def test_deploy_pod_fails(self, unused_mock_cluster_exists,
unused_mock_get_pod_status, unused_mock_call,
unused_mock_sleep):
with self.assertRaises(RuntimeError):
gke_cluster.GkeCluster('foo-cluster',
cluster_zone='foo-zone').deploy_pod(
pod_config='foo-config',
pod_name='foo-pod')
def test_delete_cluster_in_provisioning_state(self, mock_call,
unused_mock_sleep):
mock_call.side_effect = [
'foo-cluster', None, 'RUNNING', 'foo-cluster', 'PROVISIONING',
'RUNNING', 'RUNNING'
]
gke_cluster.GkeCluster(
'foo-cluster', cluster_zone='foo-zone').delete_cluster(wait=False)
self.assertEqual(mock_call.call_count, 7)
def test_create_new_cluster(self, mock_call):
gke_cluster.GkeCluster('foo-cluster',
cluster_zone='foo-zone',
alpha_cluster=True)
mock_call.assert_called_with([
'gcloud', 'alpha', 'container', 'clusters', 'create',
'foo-cluster', '--zone', 'foo-zone'
],
retry_delay_sec=1,
retries=1)
def test_delete_pod(self, unused_mock_cluster_exists,
unused_mock_pod_exists, mock_call):
gke_cluster.GkeCluster('foo-cluster',
cluster_zone='foo-zone').delete_pod(
pod_name='foo-pod', wait=True)
mock_call.assert_any_call(
['kubectl', 'delete', 'pod', 'foo-pod', '--wait'],
std_input=None,
retries=1,
retry_delay_sec=0)
def test_deploy_pod(self, unused_mock_cluster_exists,
unused_mock_get_pod_status, mock_call,
unused_mock_sleep):
gke_cluster.GkeCluster('foo-cluster',
cluster_zone='foo-zone').deploy_pod(
pod_config='foo-config', pod_name='foo-pod')
mock_call.assert_any_call(['kubectl', 'create', '-f', '-'],
std_input='foo-config',
retries=1,
retry_delay_sec=1)
def test_get_pod_status(self, unused_mock_cluster_exists, mock_call):
self.assertEqual(
gke_cluster.GkeCluster(
'foo-cluster',
cluster_zone='foo-zone').get_pod_status(pod_name='foo-pod'),
gke_cluster.PodStatus.SUCCEEDED)
mock_call.assert_any_call([
'kubectl', 'get', 'pods', 'foo-pod', '-o',
'jsonpath={.status.phase}'
],
std_input=None,
retries=0,
retry_delay_sec=1)
def _run_call_variants_with_kubernetes(pipeline_args):
"""Runs call_variants step with kubernetes."""
# Setup Kubernetes cluster.
if pipeline_args.gke_cluster_name:
# Reuse provided GKE cluster.
new_cluster_created = False
cluster = gke_cluster.GkeCluster(pipeline_args.gke_cluster_name,
pipeline_args.gke_cluster_region,
pipeline_args.gke_cluster_zone)
else:
# Create a new GKE cluster.
job_name_label = pipeline_args.job_name_prefix + _CALL_VARIANTS_JOB_NAME
extra_args = [
'--cluster-version=' + pipeline_args.gke_cluster_version,
'--num-nodes=1', '--enable-kubernetes-alpha', '--enable-ip-alias',
'--create-subnetwork=', '--node-labels=job_name=' + job_name_label,
'--scopes=cloud-platform', '--enable-tpu',
'--no-enable-autorepair', '--project', pipeline_args.project,
'--quiet'
]
cluster_name = 'deepvariant-' + _NOW_STR + uuid.uuid4().hex[:5]
cluster = gke_cluster.GkeCluster(cluster_name,
pipeline_args.gke_cluster_region,
pipeline_args.gke_cluster_zone,
alpha_cluster=True,
extra_args=extra_args)
new_cluster_created = True
# Deploy call_variants pod.
pod_name = 'deepvariant-' + _NOW_STR + '-' + uuid.uuid4().hex[:5]
try:
_deploy_call_variants_pod(pod_name, cluster, pipeline_args)
except KeyboardInterrupt:
cluster.delete_pod(pod_name)
raise RuntimeError('Job cancelled by user.')
