def _run(self,
batch: 'batch.Batch',
dry_run: bool,
verbose: bool,
delete_scratch_on_exit: bool,
wait: bool = True,
open: bool = False,
disable_progress_bar: bool = False,
callback: Optional[str] = None,
token: Optional[str] = None,
**backend_kwargs): # pylint: disable-msg=too-many-statements
"""Execute a batch.
Warning
-------
This method should not be called directly. Instead, use :meth:`.batch.Batch.run`
and pass :class:`.ServiceBackend` specific arguments as key-word arguments.
Parameters
----------
batch:
Batch to execute.
dry_run:
If `True`, don't execute code.
verbose:
If `True`, print debugging output.
delete_scratch_on_exit:
If `True`, delete temporary directories with intermediate files.
wait:
If `True`, wait for the batch to finish executing before returning.
open:
If `True`, open the UI page for the batch.
disable_progress_bar:
If `True`, disable the progress bar.
callback:
If not `None`, a URL that will receive at most one POST request
after the entire batch completes.
token:
If not `None`, a string used for idempotency of batch submission.
"""
if backend_kwargs:
raise ValueError(f'ServiceBackend does not support any of these keywords: {backend_kwargs}')
build_dag_start = time.time()
uid = uuid.uuid4().hex[:6]
remote_tmpdir = f'gs://{self._bucket_name}/batch/{uid}'
local_tmpdir = f'/io/batch/{uid}'
default_image = 'ubuntu:18.04'
attributes = copy.deepcopy(batch.attributes)
if batch.name is not None:
attributes['name'] = batch.name
bc_batch = self._batch_client.create_batch(attributes=attributes, callback=callback,
token=token, cancel_after_n_failures=batch._cancel_after_n_failures)
n_jobs_submitted = 0
used_remote_tmpdir = False
job_to_client_job_mapping: Dict[_job.Job, bc.Job] = {}
jobs_to_command = {}
commands = []
bash_flags = 'set -e' + ('x' if verbose else '')
activate_service_account = 'gcloud -q auth activate-service-account ' \
'--key-file=/gsa-key/key.json'
def copy_input(r):
if isinstance(r, resource.InputResourceFile):
return [(r._input_path, r._get_path(local_tmpdir))]
assert isinstance(r, (resource.JobResourceFile, resource.PythonResult))
return [(r._get_path(remote_tmpdir), r._get_path(local_tmpdir))]
def copy_internal_output(r):
assert isinstance(r, (resource.JobResourceFile, resource.PythonResult))
return [(r._get_path(local_tmpdir), r._get_path(remote_tmpdir))]
def copy_external_output(r):
if isinstance(r, resource.InputResourceFile):
return [(r._input_path, dest) for dest in r._output_paths]
assert isinstance(r, (resource.JobResourceFile, resource.PythonResult))
return [(r._get_path(local_tmpdir), dest) for dest in r._output_paths]
def symlink_input_resource_group(r):
symlinks = []
if isinstance(r, resource.ResourceGroup) and r._source is None:
for name, irf in r._resources.items():
src = irf._get_path(local_tmpdir)
dest = f'{r._get_path(local_tmpdir)}.{name}'
symlinks.append(f'ln -sf {shq(src)} {shq(dest)}')
return symlinks
write_external_inputs = [x for r in batch._input_resources for x in copy_external_output(r)]
if write_external_inputs:
def _cp(src, dst):
return f'gsutil -m cp -R {shq(src)} {shq(dst)}'
write_cmd = f'''
{bash_flags}
{activate_service_account}
{' && '.join([_cp(*files) for files in write_external_inputs])}
'''
if dry_run:
commands.append(write_cmd)
else:
j = bc_batch.create_job(image='gcr.io/google.com/cloudsdktool/cloud-sdk:310.0.0-alpine',
command=['/bin/bash', '-c', write_cmd],
attributes={'name': 'write_external_inputs'})
jobs_to_command[j] = write_cmd
n_jobs_submitted += 1
for job in batch._jobs:
if isinstance(job, _job.PythonJob):
if job._image is None:
version = sys.version_info
if version.major != 3 or version.minor not in (6, 7, 8):
raise BatchException(
f"You must specify 'image' for Python jobs if you are using a Python version other than 3.6, 3.7, or 3.8 (you are using {version})")
job._image = f'hailgenetics/python-dill:{version.major}.{version.minor}-slim'
job._compile(local_tmpdir, remote_tmpdir)
inputs = [x for r in job._inputs for x in copy_input(r)]
outputs = [x for r in job._internal_outputs for x in copy_internal_output(r)]
if outputs:
used_remote_tmpdir = True
outputs += [x for r in job._external_outputs for x in copy_external_output(r)]
symlinks = [x for r in job._mentioned for x in symlink_input_resource_group(r)]
env_vars = {
**job._env,
**{r._uid: r._get_path(local_tmpdir) for r in job._mentioned}}
if job._image is None:
if verbose:
print(f"Using image '{default_image}' since no image was specified.")
