Merge pull request #13 from malaval/plugin-v2

Correct bug when node ID > 9

Author: amrragab8080

.DS_Store

@@ -134,7 +134,7 @@ This JSON file specifies the groups of nodes and associated partitions that Slur
 * `Partitions`: List of partitions
    * `PartitionName`: Name of the partition. Must match the pattern `^[a-zA-Z0-9_]+$`.
    * `NodeGroups`: List of node groups for this partition. A node group is a set of nodes that share the same specifications.
-      * `NodeGroupName`: Name of the node group. Must match the pattern `^[a-zA-Z0-9_]+$`.
+      * `NodeGroupName`: Name of the node group. Must match the pattern `^[a-zA-Z0-9_]+[a-zA-Z_]$`.
       * `MaxNodes`: Maximum number of nodes that Slurm can launch for this node group. For each node group, `generate_conf.py` will issue a line with `NodeName=[partition_name]-[nodegroup_name][0-(max_nodes-1)]`
       * `Region`: Name of the AWS region where to launch EC2 instances for this node group. Example: `us-east-1`.
       * [OPTIONAL] `ProfileName`: Name of the AWS CLI profile to use to authenticate AWS requests. If you don't specify a profile name, it uses the default profile name of EC2 metadata credentials.

README.md

@@ -82,14 +82,14 @@ def validate_partitions(data):
 
     for i_partition, partition in enumerate(data['Partitions']):
         assert 'PartitionName' in partition, 'Missing "PartitionName" in root["Partitions"][%s]' %i_partition
-        assert re.match('^[a-zA-Z0-9_]+$', partition['PartitionName']), 'root["Partitions"][%s]["PartitionName"] does not match ^[a-zA-Z0-9-]+$' %i_partition
+        assert re.match('^[a-zA-Z0-9_]+$', partition['PartitionName']), 'root["Partitions"][%s]["PartitionName"] does not match ^[a-zA-Z0-9_]+$' %i_partition
 
         assert 'NodeGroups' in partition, 'Missing "NodeGroups" in root["Partitions"][%s]' %i_partition
         assert isinstance(partition['NodeGroups'], list), 'root["Partitions"][%s]["NodeGroups"] is not an array' %i_partition
 
         for i_nodegroup, nodegroup in enumerate(partition['NodeGroups']):
             assert 'NodeGroupName' in nodegroup, 'Missing "NodeGroupName" in root["Partitions"][%s]["NodeGroups"][%s]' %(i_partition, i_nodegroup)
-            assert re.match('^[a-zA-Z0-9_]+$', nodegroup['NodeGroupName']), 'root["Partitions"][%s]["NodeGroups"][%s]["NodeGroupName"] does not match ^[a-zA-Z0-9-]+$' %(i_partition, i_nodegroup)
+            assert re.match('^[a-zA-Z0-9_]+[a-zA-Z_]$', nodegroup['NodeGroupName']), 'root["Partitions"][%s]["NodeGroups"][%s]["NodeGroupName"] does not match ^[a-zA-Z0-9_]+[a-zA-Z_]$' %(i_partition, i_nodegroup)
 
             assert 'MaxNodes' in nodegroup, 'Missing "MaxNodes" in root["Partitions"][%s]["NodeGroups"][%s]' %(i_partition, i_nodegroup)
             assert isinstance(nodegroup['MaxNodes'], int), 'root["Partitions"][%s]["NodeGroups"][%s]["MaxNodes"] is not a number' %(i_partition, i_nodegroup)
@@ -238,7 +238,7 @@ def parse_node_names(node_names):
     for node_name in node_names:
 
         # For each node: extract partition name, node group name and node id
-        pattern = '^([a-zA-Z0-9_]+)-([a-zA-Z0-9_]+)([0-9]+)$'
+        pattern = '^([a-zA-Z0-9_]+)-([a-zA-Z0-9_]+[a-zA-Z_])([0-9]+)$'
         match = re.match(pattern, node_name)
         if match:
             partition_name, nodegroup_name, node_id = match.groups()

common.py

@@ -30,13 +30,14 @@
 
         nb_nodes_to_resume = len(node_ids)
         nodegroup = common.get_partition_nodegroup(partition_name, nodegroup_name)
-        client = common.get_ec2_client(nodegroup)
 
         # Ignore if the partition and the node group are not in partitions.json
         if nodegroup is None:
-            logger.debug('Skipping partition=%s nodegroup=%s: not in partition.json' %(partition_name, nodegroup_name))
+            logger.warning('Skipping partition=%s nodegroup=%s: not in partition.json' %(partition_name, nodegroup_name))
             continue
 
+        client = common.get_ec2_client(nodegroup)
+
         # Create a dict for the EC2 CreateFleet request
         request_fleet = {
             'LaunchTemplateConfigs': [

resume.py

@@ -27,6 +27,12 @@
     for nodegroup_name, node_ids in nodegroups.items():
 
         nodegroup = common.get_partition_nodegroup(partition_name, nodegroup_name)
+
+        # Ignore if the partition and the node group are not in partitions.json
+        if nodegroup is None:
+            logger.warning('Skipping partition=%s nodegroup=%s: not in partition.json' %(partition_name, nodegroup_name))
+            continue
+
         client = common.get_ec2_client(nodegroup)
 
         # Retrieve the list of instances to terminate based on the tag Name