paasta-tools 1.15.1__tar.gz → 1.16.1__tar.gz

{paasta-tools-1.15.1/paasta_tools.egg-info → paasta-tools-1.16.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 1.2
 Name: paasta-tools
-Version: 1.15.1
+Version: 1.16.1
 Summary: Tools for Yelps SOA infrastructure
 Home-page: UNKNOWN
 Author: Compute Infrastructure @ Yelp

{paasta-tools-1.15.1 → paasta-tools-1.16.1}/paasta_tools/__init__.py

@@ -17,4 +17,4 @@
 # setup phase, the dependencies may not exist on disk yet.
 #
 # Don't bump version manually. See `make release` docs in ./Makefile
-__version__ = "1.15.1"
+__version__ = "1.16.1"

{paasta-tools-1.15.1 → paasta-tools-1.16.1}/paasta_tools/api/views/remote_run.py

@@ -17,20 +17,18 @@ from pyramid.view import view_config
 
 from paasta_tools.api import settings
 from paasta_tools.api.views.exception import ApiFailure
+from paasta_tools.kubernetes.remote_run import get_max_job_duration_limit
 from paasta_tools.kubernetes.remote_run import remote_run_ready
 from paasta_tools.kubernetes.remote_run import remote_run_start
 from paasta_tools.kubernetes.remote_run import remote_run_stop
 from paasta_tools.kubernetes.remote_run import remote_run_token
-from paasta_tools.utils import load_system_paasta_config
 
 
 DEFAULT_MAX_DURATION = 60 * 60  # 1 hour
-DEFAULT_MAX_DURATION_LIMIT = 8 * 60 * 60  # 8 hours
 
 
 @view_config(route_name="remote_run.start", request_method="POST", renderer="json")
 def view_remote_run_start(request):
-    system_config = load_system_paasta_config()
     service = request.swagger_data["service"]
     instance = request.swagger_data["instance"]
     user = request.swagger_data["json_body"]["user"]
@@ -38,7 +36,7 @@ def view_remote_run_start(request):
     recreate = request.swagger_data["json_body"].get("recreate", False)
     max_duration = min(
         request.swagger_data["json_body"].get("max_duration", DEFAULT_MAX_DURATION),
-        system_config.get_remote_run_duration_limit(DEFAULT_MAX_DURATION_LIMIT),
+        get_max_job_duration_limit(),
     )
     try:
         return remote_run_start(

{paasta-tools-1.15.1 → paasta-tools-1.16.1}/paasta_tools/kubernetes/bin/paasta_cleanup_remote_run_resources.py

@@ -22,6 +22,8 @@ from typing import Callable
 from typing import Sequence
 from typing import Tuple
 
+from paasta_tools.kubernetes.remote_run import get_max_job_duration_limit
+from paasta_tools.kubernetes.remote_run import get_remote_run_jobs
 from paasta_tools.kubernetes.remote_run import get_remote_run_role_bindings
 from paasta_tools.kubernetes.remote_run import get_remote_run_roles
 from paasta_tools.kubernetes.remote_run import get_remote_run_service_accounts
@@ -35,24 +37,44 @@ log = logging.getLogger(__name__)
 
 
 def clean_namespace(
-    kube_client: KubeClient, namespace: str, age_limit: datetime, dry_run: bool = False
+    kube_client: KubeClient,
+    namespace: str,
+    auth_age_limit: datetime,
+    job_age_limit: datetime,
+    dry_run: bool = False,
 ):
     """Clean ephemeral remote-run resource in a namespace
 
