def test_docker_shm_override(run_option_type):
process_runner = MockProcessRunner()
provider = MockProvider()
provider.create_node({}, {}, 1)
cluster_name = "cluster"
docker_config = {
"container_name": "container",
"image": "rayproject/ray:latest",
run_option_type: ["--shm-size=80g"]
}
args = {
"log_prefix": "prefix",
"node_id": 0,
"provider": provider,
"auth_config": auth_config,
"cluster_name": cluster_name,
"process_runner": process_runner,
"use_internal_ip": False,
"docker_config": docker_config,
}
cmd_runner = DockerCommandRunner(**args)
process_runner.respond_to_call("json .Config.Env", 2 * ["[]"])
cmd_runner.run_init(as_head=True, file_mounts={}, sync_run_yet=True)
# Ensure the user-provided SHM size is used.
process_runner.assert_has_call("1.2.3.4", pattern="--shm-size=80g")
# Ensure that SHM auto detection is bypassed
process_runner.assert_not_has_call("1.2.3.4", pattern="/proc/meminfo")
def test_docker_command_runner():
process_runner = MockProcessRunner()
provider = MockProvider()
provider.create_node({}, {}, 1)
cluster_name = "cluster"
ssh_control_hash = hashlib.md5(cluster_name.encode()).hexdigest()
ssh_user_hash = hashlib.md5(getuser().encode()).hexdigest()
ssh_control_path = "/tmp/ray_ssh_{}/{}".format(ssh_user_hash[:10],
ssh_control_hash[:10])
docker_config = {"container_name": "container"}
args = {
"log_prefix": "prefix",
"node_id": 0,
"provider": provider,
"auth_config": auth_config,
"cluster_name": cluster_name,
"process_runner": process_runner,
"use_internal_ip": False,
"docker_config": docker_config,
}
cmd_runner = DockerCommandRunner(**args)
assert len(process_runner.calls) == 0, "No calls should be made in ctor"
env_vars = {"var1": "quote between this \" and this", "var2": "123"}
cmd_runner.run("echo hello", environment_variables=env_vars)
# This string is insane because there are an absurd number of embedded
# quotes. While this is a ridiculous string, the escape behavior is
# important and somewhat difficult to get right for environment variables.
cmd = """'true && source ~/.bashrc && export OMP_NUM_THREADS=1 PYTHONWARNINGS=ignore && (docker exec -it container /bin/bash -c '"'"'bash --login -c -i '"'"'"'"'"'"'"'"'true && source ~/.bashrc && export OMP_NUM_THREADS=1 PYTHONWARNINGS=ignore && (export var1='"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"quote between this \\" and this"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"';export var2='"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"123"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"'"';echo hello)'"'"'"'"'"'"'"'"''"'"' )'""" # noqa: E501
expected = [
"ssh", "-tt", "-i", "8265.pem", "-o", "StrictHostKeyChecking=no", "-o",
"UserKnownHostsFile=/dev/null", "-o", "IdentitiesOnly=yes", "-o",
"ExitOnForwardFailure=yes", "-o", "ServerAliveInterval=5", "-o",
"ServerAliveCountMax=3", "-o", "ControlMaster=auto", "-o",
"ControlPath={}/%C".format(ssh_control_path), "-o",
"ControlPersist=10s", "-o", "ConnectTimeout=120s", "[email protected]",
"bash", "--login", "-c", "-i", cmd
]
# Much easier to debug this loop than the function call.
for x, y in zip(process_runner.calls[0], expected):
print(f"expeted:\t{y}")
print(f"actual: \t{x}")
assert x == y
process_runner.assert_has_call("1.2.3.4", exact=expected)
def get_command_runner(
self,
log_prefix: str,
node_id: str,
auth_config: Dict[str, Any],
cluster_name: str,
process_runner: ModuleType,
use_internal_ip: bool,
docker_config: Optional[Dict[str, Any]] = None,
) -> CommandRunnerInterface:
"""Returns the CommandRunner class used to perform SSH commands.
