paasta-tools 1.4.6__tar.gz → 1.4.7__tar.gz

{paasta-tools-1.4.6/paasta_tools.egg-info → paasta-tools-1.4.7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 1.2
 Name: paasta-tools
-Version: 1.4.6
+Version: 1.4.7
 Summary: Tools for Yelps SOA infrastructure
 Home-page: UNKNOWN
 Author: Compute Infrastructure @ Yelp

{paasta-tools-1.4.6 → paasta-tools-1.4.7}/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__ = "1.4.6"
+__version__ = "1.4.7"

{paasta-tools-1.4.6 → paasta-tools-1.4.7}/paasta_tools/cli/cmds/logs.py

@@ -40,7 +40,9 @@ from typing import Tuple
 from typing import Type
 from typing import Union
 
+import a_sync
 import isodate
+import nats
 import pytz
 from dateutil import tz
 
@@ -1166,19 +1168,26 @@ class ScribeLogReader(LogReader):
 
 @register_log_reader("vector-logs")
 class VectorLogsReader(LogReader):
+    SUPPORTS_TAILING = True
     SUPPORTS_TIME = True
 
-    def __init__(self, cluster_map: Mapping[str, Any]) -> None:
+    def __init__(
+        self, cluster_map: Mapping[str, Any], nats_endpoint_map: Mapping[str, Any]
+    ) -> None:
         super().__init__()
 
         if S3LogsReader is None:
             raise Exception("yelp_clog package must be available to use S3LogsReader")
 
         self.cluster_map = cluster_map
+        self.nats_endpoint_map = nats_endpoint_map
 
     def get_superregion_for_cluster(self, cluster: str) -> Optional[str]:
         return self.cluster_map.get(cluster, None)
 
+    def get_nats_endpoint_for_cluster(self, cluster: str) -> Optional[str]:
+        return self.nats_endpoint_map.get(cluster, None)
+
     def print_logs_by_time(
         self,
         service,
@@ -1230,6 +1239,48 @@ class VectorLogsReader(LogReader):
         for line in aggregated_logs:
             print_log(line["raw_line"], levels, raw_mode, strip_headers)
 
+    def tail_logs(
+        self,
+        service: str,
+        levels: Sequence[str],
+        components: Iterable[str],
+        clusters: Sequence[str],
+        instances: List[str],
+        pods: Iterable[str] = None,
+        raw_mode: bool = False,
+        strip_headers: bool = False,
+    ) -> None:
+        stream_name = get_log_name_for_service(service, prefix="app_output")
+        endpoint = self.get_nats_endpoint_for_cluster(clusters[0])
+        if not endpoint:
+            raise NotImplementedError(
+                "Tailing logs is not supported in this cluster yet, sorry"
+            )
+
+        async def tail_logs_from_nats() -> None:
+            nc = await nats.connect(f"nats://{endpoint}")
+            sub = await nc.subscribe(stream_name)
+
+            while True:
+                # Wait indefinitely for a new message (no timeout)
+                msg = await sub.next_msg(timeout=None)
+                decoded_data = msg.data.decode("utf-8")
+
+                if paasta_log_line_passes_filter(
+                    decoded_data,
+                    levels,
+                    service,
+                    components,
+                    clusters,
+                    instances,
+                    pods,
+                ):
+                    await a_sync.run(
+                        print_log, decoded_data, levels, raw_mode, strip_headers
+                    )
+
+        a_sync.block(tail_logs_from_nats)
+
 
 def scribe_env_to_locations(scribe_env) -> Mapping[str, Any]:
     """Converts a scribe environment to a dictionary of locations. The

{paasta-tools-1.4.6 → paasta-tools-1.4.7}/paasta_tools/cli/cmds/status.py

@@ -134,7 +134,11 @@ InstanceStatusWriter = Callable[
     int,
 ]
 
-EKS_DEPLOYMENT_CONFIGS = [EksDeploymentConfig, FlinkEksDeploymentConfig]
+EKS_DEPLOYMENT_CONFIGS = [
+    EksDeploymentConfig,
+    FlinkEksDeploymentConfig,
+    VitessDeploymentConfig,
+]
 FLINK_DEPLOYMENT_CONFIGS = [FlinkDeploymentConfig, FlinkEksDeploymentConfig]
 
 