     :param KubeClient kube_client: kubernetes client
     :param str namepsace: kubernetes namespace
-    :param datetime age_limit: expiration time for resources
+    :param datetime auth_age_limit: expiration time for authentication resources
+    :param datetime job_age_limit: expiration time for job resources
+    :param bool dry_run: delete resources for real or not
     """
     dry_run_msg = " (dry_run)" if dry_run else ""
-    cleanup_actions: Sequence[Tuple[DeletionFuncType, ListingFuncType]] = (
+    cleanup_actions: Sequence[Tuple[DeletionFuncType, ListingFuncType, datetime]] = (
         (
             kube_client.core.delete_namespaced_service_account,
             get_remote_run_service_accounts,
+            auth_age_limit,
+        ),
+        (
+            kube_client.rbac.delete_namespaced_role,
+            get_remote_run_roles,
+            auth_age_limit,
+        ),
+        (
+            kube_client.rbac.delete_namespaced_role_binding,
+            get_remote_run_role_bindings,
+            auth_age_limit,
+        ),
+        (
+            kube_client.batches.delete_namespaced_job,
+            get_remote_run_jobs,
+            job_age_limit,
         ),
-        (kube_client.rbac.delete_namespaced_role, get_remote_run_roles),
-        (kube_client.rbac.delete_namespaced_role_binding, get_remote_run_role_bindings),
     )
-    for delete_func, list_func in cleanup_actions:
+    for delete_func, list_func, age_limit in cleanup_actions:
         for entity in list_func(kube_client, namespace):
             if (
                 not entity.metadata.name.startswith("remote-run-")
@@ -96,9 +118,17 @@ def main():
     args = parse_args()
     logging.basicConfig(level=logging.DEBUG if args.verbose else logging.INFO)
     kube_client = KubeClient()
-    age_limit = datetime.now(tz=timezone.utc) - timedelta(seconds=args.max_age)
+    now = datetime.now(tz=timezone.utc)
+    age_limit = now - timedelta(seconds=args.max_age)
+    job_age_limit = now - timedelta(seconds=get_max_job_duration_limit())
     for namespace in get_all_managed_namespaces(kube_client):
-        clean_namespace(kube_client, namespace, age_limit, args.dry_run)
+        clean_namespace(
+            kube_client,
+            namespace,
+            auth_age_limit=age_limit,
+            job_age_limit=job_age_limit,
+            dry_run=args.dry_run,
+        )
 
 
 if __name__ == "__main__":

{paasta-tools-1.15.1 → paasta-tools-1.16.1}/paasta_tools/kubernetes/remote_run.py

@@ -41,11 +41,13 @@ from paasta_tools.kubernetes_tools import JOB_TYPE_LABEL_NAME
 from paasta_tools.kubernetes_tools import KubeClient
 from paasta_tools.kubernetes_tools import limit_size_with_hash
 from paasta_tools.kubernetes_tools import paasta_prefixed
+from paasta_tools.utils import load_system_paasta_config
 
 
 logger = logging.getLogger(__name__)
 REMOTE_RUN_JOB_LABEL = "remote-run"
 POD_OWNER_LABEL = paasta_prefixed("pod_owner")
+DEFAULT_MAX_DURATION_LIMIT = 8 * 60 * 60  # 8 hours
 
 
 class RemoteRunError(Exception):
@@ -93,7 +95,7 @@ def remote_run_start(
     :param int max_duration: maximum allowed duration for the remote-ruh job
     :return: outcome of the operation, and resulting Kubernetes pod information
     """
-    kube_client = KubeClient(config_file="/etc/kubernetes/admin.conf")
+    kube_client = KubeClient()
 
     # Load the service deployment settings
     deployment_config = load_eks_service_config(service, instance, cluster)
@@ -149,7 +151,7 @@ def remote_run_ready(
     :param str job_name: name of the remote-run job to check
     :return: job status, with pod info
     """
-    kube_client = KubeClient(config_file="/etc/kubernetes/admin.conf")
+    kube_client = KubeClient()
 
     # Load the service deployment settings
     deployment_config = load_eks_service_config(service, instance, cluster)
@@ -182,7 +184,7 @@ def remote_run_stop(
     :param str user: the user requesting the remote-run sandbox
     :return: outcome of the operation
     """
-    kube_client = KubeClient(config_file="/etc/kubernetes/admin.conf")
+    kube_client = KubeClient()
 
     # Load the service deployment settings
     deployment_config = load_eks_service_config(service, instance, cluster)
@@ -216,7 +218,7 @@ def remote_run_token(
     :param str cluster: paasta cluster
     :param str user: the user requesting the remote-run sandbox
     """
-    kube_client = KubeClient(config_file="/etc/kubernetes/admin.conf")
+    kube_client = KubeClient()
 
     # Load the service deployment settings
     deployment_config = load_eks_service_config(service, instance, cluster)
@@ -419,11 +421,12 @@ def get_remote_run_roles(kube_client: KubeClient, namespace: str) -> List[V1Role
     """List all temporary roles related to remote-run
 
     :param KubeClient kube_client: Kubernetes client
-    :param str namespace: pod namespace
+    :param str namespace: role namespace
     :return: list of roles
     """
     return kube_client.rbac.list_namespaced_role(
-        namespace, label_selector=POD_OWNER_LABEL
+        namespace,
+        label_selector=POD_OWNER_LABEL,
     ).items
 
 
@@ -433,9 +436,31 @@ def get_remote_run_role_bindings(
     """List all temporary role bindings related to remote-run
 
