Add a 'restart --replica' option

paasta_tools/api/api.py

@@ -152,6 +152,10 @@ def make_app(global_config=None):
         "service.instance.set_state",
         "/v1/services/{service}/{instance}/state/{desired_state}",
     )
+    config.add_route(
+        "service.instance.replica.restart",
+        "/v1/services/{service}/{instance}/replicas/{replica_name}/restart",
+    )
     config.add_route(
         "service.instance.delay", "/v1/services/{service}/{instance}/delay"
     )

paasta_tools/api/api_docs/oapi.yaml

@@ -358,6 +358,17 @@ components:
     InstanceSpecCassandraCluster:
       description: Cassandra instance spec
       type: object
+    InstanceReplicaRestartOutcome:
+      properties:
+        message:
+          type: string
+        service:
+          type: string
+        instance:
+          type: string
+        replica_name:
+          type: string
+      type: object
     InstanceStatus:
       properties:
         adhoc:
@@ -1923,6 +1934,49 @@ paths:
       summary: Get mesos task of service_name.instance_name by task_id
       tags:
       - service
+  /services/{service}/{instance}/replicas/{replica_name}/restart:
+    post:
+      operationId: instance_replica_restart
+      parameters:
+      - description: Service name
+        in: path
+        name: service
+        required: true
+        schema:
+          type: string
+      - description: Instance name
+        in: path
+        name: instance
+        required: true
+        schema:
+          type: string
+      - description: Replica name to restart
+        in: path
+        name: replica_name
+        required: true
+        schema:
+          type: string
+      - description: Force immediate deletion (grace_period_seconds=0) instead of graceful termination
+        in: query
+        name: force
+        required: false
+        schema:
+          type: boolean
+          default: false
+      responses:
+        "200":
+          content:
+            application/json:
+              schema:
+                $ref: '#/components/schemas/InstanceReplicaRestartOutcome'
+          description: Replica restart initiated successfully
+        "404":
+          description: Replica not found or service/instance not found
+        "500":
+          description: Replica restart failed or instance type not supported
+      summary: Restart a specific replica for service_name.instance_name
+      tags:
+      - service
   /version:
     get:
       operationId: showVersion

paasta_tools/api/api_docs/swagger.json

@@ -1157,6 +1157,60 @@
                     }
                 ]
             }
+        },
+        "/services/{service}/{instance}/replicas/{replica_name}/restart": {
+            "post": {
+                "responses": {
+                    "200": {
+                        "description": "Replica restart initiated successfully",
+                        "schema": {
+                            "$ref": "#/definitions/InstanceReplicaRestartOutcome"
+                        }
+                    },
+                    "404": {
+                        "description": "Replica not found or service/instance not found"
+                    },
+                    "500": {
+                        "description": "Replica restart failed or instance type not supported"
+                    }
+                },
+                "summary": "Restart a specific replica for service_name.instance_name",
+                "operationId": "instance_replica_restart",
+                "tags": [
+                    "service"
+                ],
+                "parameters": [
+                    {
+                        "in": "path",
+                        "description": "Service name",
+                        "name": "service",
+                        "required": true,
+                        "type": "string"
+                    },
+                    {
+                        "in": "path",
+                        "description": "Instance name",
+                        "name": "instance",
+                        "required": true,
+                        "type": "string"
+                    },
+                    {
+                        "in": "path",
+                        "description": "Replica name to restart",
+                        "name": "replica_name",
+                        "required": true,
+                        "type": "string"
+                    },
+                    {
+                        "in": "query",
+                        "description": "Force immediate deletion (grace_period_seconds=0) instead of graceful termination",
+                        "name": "force",
+                        "required": false,
+                        "type": "boolean",
+                        "default": false
+                    }
+                ]
+            }
         }
     },
     "definitions": {
@@ -1196,6 +1250,23 @@
                 }
             }
         },
+        "InstanceReplicaRestartOutcome": {
+            "type": "object",
+            "properties": {
+                "message": {
+                    "type": "string"
+                },
+                "service": {
+                    "type": "string"
+                },
+                "instance": {
+                    "type": "string"
+                },
+                "replica_name": {
+                    "type": "string"
+                }
+            }
+        },
         "InstanceStatus": {
             "type": "object",
             "properties": {

paasta_tools/api/views/instance.py

@@ -400,3 +400,63 @@ def instance_mesh_status(
         raise ApiFailure(error_message, 500)
 
     return instance_mesh
+
+
+@view_config(
+    route_name="service.instance.replica.restart",
+    request_method="POST",
+    renderer="json",
+)
+def instance_replica_restart(
+    request: Request,
+) -> dict[str, Any]:
+    service = request.swagger_data.get("service")
+    instance = request.swagger_data.get("instance")
+    replica_name = request.swagger_data.get("replica_name")
+    force = request.swagger_data.get("force", False)
+
+    try:
+        instance_type = validate_service_instance(
+            service, instance, settings.cluster, settings.soa_dir
+        )
+    except NoConfigurationForServiceError:
+        error_message = no_configuration_for_service_message(
+            settings.cluster,
+            service,
+            instance,
+        )
+        raise ApiFailure(error_message, 404)
+    except Exception:
+        error_message = traceback.format_exc()
+        raise ApiFailure(error_message, 500)
+
+    if pik.can_restart_replica(instance_type):
+        try:
+            replica_restarted = pik.restart_replica_by_name(
+                service=service,
+                instance=instance,
+                instance_type=instance_type,
+                replica_name=replica_name,
+                settings=settings,
+                force=force,
+            )
+            if replica_restarted:
+                return {
+                    "message": f"Initiated replica restart of {replica_name} successfully",
+                    "service": service,
+                    "instance": instance,
+                    "replica_name": replica_name,
+                }
+            else:
+                raise ApiFailure(
+                    f"Replica {replica_name} not found in service {service}.{instance}",
+                    404,
+                )
+        except RuntimeError as e:
+            raise ApiFailure(str(e), 500)
+    else:
+        error_message = (
+            f"instance_type {instance_type} of {service}.{instance} doesn't "
+            "support replica restart"
+        )
+        raise ApiFailure(error_message, 500)

paasta_tools/cli/cmds/remote_run.py

@@ -13,7 +13,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 import argparse
-import json
 import os
 import shutil
 import subprocess
@@ -24,6 +23,7 @@
 from paasta_tools.cli.utils import get_paasta_oapi_api_clustername
 from paasta_tools.cli.utils import get_paasta_oapi_client_with_auth
 from paasta_tools.cli.utils import lazy_choices_completer
+from paasta_tools.cli.utils import parse_error
 from paasta_tools.cli.utils import run_interactive_cli
 from paasta_tools.kubernetes.remote_run import format_remote_run_job_name
 from paasta_tools.kubernetes.remote_run import load_eks_or_adhoc_deployment_config
@@ -75,18 +75,6 @@ def _get_kubectl_wrapper(cluster: str) -> str:
     return kubectl_wrapper
 
 
-def parse_error(body: str) -> str:
-    try:
-        body_object = json.loads(body)
-    except json.decoder.JSONDecodeError:
-        return body
-    return (
-        body_object.get("reason")
-        or body_object.get("message")
-        or json.dumps(body_object, indent=4)
-    )
-
-
 def paasta_remote_run_copy(
     args: argparse.Namespace,
     system_paasta_config: SystemPaastaConfig,