.now() < commissioning_time + STARTING_TIMEOUT:
print(f"instances: {instances}")
await asyncio.sleep(5)
if still_starting():
raise Exception(
f"Failed to start instances after {STARTING_TIMEOUT.total_seconds()} seconds"
)
# service instances started, start the local HTTP server
proxy = LocalHttpProxy(cluster, port)
await proxy.run()
print(
f"{TEXT_COLOR_CYAN}Local HTTP server listening on:\nhttp://localhost:{port}{TEXT_COLOR_DEFAULT}"
)
# wait until Ctrl-C
start_time = datetime.now()
while datetime.now() < start_time + timedelta(seconds=running_time):
print(instances)
try:
await asyncio.sleep(10)
except (KeyboardInterrupt, asyncio.CancelledError):
break
# perform the shutdown of the local http server and the service cluster
await proxy.stop()
print(f"{TEXT_COLOR_CYAN}HTTP server stopped{TEXT_COLOR_DEFAULT}")
cluster.stop()
cnt = 0
while cnt < 3 and any(s.is_available for s in instances):
print(instances)
await asyncio.sleep(5)
cnt += 1
await network.remove()
async def test_incorrect_event_class_str():
golem = Golem(budget=1, app_key="NOT_A_REAL_APPKEY")
with pytest.raises(ValueError):
golem.add_event_consumer(lambda event: event, ["NoSuchEvent"])
async def main():
vm_package = await vm.repo(
image_hash="9a3b5d67b0b27746283cb5f287c13eab1beaa12d92a9f536b747c7ae",
min_mem_gib=0.5,
min_storage_gib=2.0,
)
first_task = True
async def duplicator(work_ctx, tasks):
"""A worker that executes `echo {task.data} {task.data}` command on provider.
The command's output is the result of the whole task.
The first ever task computed by an instance of this worker fails.
It succeeds when re-tried. This may be used to test re-trying tasks,
creating new agreements and displaying summary information on task/activity
failures.
"""
async for task in tasks:
nonlocal first_task
script = work_ctx.new_script()
if first_task:
first_task = False
script.run("/bin/sleep", "1")
script.run("/command/not/found")
script.run("/bin/sleep", "1")
future_result = script.run("/bin/echo", task.data, task.data)
yield script
result = await future_result
output = result.stdout.strip()
task.accept_result(output)
async with Golem(budget=1.0, subnet_tag="goth") as golem:
# Construct a pipeline:
#
# input_tasks
# |
# V
# [ Job ALEF ]
# |
# V
# intermediate_tasks
# |
# V
# [ Job BET ]
# |
# V
# output_tasks
input_tasks = [Task(s) for s in "01234567"]
computed_input_tasks = golem.execute_tasks(
duplicator,
input_tasks,
payload=vm_package,
max_workers=1,
job_id="ALEF",
)
async def intermediate_tasks():
async for task in computed_input_tasks:
print(f"ALEF computed task: {task.data} -> {task.result}", file=sys.stderr)
yield Task(data=task.result)
output_tasks = golem.execute_tasks(
duplicator,
intermediate_tasks(),
payload=vm_package,
max_workers=1,
job_id="BET",
)
async for task in output_tasks:
print(f"BET computed task: {task.data} -> {task.result}", file=sys.stderr)
async def main(
subnet_tag,
running_time,
payment_driver=None,
payment_network=None,
num_instances=1,
show_usage=False,
):
async with Golem(
budget=1.0,
subnet_tag=subnet_tag,
payment_driver=payment_driver,
payment_network=payment_network,
) as golem:
print_env_info(golem)
commissioning_time = datetime.now()
# start the service
cluster = await golem.run_service(
SimpleService,
instance_params=[
{"instance_name": f"simple-service-{i+1}", "show_usage": show_usage}
for i in range(num_instances)
],
expiration=datetime.now(timezone.utc)
+ STARTING_TIMEOUT
+ EXPIRATION_MARGIN
+ timedelta(seconds=running_time),
)
print(f"{TEXT_COLOR_YELLOW}" f"Starting {cluster}..." f"{TEXT_COLOR_DEFAULT}")
def print_instances():
print(
f"instances: "
+ str(
[
f"{s.name}: {s.state.value}"
+ (f" on {s.provider_name}" if s.provider_id else "")
for s in cluster.instances
]
)
)
def still_starting():
return any(
i.state in (ServiceState.pending, ServiceState.starting) for i in cluster.instances
)
# wait until instances are started
while still_starting() and datetime.now() < commissioning_time + STARTING_TIMEOUT:
print_instances()
await asyncio.sleep(5)
if still_starting():
raise Exception(f"Failed to start instances before {STARTING_TIMEOUT} elapsed :( ...")
print(f"{TEXT_COLOR_YELLOW}All instances started :){TEXT_COLOR_DEFAULT}")
# allow the service to run for a short while
# (and allowing its requestor-end handlers to interact with it)
start_time = datetime.now()
while datetime.now() < start_time + timedelta(seconds=running_time):
print_instances()
await asyncio.sleep(5)
print(f"{TEXT_COLOR_YELLOW}Stopping {cluster}...{TEXT_COLOR_DEFAULT}")
cluster.stop()
# wait for instances to stop
cnt = 0
while cnt < 10 and any(s.is_available for s in cluster.instances):
print_instances()
await asyncio.sleep(5)
print_instances()
async def test_demand_resubscription(log_dir: Path, goth_config_path: Path,
monkeypatch) -> None:
"""Test that checks that a demand is re-submitted after its previous submission expires."""
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
},
]
goth_config = load_yaml(goth_config_path, [("nodes", nodes)])
vm_package = await vm.repo(
image_hash="9a3b5d67b0b27746283cb5f287c13eab1beaa12d92a9f536b747c7ae",
min_mem_gib=0.5,
min_storage_gib=2.0,
)
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]
env = dict(os.environ)
env.update(requestor.get_agent_env_vars())
# Setup the environment for the requestor
for key, val in env.items():
monkeypatch.setenv(key, val)
monitor = EventMonitor()
monitor.add_assertion(assert_demand_resubscribed)
monitor.start()
# The requestor
enable_default_logger()
async def worker(work_ctx, tasks):
async for task in tasks:
script = work_ctx.new_script()
script.run("/bin/sleep", "5")
yield script
task.accept_result()
async with Golem(
budget=10.0,
event_consumer=monitor.add_event_sync,
) as golem:
task: Task # mypy needs this for some reason
async for task in golem.execute_tasks(
worker,
[Task(data=n) for n in range(20)],
vm_package,
max_workers=1,
timeout=timedelta(seconds=30),
):
logger.info("Task %d computed", task.data)
await monitor.stop()
for a in monitor.failed:
raise a.result()
