paasta-tools 1.31.1__tar.gz → 1.31.3__tar.gz

{paasta_tools-1.31.1/paasta_tools.egg-info → paasta_tools-1.31.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: paasta-tools
-Version: 1.31.1
+Version: 1.31.3
 Summary: Tools for Yelps SOA infrastructure
 Author: Compute Infrastructure @ Yelp
 Author-email: compute-infra@yelp.com

{paasta_tools-1.31.1 → paasta_tools-1.31.3}/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__ = "1.31.1"
+__version__ = "1.31.3"

{paasta_tools-1.31.1 → paasta_tools-1.31.3}/paasta_tools/api/api_docs/swagger.json

@@ -1788,6 +1788,11 @@
         "KubernetesVersion": {
             "type": "object",
             "properties": {
+                "container_port": {
+                    "description": "Port the container is expecting to receive traffic on",
+                    "type": "integer",
+                    "format": "int32"
+                },
                 "type": {
                     "description": "Type of version (ReplicaSet or ControllerRevision)",
                     "type": "string"

{paasta_tools-1.31.1 → paasta_tools-1.31.3}/paasta_tools/cli/cmds/status.py

@@ -72,7 +72,6 @@ from paasta_tools.kubernetes_tools import format_tail_lines_for_kubernetes_pod
 from paasta_tools.kubernetes_tools import KubernetesDeploymentConfig
 from paasta_tools.kubernetes_tools import KubernetesDeployStatus
 from paasta_tools.kubernetes_tools import paasta_prefixed
-from paasta_tools.long_running_service_tools import DEFAULT_CONTAINER_PORT
 from paasta_tools.monitoring_tools import get_runbook
 from paasta_tools.monitoring_tools import get_team
 from paasta_tools.monitoring_tools import list_teams
@@ -918,11 +917,13 @@ def _print_flink_status_from_job_manager(
             else f"{pod_evicted_count}"
         )
 
+    pods_total_count = pod_running_count + pod_evicted_count + pod_other_count
     output.append(
         "    Pods:"
         f" {pod_running_count} running,"
         f" {evicted} evicted,"
-        f" {pod_other_count} other"
+        f" {pod_other_count} other,"
+        f" {pods_total_count} total"
     )
 
     if not should_job_info_be_shown(status["state"]):
@@ -945,12 +946,19 @@ def _print_flink_status_from_job_manager(
             output.append(str(e))
             return 1
 
+        jobs_total_count = (
+            overview.jobs_running
+            + overview.jobs_finished
+            + overview.jobs_failed
+            + overview.jobs_cancelled
+        )
         output.append(
             "    Jobs:"
             f" {overview.jobs_running} running,"
             f" {overview.jobs_finished} finished,"
             f" {overview.jobs_failed} failed,"
-            f" {overview.jobs_cancelled} cancelled"
+            f" {overview.jobs_cancelled} cancelled,"
+            f" {jobs_total_count} total"
         )
         output.append(
             "   "
@@ -1314,7 +1322,9 @@ def get_version_table_entry(
             for state in ReplicaState
             if state in replica_state_counts
         ]
-        entry.append(f"  Replica States: {' / '.join(replica_state_display)}")
+        entry.append(
+            f"  Replica States: {' / '.join(replica_state_display)} / {replica_state_counts.total()} total"
+        )
         if not verbose:
             unhealthy_replicas = [
                 (state, pod) for state, pod in replica_states if state.is_unhealthy()
@@ -1322,13 +1332,23 @@ def get_version_table_entry(
             if unhealthy_replicas:
                 entry.append("    Unhealthy Replicas:")
                 replica_table = create_replica_table(
-                    unhealthy_replicas, service, instance, cluster, verbose
+                    unhealthy_replicas,
+                    service,
+                    instance,
+                    cluster,
+                    version.container_port,
+                    verbose,
                 )
                 for line in replica_table:
                     entry.append(f"      {line}")
         else:
             replica_table = create_replica_table(
-                replica_states, service, instance, cluster, verbose
+                replica_states,
+                service,
+                instance,
+                cluster,
+                version.container_port,
+                verbose,
             )
             for line in replica_table:
                 entry.append(f"    {line}")
@@ -1403,12 +1423,6 @@ def get_main_container(pod: KubernetesPodV2) -> Optional[KubernetesContainerV2]:
     )
 
 
-def get_container_env(container: Optional[KubernetesContainerV2]) -> Dict[str, str]:
-    if not container or not container.env:
-        return {}
-    return {env_var.name: env_var.value for env_var in container.env}
-
-
 def get_replica_state(pod: KubernetesPodV2) -> ReplicaState:
     phase = pod.phase
     state = ReplicaState.UNKNOWN
@@ -1474,6 +1488,7 @@ def create_replica_table(
     service: str,
     instance: str,
     cluster: str,
+    container_port: int,
     verbose: int = 0,
 ) -> List[str]:
     header = ["ID", "IP/Port", "Host deployed to", "Started at what localtime", "State"]
@@ -1481,14 +1496,10 @@ def create_replica_table(
     for state, pod in pods:
         start_datetime = datetime.fromtimestamp(pod.create_timestamp)
         humanized_start_time = humanize.naturaltime(start_datetime)
-        # Get port from the main container's PAASTA_PORT env var, fallback to DEFAULT_CONTAINER_PORT
-        main_container = get_main_container(pod)
-        container_env = get_container_env(main_container)
-        paasta_port = container_env.get("PAASTA_PORT", str(DEFAULT_CONTAINER_PORT))
 
         row = [
             pod.name,
-            f"{pod.ip}:{paasta_port}" if pod.ip else "None",
+            f"{pod.ip}:{container_port}" if pod.ip else "None",
             pod.host or "None",
             humanized_start_time,
             state.formatted_message,
@@ -1503,6 +1514,7 @@ def create_replica_table(
                 )
             )
 
+        main_container = get_main_container(pod)
         if main_container:
             if main_container.timestamp:
                 timestamp = datetime.fromtimestamp(main_container.timestamp)

{paasta_tools-1.31.1 → paasta_tools-1.31.3}/paasta_tools/instance/kubernetes.py

@@ -1,4 +1,5 @@
 import asyncio
+import logging
 from asyncio.tasks import Task
 from collections import defaultdict
 from enum import Enum
@@ -17,6 +18,7 @@ from typing import Union
 
 import a_sync
 import pytz
+import requests.exceptions
 from kubernetes.client import V1Container
 from kubernetes.client import V1ControllerRevision
 from kubernetes.client import V1Pod
@@ -75,6 +77,8 @@ INSTANCE_TYPE_CR_ID = dict(
     monkrelaycluster=monkrelaycluster_tools.cr_id,
 )
 
+logger = logging.getLogger(__name__)
+
 
 class ServiceMesh(Enum):
     SMARTSTACK = "smartstack"
@@ -100,6 +104,7 @@ class KubernetesVersionDict(TypedDict, total=False):
     config_sha: str
     pods: Sequence[Mapping[str, Any]]
     namespace: str
+    container_port: Optional[int]
 
 
 def cr_id(service: str, instance: str, instance_type: str) -> Mapping[str, str]:
@@ -347,31 +352,49 @@ async def mesh_status(
 
     pods = await pods_task
     for location, hosts in node_hostname_by_location.items():
-        host = replication_checker.get_hostname_in_pool(hosts, instance_pool)
-        if service_mesh == ServiceMesh.SMARTSTACK:
-            mesh_status["locations"].append(
-                _build_smartstack_location_dict(
-                    synapse_host=host,
-                    synapse_port=settings.system_paasta_config.get_synapse_port(),
-                    synapse_haproxy_url_format=settings.system_paasta_config.get_synapse_haproxy_url_format(),
-                    registration=registration,
-                    pods=pods,
-                    location=location,
-                    should_return_individual_backends=should_return_individual_backends,
-                )
-            )
-        elif service_mesh == ServiceMesh.ENVOY:
-            mesh_status["locations"].append(
-                _build_envoy_location_dict(
-                    envoy_host=host,
-                    envoy_admin_port=settings.system_paasta_config.get_envoy_admin_port(),
-                    envoy_admin_endpoint_format=settings.system_paasta_config.get_envoy_admin_endpoint_format(),
-                    registration=registration,
-                    pods=pods,
-                    location=location,
-                    should_return_individual_backends=should_return_individual_backends,
-                )
-            )
+        max_retries = 3
+
+        for attempt in range(max_retries):
+            host = replication_checker.get_hostname_in_pool(hosts, instance_pool)
+            try:
+                if service_mesh == ServiceMesh.SMARTSTACK:
+                    location_dict = _build_smartstack_location_dict(
+                        synapse_host=host,
+                        synapse_port=settings.system_paasta_config.get_synapse_port(),
+                        synapse_haproxy_url_format=settings.system_paasta_config.get_synapse_haproxy_url_format(),
+                        registration=registration,
+                        pods=pods,
+                        location=location,
+                        should_return_individual_backends=should_return_individual_backends,
+                    )
+                elif service_mesh == ServiceMesh.ENVOY:
+                    location_dict = _build_envoy_location_dict(
+                        envoy_host=host,
+                        envoy_admin_port=settings.system_paasta_config.get_envoy_admin_port(),
+                        envoy_admin_endpoint_format=settings.system_paasta_config.get_envoy_admin_endpoint_format(),
+                        registration=registration,
+                        pods=pods,
+                        location=location,
+                        should_return_individual_backends=should_return_individual_backends,
+                    )
+
+                mesh_status["locations"].append(location_dict)
+                return mesh_status
+
+            except requests.exceptions.ConnectTimeout:
+                if attempt < max_retries - 1:
+                    logger.warning(
+                        "attempt %s/%s: Unable to connect to %s, retrying (on another host, hopefully)...",
+                        attempt,
+                        max_retries,
+                        host,
+                    )
+                    continue
+                else:
+                    logger.critical(
+                        "Unable to connect to %s, not retrying again.", host
+                    )
+                    raise
     return mesh_status
 
 
@@ -698,6 +721,7 @@ async def kubernetes_status_v2(
                 instance=instance,
                 namespaces=relevant_namespaces,
                 pod_status_by_sha_and_readiness_task=pod_status_by_sha_and_readiness_task,  # type: ignore  # PAASTA-18698; ignoring due to unexpected type mismatch
+                container_port=job_config.get_container_port(),
             )
         )
         tasks.extend([pod_status_by_sha_and_readiness_task, versions_task])  # type: ignore  # PAASTA-18698; ignoring due to unexpected type mismatch
@@ -717,6 +741,7 @@ async def kubernetes_status_v2(
                 instance=instance,
                 namespaces=relevant_namespaces,
                 pod_status_by_replicaset_task=pod_status_by_replicaset_task,  # type: ignore  # PAASTA-18698; ignoring due to unexpected type mismatch
+                container_port=job_config.get_container_port(),
             )
         )
         tasks.extend([pod_status_by_replicaset_task, versions_task])  # type: ignore  # PAASTA-18698; ignoring due to unexpected type mismatch
@@ -788,6 +813,7 @@ async def get_versions_for_replicasets(
     instance: str,
     namespaces: Iterable[str],
     pod_status_by_replicaset_task: "asyncio.Future[Mapping[str, Sequence[asyncio.Future[Dict[str, Any]]]]]",
+    container_port: Optional[int],
 ) -> List[KubernetesVersionDict]:
 