finally:
if new_cluster_created:
cluster.delete_cluster(wait=False)
def test_deploy_pod_retries(self, unused_mock_cluster_exists,
unused_mock_delete_pod, mock_get_pod_status,
mock_call, unused_mock_sleep):
mock_get_pod_status.side_effect = [
gke_cluster.PodStatus.FAILED, gke_cluster.PodStatus.SUCCEEDED
]
gke_cluster.GkeCluster('foo-cluster',
cluster_zone='foo-zone').deploy_pod(
pod_config='foo-config',
pod_name='foo-pod',
retries=1)
mock_call.assert_any_call(['kubectl', 'create', '-f', '-'],
std_input='foo-config',
retries=1,
retry_delay_sec=1)
mock_call.assert_any_call(['kubectl', 'replace', '--force', '-f', '-'],
std_input='foo-config',
retries=1,
retry_delay_sec=1)
def test_reuse_existing_cluster(self, mock_call):
gke_cluster.GkeCluster('foo-cluster', cluster_zone='foo-zone')
mock_call.assert_any_call([
'gcloud', 'container', 'clusters', 'list', '--format=value(name)',
'--zone', 'foo-zone'
],
retry_delay_sec=1,
retries=1)
mock_call.assert_any_call([
'gcloud', 'container', 'clusters', 'get-credentials',
'foo-cluster', '--zone', 'foo-zone'
],
retry_delay_sec=1,
retries=1)
mock_call.assert_any_call([
'gcloud', 'container', 'clusters', 'describe', 'foo-cluster',
'--format=value(status)', '--zone', 'foo-zone'
],
retry_delay_sec=1,
retries=1)
def test_create_new_cluster_with_keyboard_interrupt(
self, unused_mock_cluster_exists, mock_call):
gke_cluster.GkeCluster('foo-cluster',
cluster_zone='foo-zone',
alpha_cluster=True)
mock_call.assert_any_call([
'gcloud', 'alpha', 'container', 'clusters', 'create',
'foo-cluster', '--zone', 'foo-zone'
],
retry_delay_sec=1,
retries=1)
mock_call.assert_any_call([
'gcloud', 'container', 'clusters', 'describe', 'foo-cluster',
'--format=value(status)', '--zone', 'foo-zone'
],
retry_delay_sec=1,
retries=1)
mock_call.assert_any_call([
'gcloud', 'container', 'clusters', 'describe', 'foo-cluster',
'--format=value(status)', '--zone', 'foo-zone'
],
retry_delay_sec=1,
retries=1)
def test_get_pod_status_pulling_image_failed(self,
unused_mock_cluster_exists,
mock_call):
self.assertEqual(
gke_cluster.GkeCluster(
'foo-cluster',
cluster_zone='foo-zone').get_pod_status(pod_name='foo-pod'),
gke_cluster.PodStatus.FAILED)
mock_call.assert_any_call([
'kubectl', 'get', 'pods', 'foo-pod', '-o',
'jsonpath={.status.phase}'
],
std_input=None,
retries=0,
retry_delay_sec=1)
mock_call.assert_any_call([
'kubectl', 'get', 'pods', 'foo-pod', '-o',
'go-template="{{range .status.containerStatuses}}{{if not .ready}}'
'{{.state.waiting.reason}}{{end}}{{end}}"'
],
std_input=None,
retries=0,
retry_delay_sec=1)
def test_get_cluster_status(self, mock_call):
self.assertEqual(
gke_cluster.GkeCluster(
'foo-cluster', cluster_zone='foo-zone')._get_cluster_status(),
gke_cluster.ClusterStatus.RUNNING)
mock_call.assert_any_call([
'gcloud', 'container', 'clusters', 'list', '--format=value(name)',
'--zone', 'foo-zone'
],
retry_delay_sec=1,
retries=1)
mock_call.assert_any_call([
'gcloud', 'container', 'clusters', 'get-credentials',
'foo-cluster', '--zone', 'foo-zone'
],
retry_delay_sec=1,
retries=1)
mock_call.assert_any_call([
'gcloud', 'container', 'clusters', 'describe', 'foo-cluster',
'--format=value(status)', '--zone', 'foo-zone'
],
retry_delay_sec=1,
retries=1)
def _validate_and_complete_args(pipeline_args):
"""Validates pipeline arguments and fills some missing args (if any)."""
# Basic validation logic. More detailed validation is done by pipelines API.
if pipeline_args.preemptible and pipeline_args.max_preemptible_tries <= 0:
raise ValueError('--max_preemptible_tries must be greater than zero.')
if pipeline_args.max_non_preemptible_tries <= 0:
raise ValueError(
'--max_non_preemptible_tries must be greater than zero.')
if pipeline_args.make_examples_workers <= 0:
raise ValueError('--make_examples_workers must be greater than zero.')
if pipeline_args.call_variants_workers <= 0:
raise ValueError('--call_variants_workers must be greater than zero.')
if pipeline_args.shards <= 0:
raise ValueError('--shards must be greater than zero.')