make_local_tmpdir = f'mkdir -p {local_tmpdir}/{job._job_id}'
job_command = [cmd.strip() for cmd in job._command]
prepared_job_command = (f'{{\n{x}\n}}' for x in job_command)
cmd = f'''
{bash_flags}
{make_local_tmpdir}
{"; ".join(symlinks)}
{" && ".join(prepared_job_command)}
'''
if dry_run:
commands.append(cmd)
continue
parents = [job_to_client_job_mapping[j] for j in job._dependencies]
attributes = copy.deepcopy(job.attributes) if job.attributes else dict()
if job.name:
attributes['name'] = job.name
resources: Dict[str, Any] = {}
if job._cpu:
resources['cpu'] = job._cpu
if job._memory:
resources['memory'] = job._memory
if job._storage:
resources['storage'] = job._storage
if job._machine_type:
resources['machine_type'] = job._machine_type
if job._preemptible is not None:
resources['preemptible'] = job._preemptible
image = job._image if job._image else default_image
image_ref = parse_docker_image_reference(image)
if not is_google_registry_domain(image_ref.domain) and image_ref.name() not in HAIL_GENETICS_IMAGES:
warnings.warn(f'Using an image {image} not in GCR. '
f'Jobs may fail due to Docker Hub rate limits.')
j = bc_batch.create_job(image=image,
command=[job._shell if job._shell else self._DEFAULT_SHELL, '-c', cmd],
parents=parents,
attributes=attributes,
resources=resources,
input_files=inputs if len(inputs) > 0 else None,
output_files=outputs if len(outputs) > 0 else None,
always_run=job._always_run,
timeout=job._timeout,
gcsfuse=job._gcsfuse if len(job._gcsfuse) > 0 else None,
env=env_vars,
requester_pays_project=batch.requester_pays_project,
mount_tokens=True)
n_jobs_submitted += 1
job_to_client_job_mapping[job] = j
jobs_to_command[j] = cmd
if dry_run:
print("\n\n".join(commands))
return None
if delete_scratch_on_exit and used_remote_tmpdir:
parents = list(jobs_to_command.keys())
rm_cmd = f'gsutil -m rm -r {remote_tmpdir}'
cmd = f'''
{bash_flags}
{activate_service_account}
{rm_cmd}
'''
j = bc_batch.create_job(
image='gcr.io/google.com/cloudsdktool/cloud-sdk:310.0.0-alpine',
command=['/bin/bash', '-c', cmd],
parents=parents,
attributes={'name': 'remove_tmpdir'},
always_run=True)
jobs_to_command[j] = cmd
n_jobs_submitted += 1
if verbose:
print(f'Built DAG with {n_jobs_submitted} jobs in {round(time.time() - build_dag_start, 3)} seconds.')
submit_batch_start = time.time()
bc_batch = bc_batch.submit(disable_progress_bar=disable_progress_bar)
jobs_to_command = {j.id: cmd for j, cmd in jobs_to_command.items()}
if verbose:
print(f'Submitted batch {bc_batch.id} with {n_jobs_submitted} jobs in {round(time.time() - submit_batch_start, 3)} seconds:')
for jid, cmd in jobs_to_command.items():
print(f'{jid}: {cmd}')
print('')
deploy_config = get_deploy_config()
url = deploy_config.url('batch', f'/batches/{bc_batch.id}')
print(f'Submitted batch {bc_batch.id}, see {url}')
if open:
webbrowser.open(url)
if wait:
print(f'Waiting for batch {bc_batch.id}...')