Args:
log_prefix(str): stores "NodeUpdater: {}: ".format(<node_id>). Used
to print progress in the CommandRunner.
node_id(str): the node ID.
auth_config(dict): the authentication configs from the autoscaler
yaml file.
cluster_name(str): the name of the cluster.
process_runner(module): the module to use to run the commands
in the CommandRunner. E.g., subprocess.
use_internal_ip(bool): whether the node_id belongs to an internal ip
or external ip.
docker_config(dict): If set, the docker information of the docker
container that commands should be run on.
"""
common_args = {
"log_prefix": log_prefix,
"node_id": node_id,
"provider": self,
"auth_config": auth_config,
"cluster_name": cluster_name,
"process_runner": process_runner,
"use_internal_ip": use_internal_ip,
}
if docker_config and docker_config["container_name"] != "":
return DockerCommandRunner(docker_config, **common_args)
else:
return SSHCommandRunner(**common_args)
def test_docker_rsync():
process_runner = MockProcessRunner()
provider = MockProvider()
provider.create_node({}, {}, 1)
cluster_name = "cluster"
docker_config = {"container_name": "container"}
args = {
"log_prefix": "prefix",
"node_id": 0,
"provider": provider,
"auth_config": auth_config,
"cluster_name": cluster_name,
"process_runner": process_runner,
"use_internal_ip": False,
"docker_config": docker_config,
}
cmd_runner = DockerCommandRunner(**args)
local_mount = "/home/ubuntu/base/mount/"
remote_mount = "/root/protected_mount/"
docker_mount_prefix = get_docker_host_mount_location(cluster_name)
remote_host_mount = f"{docker_mount_prefix}{remote_mount}"
local_file = "/home/ubuntu/base-file"
remote_file = "/root/protected-file"
remote_host_file = f"{docker_mount_prefix}{remote_file}"
process_runner.respond_to_call("docker inspect -f", ["true"])
cmd_runner.run_rsync_up(local_mount,
remote_mount,
options={"docker_mount_if_possible": True})
# Make sure we do not copy directly to raw destination
process_runner.assert_not_has_call(
"1.2.3.4", pattern=f"-avz {local_mount} [email protected]:{remote_mount}")
process_runner.assert_not_has_call("1.2.3.4",
pattern=f"mkdir -p {remote_mount}")
# No docker cp for file_mounts
process_runner.assert_not_has_call("1.2.3.4", pattern="docker cp")
process_runner.assert_has_call(
"1.2.3.4",
pattern=f"-avz {local_mount} [email protected]:{remote_host_mount}")
process_runner.clear_history()
##############################
process_runner.respond_to_call("docker inspect -f", ["true"])
cmd_runner.run_rsync_up(local_file,
remote_file,
options={"docker_mount_if_possible": False})
# Make sure we do not copy directly to raw destination
process_runner.assert_not_has_call(
"1.2.3.4", pattern=f"-avz {local_file} [email protected]:{remote_file}")
process_runner.assert_not_has_call("1.2.3.4",
pattern=f"mkdir -p {remote_file}")
process_runner.assert_has_call("1.2.3.4", pattern="docker cp")
process_runner.assert_has_call(
"1.2.3.4", pattern=f"-avz {local_file} [email protected]:{remote_host_file}")
process_runner.clear_history()
##############################
cmd_runner.run_rsync_down(remote_mount,
local_mount,
options={"docker_mount_if_possible": True})
process_runner.assert_not_has_call("1.2.3.4", pattern="docker cp")
process_runner.assert_not_has_call(
"1.2.3.4", pattern=f"-avz [email protected]:{remote_mount} {local_mount}")
process_runner.assert_has_call(
"1.2.3.4",
pattern=f"-avz [email protected]:{remote_host_mount} {local_mount}")
process_runner.clear_history()
##############################
cmd_runner.run_rsync_down(remote_file,
local_file,
options={"docker_mount_if_possible": False})
process_runner.assert_has_call("1.2.3.4", pattern="docker cp")
process_runner.assert_not_has_call(
"1.2.3.4", pattern=f"-avz [email protected]:{remote_file} {local_file}")
process_runner.assert_has_call(
"1.2.3.4", pattern=f"-avz [email protected]:{remote_host_file} {local_file}")