     :param KubeClient kube_client: Kubernetes client
-    :param str namespace: pod namespace
+    :param str namespace: role namespace
     :return: list of roles
     """
     return kube_client.rbac.list_namespaced_role_binding(
-        namespace, label_selector=POD_OWNER_LABEL
+        namespace,
+        label_selector=POD_OWNER_LABEL,
+    ).items
+
+
+def get_remote_run_jobs(kube_client: KubeClient, namespace: str) -> List[V1Job]:
+    """List all remote-run jobs
+
+    :param KubeClient kube_client: Kubernetes client
+    :param str namespace: job namespace
+    """
+    return kube_client.batches.list_namespaced_job(
+        namespace,
+        label_selector=f"{paasta_prefixed(JOB_TYPE_LABEL_NAME)}={REMOTE_RUN_JOB_LABEL}",
     ).items
+
+
+def get_max_job_duration_limit() -> int:
+    """Get maximum configured duration for a remote run job
+
+    :return: max duration in seconds
+    """
+    system_config = load_system_paasta_config()
+    return system_config.get_remote_run_duration_limit(DEFAULT_MAX_DURATION_LIMIT)

{paasta-tools-1.15.1 → paasta-tools-1.16.1}/paasta_tools/kubernetes_tools.py

@@ -2088,6 +2088,7 @@ class KubernetesDeploymentConfig(LongRunningServiceConfig):
                 metadata=self.get_kubernetes_metadata(git_sha),
                 spec=V1JobSpec(
                     active_deadline_seconds=deadline_seconds,
+                    ttl_seconds_after_finished=0,  # remove job resource after completion
                     template=pod_template,
                 ),
             )

{paasta-tools-1.15.1 → paasta-tools-1.16.1}/paasta_tools/tron_tools.py

@@ -697,13 +697,14 @@ class TronActionConfig(InstanceConfig):
         override this value. To control this, we have an optional config item that we'll puppet onto Tron masters
         which this function will read.
         """
-        return (
-            self.config_dict.get(
+        if self.get_executor() == "spark":
+            pool = load_system_paasta_config().get_default_spark_driver_pool_override()
+        else:
+            pool = self.config_dict.get(
                 "pool", load_system_paasta_config().get_tron_default_pool_override()
             )
-            if not self.get_executor() == "spark"
-            else spark_tools.SPARK_DRIVER_POOL
-        )
+
+        return pool
 
     def get_spark_executor_pool(self) -> str:
         return self.config_dict.get("pool", DEFAULT_SPARK_EXECUTOR_POOL)

{paasta-tools-1.15.1 → paasta-tools-1.16.1}/paasta_tools/utils.py

@@ -123,6 +123,10 @@ DEFAULT_CPU_BURST_ADD = 1
 
 DEFAULT_SOA_CONFIGS_GIT_URL = "sysgit.yelpcorp.com"
 
+# To ensure the Spark driver not being interrupted due to spot instances,
+# we use stable pool for drivers
+DEFAULT_SPARK_DRIVER_POOL = "stable"
+
 log = logging.getLogger(__name__)
 log.addHandler(logging.NullHandler())
 
@@ -2045,6 +2049,7 @@ class SystemPaastaConfigDict(TypedDict, total=False):
     enable_automated_redeploys_default: bool
     enable_tron_tsc: bool
     default_spark_iam_user: str
+    default_spark_driver_pool_override: str
 
 
 def load_system_paasta_config(
@@ -2148,6 +2153,17 @@ class SystemPaastaConfig:
             "default_spark_iam_user", "/etc/boto_cfg/mrjob.yaml"
         )
 
+    def get_default_spark_driver_pool_override(self) -> str:
+        """
+        If defined, fetches the override for what pool to run a Spark driver in.
+        Otherwise, returns the default Spark driver pool.
+
+        :returns: The default_spark_driver_pool_override specified in the paasta configuration
+        """
+        return self.config_dict.get(
+            "default_spark_driver_pool_override", DEFAULT_SPARK_DRIVER_POOL
+        )
+
     def get_sidecar_requirements_config(
         self,
     ) -> Dict[str, KubeContainerResourceRequest]:

{paasta-tools-1.15.1 → paasta-tools-1.16.1/paasta_tools.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 1.2
 Name: paasta-tools
-Version: 1.15.1
+Version: 1.16.1
 Summary: Tools for Yelps SOA infrastructure
 Home-page: UNKNOWN
 Author: Compute Infrastructure @ Yelp