status = bc_batch.wait()
print(f'batch {bc_batch.id} complete: {status["state"]}')
return bc_batch
def test_parse_docker_image_reference():
x = parse_docker_image_reference('animage')
assert x.domain is None
assert x.path == 'animage'
assert x.tag is None
assert x.digest is None
assert x.name() == 'animage'
assert str(x) == 'animage'
x = parse_docker_image_reference('hailgenetics/animage')
assert x.domain == 'hailgenetics'
assert x.path == 'animage'
assert x.tag is None
assert x.digest is None
assert x.name() == 'hailgenetics/animage'
assert str(x) == 'hailgenetics/animage'
x = parse_docker_image_reference('localhost:5000/animage')
assert x.domain == 'localhost:5000'
assert x.path == 'animage'
assert x.tag is None
assert x.digest is None
assert x.name() == 'localhost:5000/animage'
assert str(x) == 'localhost:5000/animage'
x = parse_docker_image_reference('localhost:5000/a/b/name')
assert x.domain == 'localhost:5000'
assert x.path == 'a/b/name'
assert x.tag is None
assert x.digest is None
assert x.name() == 'localhost:5000/a/b/name'
assert str(x) == 'localhost:5000/a/b/name'
x = parse_docker_image_reference('localhost:5000/a/b/name:tag')
assert x.domain == 'localhost:5000'
assert x.path == 'a/b/name'
assert x.tag == 'tag'
assert x.digest is None
assert x.name() == 'localhost:5000/a/b/name'
assert str(x) == 'localhost:5000/a/b/name:tag'
x = parse_docker_image_reference(
'localhost:5000/a/b/name:tag@sha256:abc123')
assert x.domain == 'localhost:5000'
assert x.path == 'a/b/name'
assert x.tag == 'tag'
assert x.digest == 'sha256:abc123'
assert x.name() == 'localhost:5000/a/b/name'
assert str(x) == 'localhost:5000/a/b/name:tag@sha256:abc123'
x = parse_docker_image_reference('localhost:5000/a/b/name@sha256:abc123')
assert x.domain == 'localhost:5000'
assert x.path == 'a/b/name'
assert x.tag is None
assert x.digest == 'sha256:abc123'
assert x.name() == 'localhost:5000/a/b/name'
assert str(x) == 'localhost:5000/a/b/name@sha256:abc123'
x = parse_docker_image_reference('name@sha256:abc123')
assert x.domain is None
assert x.path == 'name'
assert x.tag is None
assert x.digest == 'sha256:abc123'
assert x.name() == 'name'
assert str(x) == 'name@sha256:abc123'
x = parse_docker_image_reference('gcr.io/hail-vdc/batch-worker:123fds312')
assert x.domain == 'gcr.io'
assert x.path == 'hail-vdc/batch-worker'
assert x.tag == '123fds312'
assert x.digest is None
assert x.name() == 'gcr.io/hail-vdc/batch-worker'
assert str(x) == 'gcr.io/hail-vdc/batch-worker:123fds312'
x = parse_docker_image_reference(
'us-docker.pkg.dev/my-project/my-repo/test-image')
assert x.domain == 'us-docker.pkg.dev'
assert x.path == 'my-project/my-repo/test-image'
assert x.tag is None
assert x.digest is None
assert x.name() == 'us-docker.pkg.dev/my-project/my-repo/test-image'
assert str(x) == 'us-docker.pkg.dev/my-project/my-repo/test-image'
async def _async_run(
self,
batch: 'batch.Batch',
dry_run: bool,
verbose: bool,
delete_scratch_on_exit: bool,
wait: bool = True,
open: bool = False,
disable_progress_bar: bool = False,
callback: Optional[str] = None,
token: Optional[str] = None,
**backend_kwargs): # pylint: disable-msg=too-many-statements
if backend_kwargs:
raise ValueError(
f'ServiceBackend does not support any of these keywords: {backend_kwargs}'
)
build_dag_start = time.time()
uid = uuid.uuid4().hex[:6]
batch_remote_tmpdir = f'{self.remote_tmpdir}{uid}'
local_tmpdir = f'/io/batch/{uid}'
default_image = 'ubuntu:20.04'
