Copying and serializing the image list for each node is very costly, do it only when needed

zoe_master/backends/interface.py

@@ -25,7 +25,7 @@ from zoe_lib.state import Execution, Service, SQLManager  # pylint: disable=unus
 from zoe_master.backends.base import BaseBackend
 from zoe_master.backends.service_instance import ServiceInstance
 from zoe_master.exceptions import ZoeStartExecutionFatalException, ZoeStartExecutionRetryException, ZoeException
-from zoe_master.stats import ClusterStats  # pylint: disable=unused-import
+from zoe_master.stats import ClusterStats, NodeStats  # pylint: disable=unused-import
 
 try:
     from zoe_master.backends.swarm.backend import SwarmBackend
@@ -167,12 +167,14 @@ def terminate_execution(execution: Execution) -> None:
     execution.set_terminated()
 
 
-def get_platform_state(state: SQLManager) -> ClusterStats:
+def get_platform_state(state: SQLManager, with_images=False) -> ClusterStats:
     """Retrieves the state of the platform by querying the container backend. Platform state includes information on free/reserved resources for each node. This information is used for advanced scheduling."""
     backend = _get_backend()
     platform_state = backend.platform_state()
-    for node in platform_state.nodes:
+    for node in platform_state.nodes:  # type: NodeStats
         node.services = state.service_list(backend_host=node.name, backend_status=Service.BACKEND_START_STATUS)
+        if not with_images:
+            node.image_list = []
     return platform_state
 
 

zoe_master/scheduler/elastic_scheduler.py

@@ -190,7 +190,7 @@ class ZoeElasticScheduler:
                     log.debug("-> {}".format(job))
 
                 try:
-                    platform_state = get_platform_state(self.state)
+                    platform_state = get_platform_state(self.state, with_images=True)
                 except ZoeException:
                     log.error('Cannot retrieve platform state, cannot schedule')
                     for job in jobs_to_attempt_scheduling: