Add more optional parameters

fairing/builders/docker_builder.py

@@ -25,24 +25,29 @@
 class DockerBuilder(BuilderInterface):
     """A builder using the local Docker client"""
 
-    def __init__(self, 
-                repository, 
-                image_name=DEFAULT_IMAGE_NAME, 
-                image_tag=None, 
-                base_image=None, 
-                dockerfile_path=None):
+    def __init__(self,
+                 repository,
+                 image_name=DEFAULT_IMAGE_NAME,
+                 image_tag=None,
+                 base_image=None,
+                 dockerfile_path=None,
+                 image_pull_policy='Always',
+                 restart_policy='Never'):
 
         self.repository = repository
         self.image_name = image_name
         self.base_image = base_image
         self.dockerfile_path = dockerfile_path
+        self.image_pull_policy = image_pull_policy
+        self.restart_policy = restart_policy
 
         if image_tag is None:
             self.image_tag = utils.get_unique_tag()
         else: 
             self.image_tag = image_tag
+
         self.full_image_name = utils.get_image_full_name(
-            self.repository, 
+            self.repository,
             self.image_name,
             self.image_tag
         )
@@ -55,8 +60,9 @@ def generate_pod_spec(self):
             containers=[client.V1Container(
                 name='model',
                 image=self.full_image_name,
+                image_pull_policy=self.image_pull_policy,
             )],
-            restart_policy='Never'
+            restart_policy=self.restart_policy
         )
 
     def execute(self):

fairing/training/kubeflow/decorators.py

@@ -1,28 +1,30 @@
 
 from ..native import decorators
 from .deployment import KubeflowDeployment
-
+
+
 class Training(decorators.Training):
 
-    def __init__(self, namespace=None):
-        super(Training, self).__init__(namespace)
+    def __init__(self, namespace=None, job_name=None):
+        super(Training, self).__init__(namespace, job_name)
         self.distribution = {
             'Worker': 1
         }
 
     def _deploy(self, user_object):
         deployment = KubeflowDeployment(
             self.namespace,
+            self.job_name,
             self.runs,
             self.distribution)
         deployment.execute()
 
 
 class DistributedTraining(Training):
 
-    def __init__(self, worker_count=0, ps_count=0, namespace=None):
+    def __init__(self, worker_count=0, ps_count=0, namespace=None, job_name=None):
         # By default we set worker to 0 as we always add a Chief
-        super(DistributedTraining, self).__init__(namespace)
+        super(DistributedTraining, self).__init__(namespace, job_name)
         self.distribution = {
             'Worker': worker_count,
             'PS': ps_count,

fairing/training/kubeflow/deployment.py

@@ -2,10 +2,11 @@
 
 from ..native import deployment
 
+
 class KubeflowDeployment(deployment.NativeDeployment):
 
-    def __init__(self, namespace, runs, distribution):
-        super(KubeflowDeployment, self).__init__(namespace, runs)
+    def __init__(self, namespace, job_name, runs, distribution):
+        super(KubeflowDeployment, self).__init__(namespace, job_name, runs)
         self.distribution = distribution
 
     def deploy(self):
@@ -50,4 +51,4 @@ def set_container_name(self, pod_template_spec):
 
     def get_logs(self):
         selector='tf-replica-index=0,tf-replica-type=worker'
-        self.backend.log(self.name, self.namespace, selector)
+        self.backend.log(self.name, self.namespace, selector)

fairing/training/native/decorators.py

@@ -14,14 +14,16 @@
 
 logger = logging.getLogger(__name__)
 
+
 class Training(base.TrainingDecoratorInterface):
     """A simple Kubernetes training.
     
     Args:
         namespace {string} -- (optional) here the training should be deployed
     """
 
-    def __init__(self, namespace=None):
+    def __init__(self, namespace=None, job_name=None):
+        self.job_name = job_name
         self.namespace = namespace
         self.runs = 1
 
@@ -35,7 +37,7 @@ def _train(self, user_object):
         runtime.execute(user_object)
 
     def _deploy(self, user_object):
-        deployment = NativeDeployment(self.namespace, self.runs)
+        deployment = NativeDeployment(self.namespace, self.job_name, self.runs)
         deployment.execute()
 
 

fairing/training/native/deployment.py

@@ -25,14 +25,17 @@ class NativeDeployment(object):
             will generate multiple jobs.
     """
 
-    def __init__(self, namespace, runs):
+    def __init__(self, namespace, job_name, runs):
         if namespace is None:
             self.namespace = utils.get_default_target_namespace()
         else:
             self.namespace = namespace
 
         # Used as pod and job name
-        self.name = "{}-{}".format(DEFAULT_JOB_NAME, utils.get_unique_tag())
+        if job_name is None:
+            job_name = DEFAULT_JOB_NAME
+
+        self.name = "{}-{}".format(job_name, utils.get_unique_tag())
         self.job_spec = None
         self.runs = runs