{paasta-tools-1.4.6 → paasta-tools-1.4.7}/paasta_tools/kubernetes/application/controller_wrappers.py

@@ -15,6 +15,7 @@ from paasta_tools.eks_tools import load_eks_service_config_no_cache
 from paasta_tools.kubernetes_tools import create_deployment
 from paasta_tools.kubernetes_tools import create_pod_disruption_budget
 from paasta_tools.kubernetes_tools import create_stateful_set
+from paasta_tools.kubernetes_tools import ensure_service_account
 from paasta_tools.kubernetes_tools import KubeClient
 from paasta_tools.kubernetes_tools import KubeDeployment
 from paasta_tools.kubernetes_tools import KubernetesDeploymentConfig
@@ -120,6 +121,25 @@ class Application(ABC):
         """
         self.ensure_pod_disruption_budget(kube_client, self.soa_config.get_namespace())
 
+    def update_dependency_api_objects(self, kube_client: KubeClient) -> None:
+        """
+        Update related Kubernetes API objects that should be updated before the main object,
+        such as service accounts.
+        :param kube_client:
+        """
+        self.ensure_service_account(kube_client)
+
+    def ensure_service_account(self, kube_client: KubeClient) -> None:
+        """
+        Ensure that the service account for this application exists
+        :param kube_client:
+        """
+        ensure_service_account(
+            iam_role=self.soa_config.get_iam_role(),
+            namespace=self.soa_config.get_namespace(),
+            kube_client=kube_client,
+        )
+
     def delete_pod_disruption_budget(self, kube_client: KubeClient) -> None:
         try:
             kube_client.policy.delete_namespaced_pod_disruption_budget(

{paasta-tools-1.4.6 → paasta-tools-1.4.7}/paasta_tools/kubernetes_tools.py

@@ -2220,9 +2220,9 @@ class KubernetesDeploymentConfig(LongRunningServiceConfig):
             annotations["iam.amazonaws.com/role"] = ""
             iam_role = self.get_iam_role()
             if iam_role:
-                pod_spec_kwargs[
-                    "service_account_name"
-                ] = create_or_find_service_account_name(iam_role, self.get_namespace())
+                pod_spec_kwargs["service_account_name"] = get_service_account_name(
+                    iam_role
+                )
                 if fs_group is None:
                     # We need some reasoable default for group id of a process
                     # running inside the container. Seems like most of such
@@ -4050,12 +4050,9 @@ def get_all_limit_ranges(
 _RE_NORMALIZE_IAM_ROLE = re.compile(r"[^0-9a-zA-Z]+")
 
 
-def create_or_find_service_account_name(
+def get_service_account_name(
     iam_role: str,
-    namespace: str,
     k8s_role: Optional[str] = None,
-    kubeconfig_file: Optional[str] = None,
-    dry_run: bool = False,
 ) -> str:
     # the service account is expected to always be prefixed with paasta- as using the actual namespace
     # potentially wastes a lot of characters (e.g., paasta-nrtsearchservices) that could be used for
@@ -4081,12 +4078,17 @@ def create_or_find_service_account_name(
             "Expected at least one of iam_role or k8s_role to be passed in!"
         )
 
-    # if someone is dry-running paasta_setup_tron_namespace or some other tool that
-    # calls this function, we probably don't want to mutate k8s state :)
-    if dry_run:
-        return sa_name
+    return sa_name
+
+
+def ensure_service_account(
+    iam_role: str,
+    namespace: str,
+    kube_client: KubeClient,
+    k8s_role: Optional[str] = None,
+) -> None:
+    sa_name = get_service_account_name(iam_role, k8s_role)
 
-    kube_client = KubeClient(config_file=kubeconfig_file)
     if not any(
         sa.metadata and sa.metadata.name == sa_name
         for sa in get_all_service_accounts(kube_client, namespace)
@@ -4135,8 +4137,6 @@ def create_or_find_service_account_name(
                 namespace=namespace, body=role_binding
             )
 
-    return sa_name
-
 
 def mode_to_int(mode: Optional[Union[str, int]]) -> Optional[int]:
     if mode is not None:

{paasta-tools-1.4.6 → paasta-tools-1.4.7}/paasta_tools/setup_kubernetes_job.py

@@ -281,6 +281,7 @@ def setup_kube_deployments(
                 "paasta_namespace": app.kube_deployment.namespace,
             }
             try:
+                app.update_dependency_api_objects(kube_client)
                 if (
                     app.kube_deployment.service,
                     app.kube_deployment.instance,

