def build_configs(self, timeout: float = 30.0):
try:
self.create_cluster_compute()
except AssertionError as e:
# If already exists, ignore
logger.warning(str(e))
except ClusterComputeCreateError as e:
raise e
except Exception as e:
raise ClusterComputeCreateError(
f"Unexpected cluster compute build error: {e}") from e
try:
self.create_cluster_env()
except AssertionError as e:
# If already exists, ignore
logger.warning(str(e))
except ClusterEnvCreateError as e:
raise e
except Exception as e:
raise ClusterEnvCreateError(
f"Unexpected cluster env create error: {e}") from e
try:
self.build_cluster_env(timeout=timeout)
except AssertionError as e:
# If already exists, ignore
logger.warning(str(e))
except (ClusterEnvBuildError, ClusterEnvBuildTimeout) as e:
raise e
except Exception as e:
raise ClusterEnvBuildError(
f"Unexpected cluster env build error: {e}") from e
def create_cluster_compute(self, _repeat: bool = True):
assert self.cluster_compute_id is None
if self.cluster_compute:
assert self.cluster_compute
logger.info(f"Tests uses compute template "
f"with name {self.cluster_compute_name}. "
f"Looking up existing cluster computes.")
paging_token = None
while not self.cluster_compute_id:
result = self.sdk.search_cluster_computes(
dict(
project_id=self.project_id,
name=dict(equals=self.cluster_compute_name),
include_anonymous=True,
paging=dict(token=paging_token),
))
paging_token = result.metadata.next_paging_token
for res in result.results:
if res.name == self.cluster_compute_name:
self.cluster_compute_id = res.id
logger.info(f"Cluster compute already exists "
f"with ID {self.cluster_compute_id}")
break
if not paging_token:
break
if not self.cluster_compute_id:
logger.info(f"Cluster compute not found. "
f"Creating with name {self.cluster_compute_name}.")
try:
result = self.sdk.create_cluster_compute(
dict(
name=self.cluster_compute_name,
project_id=self.project_id,
config=self.cluster_compute,
))
self.cluster_compute_id = result.result.id
except Exception as e:
if _repeat:
logger.warning(
f"Got exception when trying to create cluster "
f"compute: {e}. Sleeping for 10 seconds and then "
f"try again once...")
time.sleep(10)
return self.create_cluster_compute(_repeat=False)
raise ClusterComputeCreateError(
"Could not create cluster compute") from e
logger.info(f"Cluster compute template created with "
f"name {self.cluster_compute_name} and "
f"ID {self.cluster_compute_id}")