attributes = copy.deepcopy(batch.attributes)
if batch.name is not None:
attributes['name'] = batch.name
bc_batch = self._batch_client.create_batch(
attributes=attributes,
callback=callback,
token=token,
cancel_after_n_failures=batch._cancel_after_n_failures)
n_jobs_submitted = 0
used_remote_tmpdir = False
job_to_client_job_mapping: Dict[_job.Job, bc.Job] = {}
jobs_to_command = {}
commands = []
bash_flags = 'set -e' + ('x' if verbose else '')
def copy_input(r):
if isinstance(r, resource.InputResourceFile):
return [(r._input_path, r._get_path(local_tmpdir))]
assert isinstance(
r, (resource.JobResourceFile, resource.PythonResult))
return [(r._get_path(batch_remote_tmpdir),
r._get_path(local_tmpdir))]
def copy_internal_output(r):
assert isinstance(
r, (resource.JobResourceFile, resource.PythonResult))
return [(r._get_path(local_tmpdir),
r._get_path(batch_remote_tmpdir))]
def copy_external_output(r):
if isinstance(r, resource.InputResourceFile):
return [(r._input_path, dest) for dest in r._output_paths]
assert isinstance(
r, (resource.JobResourceFile, resource.PythonResult))
return [(r._get_path(local_tmpdir), dest)
for dest in r._output_paths]
def symlink_input_resource_group(r):
symlinks = []
if isinstance(r, resource.ResourceGroup) and r._source is None:
for name, irf in r._resources.items():
src = irf._get_path(local_tmpdir)
dest = f'{r._get_path(local_tmpdir)}.{name}'
symlinks.append(f'ln -sf {shq(src)} {shq(dest)}')
return symlinks
write_external_inputs = [
x for r in batch._input_resources for x in copy_external_output(r)
]
if write_external_inputs:
transfers_bytes = orjson.dumps([{
"from": src,
"to": dest
} for src, dest in write_external_inputs])
transfers = transfers_bytes.decode('utf-8')
write_cmd = [
'python3', '-m', 'hailtop.aiotools.copy', 'null', transfers
]
if dry_run:
commands.append(' '.join(shq(x) for x in write_cmd))
else:
j = bc_batch.create_job(
image=HAIL_GENETICS_HAIL_IMAGE,
command=write_cmd,
attributes={'name': 'write_external_inputs'})
jobs_to_command[j] = ' '.join(shq(x) for x in write_cmd)
n_jobs_submitted += 1
pyjobs = [j for j in batch._jobs if isinstance(j, _job.PythonJob)]
for job in pyjobs:
if job._image is None:
version = sys.version_info
if version.major != 3 or version.minor not in (6, 7, 8):
raise BatchException(
f"You must specify 'image' for Python jobs if you are using a Python version other than 3.6, 3.7, or 3.8 (you are using {version})"
)
job._image = f'hailgenetics/python-dill:{version.major}.{version.minor}-slim'
with tqdm(total=len(batch._jobs),
desc='upload code',
disable=disable_progress_bar) as pbar:
async def compile_job(job):
used_remote_tmpdir = await job._compile(local_tmpdir,
batch_remote_tmpdir,
dry_run=dry_run)
pbar.update(1)
return used_remote_tmpdir
used_remote_tmpdir_results = await bounded_gather(
*[functools.partial(compile_job, j) for j in batch._jobs],
parallelism=150)
used_remote_tmpdir |= any(used_remote_tmpdir_results)
for job in tqdm(batch._jobs,
desc='create job objects',
disable=disable_progress_bar):
inputs = [x for r in job._inputs for x in copy_input(r)]
outputs = [
x for r in job._internal_outputs
for x in copy_internal_output(r)
]
if outputs:
used_remote_tmpdir = True
outputs += [
x for r in job._external_outputs
for x in copy_external_output(r)
]
symlinks = [
x for r in job._mentioned
for x in symlink_input_resource_group(r)
]
if job._image is None:
if verbose:
print(
f"Using image '{default_image}' since no image was specified."
)
make_local_tmpdir = f'mkdir -p {local_tmpdir}/{job._dirname}'
job_command = [cmd.strip() for cmd in job._wrapper_code]
prepared_job_command = (f'{{\n{x}\n}}' for x in job_command)
cmd = f'''
{bash_flags}
{make_local_tmpdir}
{"; ".join(symlinks)}
{" && ".join(prepared_job_command)}
'''
user_code = '\n\n'.join(job._user_code) if job._user_code else None
if dry_run:
formatted_command = f'''
================================================================================
# Job {job._job_id} {f": {job.name}" if job.name else ''}
--------------------------------------------------------------------------------
## USER CODE
--------------------------------------------------------------------------------
{user_code}
--------------------------------------------------------------------------------
## COMMAND
--------------------------------------------------------------------------------
{cmd}
================================================================================
'''
commands.append(formatted_command)
continue
parents = [job_to_client_job_mapping[j] for j in job._dependencies]
attributes = copy.deepcopy(
job.attributes) if job.attributes else {}
if job.name:
attributes['name'] = job.name
resources: Dict[str, Any] = {}
if job._cpu:
resources['cpu'] = job._cpu
if job._memory:
resources['memory'] = job._memory
if job._storage:
resources['storage'] = job._storage
if job._machine_type:
resources['machine_type'] = job._machine_type
if job._preemptible is not None:
resources['preemptible'] = job._preemptible
image = job._image if job._image else default_image
image_ref = parse_docker_image_reference(image)
if image_ref.hosted_in('dockerhub') and image_ref.name(
) not in HAIL_GENETICS_IMAGES:
warnings.warn(f'Using an image {image} from Docker Hub. '
f'Jobs may fail due to Docker Hub rate limits.')
env = {**job._env, 'BATCH_TMPDIR': local_tmpdir}
j = bc_batch.create_job(
image=image,
command=[
job._shell if job._shell else DEFAULT_SHELL, '-c', cmd
],
parents=parents,
attributes=attributes,
resources=resources,
input_files=inputs if len(inputs) > 0 else None,
output_files=outputs if len(outputs) > 0 else None,
always_run=job._always_run,
timeout=job._timeout,
cloudfuse=job._cloudfuse if len(job._cloudfuse) > 0 else None,
env=env,
requester_pays_project=batch.requester_pays_project,
mount_tokens=True,
user_code=user_code)
n_jobs_submitted += 1
job_to_client_job_mapping[job] = j
jobs_to_command[j] = cmd
if dry_run:
print("\n\n".join(commands))
return None
if delete_scratch_on_exit and used_remote_tmpdir:
parents = list(jobs_to_command.keys())
j = bc_batch.create_job(image=HAIL_GENETICS_HAIL_IMAGE,
command=[
'python3', '-m',
'hailtop.aiotools.delete',
batch_remote_tmpdir
],
parents=parents,
attributes={'name': 'remove_tmpdir'},
always_run=True)
jobs_to_command[j] = cmd
n_jobs_submitted += 1
if verbose:
print(
f'Built DAG with {n_jobs_submitted} jobs in {round(time.time() - build_dag_start, 3)} seconds.'
)
submit_batch_start = time.time()
batch_handle = bc_batch.submit(
disable_progress_bar=disable_progress_bar)
jobs_to_command = {j.id: cmd for j, cmd in jobs_to_command.items()}
if verbose:
print(
f'Submitted batch {batch_handle.id} with {n_jobs_submitted} jobs in {round(time.time() - submit_batch_start, 3)} seconds:'
)
for jid, cmd in jobs_to_command.items():
print(f'{jid}: {cmd}')
print('')
deploy_config = get_deploy_config()
url = deploy_config.url('batch', f'/batches/{batch_handle.id}')
print(f'Submitted batch {batch_handle.id}, see {url}')
if open:
webbrowser.open(url)
if wait:
print(f'Waiting for batch {batch_handle.id}...')
status = batch_handle.wait()
print(f'batch {batch_handle.id} complete: {status["state"]}')
return batch_handle