{paasta-tools-1.4.6 → paasta-tools-1.4.7}/paasta_tools/setup_tron_namespace.py

@@ -26,8 +26,13 @@ import sys
 
 import ruamel.yaml as yaml
 
+from paasta_tools import spark_tools
 from paasta_tools import tron_tools
+from paasta_tools.kubernetes_tools import ensure_service_account
+from paasta_tools.kubernetes_tools import KubeClient
+from paasta_tools.tron_tools import KUBERNETES_NAMESPACE
 from paasta_tools.tron_tools import MASTER_NAMESPACE
+from paasta_tools.utils import load_system_paasta_config
 
 log = logging.getLogger(__name__)
 
@@ -63,6 +68,32 @@ def parse_args():
     return args
 
 
+def ensure_service_accounts(
+    raw_config: str, kube_client: KubeClient, spark_kube_client: KubeClient
+) -> None:
+    # this is kinda silly, but the tron create_config functions return strings
+    # we should refactor to pass the dicts around until the we're going to send the config to tron
+    # (where we can finally convert it to a string)
+    config = yaml.safe_load(raw_config)
+    for _, job in config.get("jobs", {}).items():
+        for _, action in job.get("actions", {}).items():
+            if action.get("service_account_name") is not None:
+                ensure_service_account(
+                    action["service_account_name"],
+                    namespace=KUBERNETES_NAMESPACE,
+                    kube_client=kube_client,
+                )
+                # spark executors are special in that we want the SA to exist in two namespaces:
+                # the tron namespace - for the spark driver
+                # and the spark namespace - for the spark executor
+                if action.get("executor") == "spark":
+                    ensure_service_account(
+                        action["service_account_name"],
+                        namespace=spark_tools.SPARK_EXECUTOR_NAMESPACE,
+                        kube_client=spark_kube_client,
+                    )
+
+
 def main():
     args = parse_args()
     log_level = logging.DEBUG if args.verbose else logging.INFO
@@ -133,6 +164,16 @@ def main():
                 log.info(f"{new_config}")
                 updated.append(service)
             else:
+                # NOTE: these are all lru_cache'd so it should be fine to call these for every service
+                system_paasta_config = load_system_paasta_config()
+                kube_client = KubeClient()
+                spark_kube_client = KubeClient(
+                    config_file=system_paasta_config.get_spark_kubeconfig()
+                )
+                # PaaSTA will not necessarily have created the SAs we want to use
+                # ...so let's go ahead and create them!
+                ensure_service_accounts(new_config, kube_client, spark_kube_client)
+
                 if client.update_namespace(service, new_config):
                     updated.append(service)
                     log.debug(f"Updated {service}")

{paasta-tools-1.4.6 → paasta-tools-1.4.7}/paasta_tools/spark_tools.py

@@ -6,6 +6,7 @@ from typing import Any
 from typing import cast
 from typing import Dict
 from typing import List
+from typing import Literal
 from typing import Mapping
 from typing import Set
 
@@ -23,6 +24,7 @@ SPARK_DRIVER_POOL = "stable"
 SPARK_JOB_USER = "TRON"
 SPARK_PROMETHEUS_SHARD = "ml-compute"
 SPARK_DNS_POD_TEMPLATE = "/nail/srv/configs/spark_dns_pod_template.yaml"
+MEM_MULTIPLIER = {"k": 1024, "m": 1024**2, "g": 1024**3, "t": 1024**4}
 
 log = logging.getLogger(__name__)
 
@@ -247,3 +249,22 @@ def get_spark_driver_monitoring_labels(
         "spark.yelp.com/driver_ui_port": ui_port_str,
     }
     return labels
+
+
+def get_spark_memory_in_unit(mem: str, unit: Literal["k", "m", "g", "t"]) -> float:
+    """
+    Converts Spark memory to the desired unit.
+    mem is the same format as JVM memory strings: just number or number followed by 'k', 'm', 'g' or 't'.
+    unit can be 'k', 'm', 'g' or 't'.
+    Returns memory as a float converted to the desired unit.
+    """
+    try:
+        memory_bytes = float(mem)
+    except ValueError:
+        try:
+            memory_bytes = float(mem[:-1]) * MEM_MULTIPLIER[mem[-1]]
+        except (ValueError, IndexError):
+            print(f"Unable to parse memory value {mem}. Defaulting to 2 GB.")
+            memory_bytes = 2147483648  # default to 2 GB
+    memory_unit = memory_bytes / MEM_MULTIPLIER[unit]
+    return memory_unit

{paasta-tools-1.4.6 → paasta-tools-1.4.7}/paasta_tools/tron_tools.py

@@ -62,7 +62,7 @@ from paasta_tools import spark_tools
 
 from paasta_tools.kubernetes_tools import (
     allowlist_denylist_to_requirements,
-    create_or_find_service_account_name,
+    get_service_account_name,
     limit_size_with_hash,
     raw_selectors_to_requirements,
     to_node_label,
@@ -280,9 +280,40 @@ class TronActionConfig(InstanceConfig):
             soa_dir=soa_dir,
         )
         self.job, self.action = decompose_instance(instance)
+
         # Indicate whether this config object is created for validation
         self.for_validation = for_validation
 
+        self.action_spark_config = None
+        if self.get_executor() == "spark":
+            # build the complete Spark configuration
+            # TODO: add conditional check for Spark specific commands spark-submit, pyspark etc ?
+            self.action_spark_config = self.build_spark_config()
+
+    def get_cpus(self) -> float:
+        # set Spark driver pod CPU if it is specified by Spark arguments
+        if (
+            self.action_spark_config
+            and "spark.driver.cores" in self.action_spark_config
+        ):
+            return float(self.action_spark_config["spark.driver.cores"])
+        # we fall back to this default if there's no spark.driver.cores config
+        return super().get_cpus()
+
+    def get_mem(self) -> float:
+        # set Spark driver pod memory if it is specified by Spark arguments
+        if (
+            self.action_spark_config
+            and "spark.driver.memory" in self.action_spark_config
+        ):
+            return int(
+                spark_tools.get_spark_memory_in_unit(
+                    self.action_spark_config["spark.driver.memory"], "m"
+                )
+            )
+        # we fall back to this default if there's no spark.driver.memory config
+        return super().get_mem()
+
     def build_spark_config(self) -> Dict[str, str]:
         system_paasta_config = load_system_paasta_config()
         resolved_cluster = system_paasta_config.get_eks_cluster_aliases().get(
@@ -354,15 +385,11 @@ class TronActionConfig(InstanceConfig):
             "spark.kubernetes.executor.label.yelp.com/owner", self.get_team()
         )
 
-        # We need to make sure the Service Account used by the executors has been created.
         # We are using the Service Account created using the provided or default IAM role.
         spark_conf[
             "spark.kubernetes.authenticate.executor.serviceAccountName"
-        ] = create_or_find_service_account_name(
+        ] = get_service_account_name(
             iam_role=self.get_spark_executor_iam_role(),
-            namespace=spark_tools.SPARK_EXECUTOR_NAMESPACE,
-            kubeconfig_file=system_paasta_config.get_spark_kubeconfig(),
-            dry_run=self.for_validation,
         )
 
         return spark_conf
@@ -440,7 +467,6 @@ class TronActionConfig(InstanceConfig):
         system_paasta_config: Optional["SystemPaastaConfig"] = None,
     ) -> Dict[str, str]:
         env = super().get_env(system_paasta_config=system_paasta_config)
-
         if self.get_executor() == "spark":
             # Required by some sdks like boto3 client. Throws NoRegionError otherwise.
             # AWS_REGION takes precedence if set.
@@ -605,6 +631,20 @@ class TronActionConfig(InstanceConfig):
             error_msgs.append(
                 f"{self.get_job_name()}.{self.get_action_name()} must have a deploy_group set"
             )
+        # We are not allowing users to specify `cpus` and `mem` configuration if the action is a Spark job
+        # with driver running on k8s (executor: spark), because we derive these values from `spark.driver.cores`
+        # and `spark.driver.memory` in order to avoid confusion.
+        if self.get_executor() == "spark":
+            if "cpus" in self.config_dict:
+                error_msgs.append(
+                    f"{self.get_job_name()}.{self.get_action_name()} is a Spark job. `cpus` config is not allowed. "
+                    f"Please specify the driver cores using `spark.driver.cores`."
+                )
+            if "mem" in self.config_dict:
+                error_msgs.append(
+                    f"{self.get_job_name()}.{self.get_action_name()} is a Spark job. `mem` config is not allowed. "
+                    f"Please specify the driver memory using `spark.driver.memory`."
+                )
         return error_msgs
 
     def get_pool(self) -> str:
@@ -952,20 +992,14 @@ def format_tron_action_dict(action_config: TronActionConfig):
 
         result["labels"]["yelp.com/owner"] = "compute_infra_platform_experience"
 