     replicaset_list: List[V1ReplicaSet] = []
@@ -815,6 +841,7 @@ async def get_versions_for_replicasets(
                 replicaset,
                 kube_client,
                 pod_status_by_replicaset.get(replicaset.metadata.name),
+                container_port,
             )
             for replicaset in actually_running_replicasets
         ]
@@ -826,6 +853,7 @@ async def get_replicaset_status(
     replicaset: V1ReplicaSet,
     client: kubernetes_tools.KubeClient,
     pod_status_tasks: Sequence["asyncio.Future[Dict[str, Any]]"],
+    container_port: Optional[int],
 ) -> KubernetesVersionDict:
     return {
         "name": replicaset.metadata.name,
@@ -840,6 +868,7 @@ async def get_replicaset_status(
         "config_sha": replicaset.metadata.labels.get("paasta.yelp.com/config_sha"),
         "pods": await asyncio.gather(*pod_status_tasks) if pod_status_tasks else [],
         "namespace": replicaset.metadata.namespace,
+        "container_port": container_port,
     }
 
 
@@ -1063,6 +1092,7 @@ async def get_versions_for_controller_revisions(
     instance: str,
     namespaces: Iterable[str],
     pod_status_by_sha_and_readiness_task: "asyncio.Future[Mapping[Tuple[str, str], Mapping[bool, Sequence[asyncio.Future[Mapping[str, Any]]]]]]",
+    container_port: Optional[int] = None,
 ) -> List[KubernetesVersionDict]:
     controller_revision_list: List[V1ControllerRevision] = []
 
@@ -1092,6 +1122,7 @@ async def get_versions_for_controller_revisions(
                 cr,
                 kube_client,
                 pod_status_by_sha_and_readiness[(git_sha, config_sha)],
+                container_port=container_port,
             )
             for (git_sha, config_sha), cr in cr_by_shas.items()
         ]
@@ -1106,6 +1137,7 @@ async def get_version_for_controller_revision(
     pod_status_tasks_by_readiness: Mapping[
         bool, Sequence["asyncio.Future[Mapping[str, Any]]"]
     ],
+    container_port: Optional[int] = None,
 ) -> KubernetesVersionDict:
     all_pod_status_tasks = [
         task for tasks in pod_status_tasks_by_readiness.values() for task in tasks
@@ -1122,6 +1154,7 @@ async def get_version_for_controller_revision(
         "config_sha": cr.metadata.labels.get("paasta.yelp.com/config_sha"),
         "pods": [task.result() for task in all_pod_status_tasks],
         "namespace": cr.metadata.namespace,
+        "container_port": container_port,
     }
 
 

{paasta_tools-1.31.1 → paasta_tools-1.31.3}/paasta_tools/paastaapi/model/kubernetes_version.py

@@ -81,6 +81,7 @@ class KubernetesVersion(ModelNormal):
         """
         lazy_import()
         return {
+            'container_port': (int,),  # noqa: E501
             'type': (str,),  # noqa: E501
             'create_timestamp': (float,),  # noqa: E501
             'git_sha': (str,),  # noqa: E501
@@ -99,6 +100,7 @@ class KubernetesVersion(ModelNormal):
 
 
     attribute_map = {
+        'container_port': 'container_port',  # noqa: E501
         'type': 'type',  # noqa: E501
         'create_timestamp': 'create_timestamp',  # noqa: E501
         'git_sha': 'git_sha',  # noqa: E501
@@ -157,6 +159,7 @@ class KubernetesVersion(ModelNormal):
                                 Animal class but this time we won't travel
                                 through its discriminator because we passed in
                                 _visited_composed_classes = (Animal,)
+            container_port (int): Port the container is expecting to receive traffic on. [optional]  # noqa: E501
             type (str): Type of version (ReplicaSet or ControllerRevision). [optional]  # noqa: E501
             create_timestamp (float): Unix timestamp when version was created. [optional]  # noqa: E501
             git_sha (str): Git SHA of service code for this version of the instance. [optional]  # noqa: E501

{paasta_tools-1.31.1 → paasta_tools-1.31.3/paasta_tools.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: paasta-tools
-Version: 1.31.1
+Version: 1.31.3
 Summary: Tools for Yelps SOA infrastructure
 Author: Compute Infrastructure @ Yelp
 Author-email: compute-infra@yelp.com