Merge pull request #86 from sean-smith/merge-master

Release 2.1.0

CHANGELOG.md

@@ -3,6 +3,18 @@ aws-parallelcluster-node CHANGELOG
 
 This file is used to list changes made in each version of the aws-parallelcluster-node package.
 
+2.1.0
+-----
+
+Bug Fixes:
+
+    - Don't schedule jobs on compute nodes that are terminating
+
+2.0.2
+-----
+
+- Align version to main ParallelCluster package
+
 2.0.0
 -----
 

README.rst

@@ -1,9 +1,14 @@
-=========================
-aws-parallelcluster-node
-=========================
+========================
+AWS ParallelCluster Node
+========================
 
-.. image:: https://travis-ci.org/aws/aws-parallelcluster-node.png?branch=develop
-   :target: https://travis-ci.org/aws/aws-parallelcluster-node
-   :alt: Build Status
+|Build Status| |Version|
 
-   aws-parallelcluster-node is the python package installed on the Amazon EC2 instances launched as part of AWS ParallelCluster.
+.. |Build Status| image:: https://travis-ci.org/aws/aws-parallelcluster-node.png?branch=develop
+    :target: https://travis-ci.org/aws/aws-parallelcluster-node/
+    :alt: Build Status
+.. |Version| image:: https://badge.fury.io/py/aws-parallelcluster-node.png
+    :target: https://badge.fury.io/py/aws-parallelcluster-node
+
+This repo contains the aws-parallelcluster-node package installed on the Amazon EC2 instances launched
+as part of AWS ParallelCluster.

nodewatcher/nodewatcher.py

@@ -152,7 +152,8 @@ def stackCreationComplete(stack_name, region, proxy_config):
     log.info('Checking for status of the stack %s' % stack_name)
     cfn_client = boto3.client('cloudformation', region_name=region, config=proxy_config)
     stacks = cfn_client.describe_stacks(StackName=stack_name)
-    return stacks['Stacks'][0]['StackStatus'] == 'CREATE_COMPLETE'
+    return stacks['Stacks'][0]['StackStatus'] == 'CREATE_COMPLETE' or \
+           stacks['Stacks'][0]['StackStatus'] == 'UPDATE_COMPLETE'
 
 
 def main():
@@ -183,7 +184,13 @@ def main():
         data = {_CURRENT_IDLETIME: 0}
 
     stack_creation_complete = False
+    termination_in_progress = False
     while True:
+        # if this node is terminating sleep for a long time and wait for termination
+        if termination_in_progress:
+            time.sleep(300)
+            log.info('%s is still terminating' % hostname)
+            continue
         time.sleep(60)
         if not stack_creation_complete:
             stack_creation_complete = stackCreationComplete(stack_name, region, proxy_config)
@@ -217,6 +224,7 @@ def main():
                             os.remove(_IDLETIME_FILE)
                             try:
                                 selfTerminate(asg_name, asg_conn, instance_id)
+                                termination_in_progress = True
                             except ClientError as ex:
                                 log.error('Failed to terminate instance: %s with exception %s' % (instance_id, ex))
                                 lockHost(s, hostname, unlock=True)

requirements26.txt

@@ -1,2 +1,3 @@
 pycparser==2.18
-paramiko==2.3.3
+idna==2.6
+paramiko==2.3.3

setup.py

@@ -25,13 +25,14 @@ def read(fname):
 console_scripts = ['sqswatcher = sqswatcher.sqswatcher:main',
                    'nodewatcher = nodewatcher.nodewatcher:main',
                    'jobwatcher = jobwatcher.jobwatcher:main']
-version = "2.0.0"
+version = "2.1.0"
 requires = ['boto3>=1.7.55', 'python-dateutil>=2.6.1']
 
 if sys.version_info[:2] == (2, 6):
     # For python2.6 we have to require argparse since it
     # was not in stdlib until 2.7.
     requires.append('argparse>=1.4')
+    requires.append('idna==2.6')
     requires.append('paramiko==2.3.3')
     requires.append('pycparser==2.18')
 else: