Revert a fix in get_optimal_nodes to allow additional testing with torque

Note: sge and slurm are not affected by this
Signed-off-by: Francesco De Martino <[email protected]>

Author: demartinofra

jobwatcher/plugins/utils.py

@@ -21,12 +21,19 @@ def get_optimal_nodes(nodes_requested, slots_requested, instance_properties):
         slots_required_per_node = -(-slots // num_of_nodes)
 
         if slots_required_per_node > vcpus:
-            log.warning(
-                "Slots required per node (%d) is greater than vcpus available on single node (%d), skipping job...",
-                slots_required_per_node,
-                vcpus,
-            )
-            continue
+            # FIXME: Replace this code with the commented section below after testing with Torque
+            # If slots required per node is greater than vcpus, add additional nodes
+            # and recalculate slots_required_per_node
+            log.info("Slots required per node is greater than vcpus, recalculating")
+            num_of_nodes = -(-slots // vcpus)
+            slots_required_per_node = -(-slots // num_of_nodes)
+            log.info("Recalculated: %s nodes and %s slots required per node" % (num_of_nodes, slots_required_per_node))
+            # log.warning(
+            #     "Slots required per node (%d) is greater than vcpus available on single node (%d), skipping job...",
+            #     slots_required_per_node,
+            #     vcpus,
+            # )
+            # continue
 
         # Verify if there are enough available slots in the nodes allocated in the previous rounds
         for slot_idx, slots_available in enumerate(slots_remaining_per_node):

tests/jobwatcher/unittests.py

@@ -18,7 +18,7 @@ def test_each_node_at_capacity(self):
 
     def test_slots_requested_greater_than_available(self):
         nodes = utils.get_optimal_nodes([1], [9], instance_properties)
-        expected = 0
+        expected = 2
         self.assertEqual(nodes, expected)
 
     def test_each_node_half_capacity(self):