if pipeline_args.shards % pipeline_args.make_examples_workers != 0:
raise ValueError(
'--shards must be divisible by --make_examples_workers')
if pipeline_args.shards % pipeline_args.call_variants_workers != 0:
raise ValueError(
'--shards must be divisible by --call_variants_workers')
if pipeline_args.call_variants_workers > pipeline_args.make_examples_workers:
logging.warning(
'--call_variants_workers cannot be greather than '
'--make_examples_workers. Setting call_variants_workers to %d',
pipeline_args.make_examples_workers)
pipeline_args.call_variants_workers = pipeline_args.make_examples_workers
if pipeline_args.gpu and not pipeline_args.docker_image_gpu:
raise ValueError('--docker_image_gpu must be provided with --gpu')
if (pipeline_args.call_variants_cores_per_worker <
pipeline_args.call_variants_cores_per_shard):
raise ValueError('--call_variants_cores_per_worker must be at least '
'as large as --call_variants_cores_per_shard')
if (pipeline_args.gvcf_gq_binsize is not None
and not pipeline_args.gvcf_outfile):
raise ValueError(
'--gvcf_outfile must be provided with --gvcf_gq_binsize')
if (pipeline_args.gvcf_gq_binsize is not None
and pipeline_args.gvcf_gq_binsize < 1):
raise ValueError('--gvcf_gq_binsize must be greater or equal to 1')
if pipeline_args.gpu and pipeline_args.tpu:
raise ValueError('Both --gpu and --tpu cannot be set.')
# TODO(nmousavi): Support multiple TPUs for call_variants if there is an
# interest.
if pipeline_args.tpu and pipeline_args.call_variants_workers != 1:
raise ValueError(
'--call_variants_workers must be equal to one when --tpu is set.')
if pipeline_args.tpu and bool(pipeline_args.gke_cluster_region) == bool(
pipeline_args.gke_cluster_zone):
raise ValueError(
'Exactly one of --gke_cluster_region or '
'--gke_cluster_zone must be specified if --tpu is set.')
# Verify the existing gke cluster is up and running.
if pipeline_args.gke_cluster_name:
try:
_ = gke_cluster.GkeCluster(pipeline_args.gke_cluster_name,
pipeline_args.gke_cluster_region,
pipeline_args.gke_cluster_zone,
create_if_not_exist=False)
except ValueError:
raise ValueError('Given --gke_cluster_name does not exist: %s' %
pipeline_args.gke_cluster_name)
# Automatically generate default values for missing args (if any).
if not pipeline_args.logging:
pipeline_args.logging = os.path.join(pipeline_args.staging, 'logs')
if not pipeline_args.ref_fai:
pipeline_args.ref_fai = pipeline_args.ref + _FAI_FILE_SUFFIX
if not pipeline_args.ref_gzi and pipeline_args.ref.endswith(
_GZ_FILE_SUFFIX):
pipeline_args.ref_gzi = pipeline_args.ref + _GZI_FILE_SUFFIX
if not pipeline_args.bai:
pipeline_args.bai = pipeline_args.bam + _BAI_FILE_SUFFIX
# Ensuring all input files exist...
if not _gcs_object_exist(pipeline_args.ref):
raise ValueError('Given reference file via --ref does not exist')
if not _gcs_object_exist(pipeline_args.ref_fai):
raise ValueError('Given FAI index file via --ref_fai does not exist')
if (pipeline_args.ref_gzi
and not _gcs_object_exist(pipeline_args.ref_gzi)):
raise ValueError('Given GZI index file via --ref_gzi does not exist')
if not _gcs_object_exist(pipeline_args.bam):
raise ValueError('Given BAM file via --bam does not exist')
if not _gcs_object_exist(pipeline_args.bai):
raise ValueError('Given BAM index file via --bai does not exist')
# ...and we can write to output buckets.
if not _can_write_to_bucket(_get_gcs_bucket(pipeline_args.staging)):
raise ValueError(
'Cannot write to staging bucket, change --staging value')
if not _can_write_to_bucket(_get_gcs_bucket(pipeline_args.outfile)):
raise ValueError(
'Cannot write to output bucket, change --outfile value')
def test_cluster_exists_for_non_existent_cluster(
self, unused_mock_call, unused_unused_mock_create_cluster):
self.assertEqual(
gke_cluster.GkeCluster('foo-cluster',
cluster_zone='foo-zone')._cluster_exists(),
False)
def test_cluster_exists(self, unused_mock_call,
unused_mock_create_cluster):
self.assertEqual(
gke_cluster.GkeCluster('foo-cluster',
cluster_zone='foo-zone')._cluster_exists(),
True)
def test_get_cluster_unknown_status(self, unused_mock_call):
self.assertEqual(
gke_cluster.GkeCluster(
'foo-cluster', cluster_zone='foo-zone')._get_cluster_status(),
gke_cluster.ClusterStatus.UNKNOWN)