-        # create_or_find_service_account_name requires k8s credentials, and we don't
-        # have those available for CI to use (nor do we check these for normal PaaSTA
-        # services, so we're not doing anything "new" by skipping this)
         if (
             action_config.get_iam_role_provider() == "aws"
             and action_config.get_iam_role()
-            and not action_config.for_validation
         ):
             # this service account will be used for normal Tron batches as well as for Spark drivers
-            result["service_account_name"] = create_or_find_service_account_name(
+            result["service_account_name"] = get_service_account_name(
                 iam_role=action_config.get_iam_role(),
-                namespace=EXECUTOR_TYPE_TO_NAMESPACE[executor],
                 k8s_role=None,
-                dry_run=action_config.for_validation,
             )
 
         # service account token volumes for service authentication
@@ -975,21 +1009,26 @@ def format_tron_action_dict(action_config: TronActionConfig):
         if executor == "spark":
             is_mrjob = action_config.config_dict.get("mrjob", False)
             system_paasta_config = load_system_paasta_config()
-            # inject spark configs to the original spark-submit command
-            spark_config = action_config.build_spark_config()
+            # inject additional Spark configs in case of Spark commands
             result["command"] = spark_tools.build_spark_command(
                 result["command"],
-                spark_config,
+                action_config.action_spark_config,
                 is_mrjob,
                 action_config.config_dict.get(
                     "max_runtime", spark_tools.DEFAULT_SPARK_RUNTIME_TIMEOUT
                 ),
             )
+            # point to the KUBECONFIG needed by Spark driver
             result["env"]["KUBECONFIG"] = system_paasta_config.get_spark_kubeconfig()
+
             # spark, unlike normal batches, needs to expose several ports for things like the spark
             # ui and for executor->driver communication
             result["ports"] = list(
-                set(spark_tools.get_spark_ports_from_config(spark_config))
+                set(
+                    spark_tools.get_spark_ports_from_config(
+                        action_config.action_spark_config
+                    )
+                )
             )
             # mount KUBECONFIG file for Spark drivers to communicate with EKS cluster
             extra_volumes.append(
@@ -1003,10 +1042,12 @@ def format_tron_action_dict(action_config: TronActionConfig):
             )
             # Add pod annotations and labels for Spark monitoring metrics
             monitoring_annotations = (
-                spark_tools.get_spark_driver_monitoring_annotations(spark_config)
+                spark_tools.get_spark_driver_monitoring_annotations(
+                    action_config.action_spark_config
+                )
             )
             monitoring_labels = spark_tools.get_spark_driver_monitoring_labels(
-                spark_config
+                action_config.action_spark_config
             )
             result["annotations"].update(monitoring_annotations)
             result["labels"].update(monitoring_labels)

{paasta-tools-1.4.6 → paasta-tools-1.4.7/paasta_tools.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 1.2
 Name: paasta-tools
-Version: 1.4.6
+Version: 1.4.7
 Summary: Tools for Yelps SOA infrastructure
 Home-page: UNKNOWN
 Author: Compute Infrastructure @ Yelp

{paasta-tools-1.4.6 → paasta-tools-1.4.7}/paasta_tools.egg-info/requires.txt

@@ -26,6 +26,7 @@ kubernetes<22.0.0,>=18.20.0
 ldap3
 manhole
 mypy-extensions>=0.3.0
+nats-py
 nulltype
 objgraph
 ply

{paasta-tools-1.4.6 → paasta-tools-1.4.7}/requirements-minimal.txt

@@ -31,6 +31,7 @@ kubernetes >= 18.20.0, < 22.0.0
 ldap3
 manhole
 mypy-extensions >= 0.3.0
+nats-py
 nulltype
 objgraph
 ply